AI Magazine - 2023 - Gudigantala - An AI Decision Making Framework For Business Value Maximization

Received: 23 June 2022 Revised: 20 November 2022 Accepted: 2 January 2023
DOI: 10.1002/aaai.12076
ARTICLE
An AI decision-making framework for business value

maximization
Naveen Gudigantala1 Sreedhar Madhavaram2 Pelin Bicen3
1 RobertB. Pamplin School of Business

Administration, University of Portland, Abstract
Portland, Oregon, USA This article addresses a key question of why businesses are failing to maximize
2 Jerry
S. Rawls College of Business business value from their artificial intelligence (AI) investments and proposes
Administration, Texas Tech University,
Lubbock, Texas, USA
a strategic decision-making framework for AI decision-making to address this
3 Department of Marketing, Sawyer problem. We suggest that a firm’s business strategy must drive AI-driven business
Business School, Suffolk University, outcomes and measurements, which in turn should drive the AI implementation
Boston, Massachusetts, USA
decisions. Very often, we find that businesses fail to successfully cast business
Correspondence problems into AI problems. To bridge this gap, we propose that firms use a perfor-
Naveen Gudigantala, Robert B. Pamplin mance management system such as objectives and key results (OKRs) to ensure
School of Business Administration,
that the business and AI goals & objectives are well defined, tightly aligned, and
University of Portland, Portland, OR
97203, USA. made transparent across the company, and the AI efforts are approached in an
Email: [email protected] integrated manner by the different parts of a firm. We use McDonald’s use of
AI initiatives as a business use case to demonstrate support for our AI decision-
making framework. We argue that using the business strategy as a primary driver
will enable firms to solve the right problems using AI, turning it to be a source
of technology innovation and competitive advantage.
INTRODUCTION their AI investments (Bean 2021). Indeed, most non-tech

Fortune 500 companies are stuck in pilot purgatory when
Artificial intelligence (AI) is beginning to transform many it comes to AI (Aaser, Woetzel, and Russell 2021). Studies
industries and is expected to contribute $15.7 trillion to the discussing the impact of AI on firm value have been rare
global economy by 2030 (Pumplun, Tauchert, and Heidt and have found mixed results. Lui, Lee, and Ngai (2022)
2019; PWC, 2019). This will facilitate the processing of large studied the impact of AI on firm value by studying the
scale and unstructured data in real-time, and by generating public listings of AI investments by 62 firms. They found
predictive insights that enhance business decision making that the stock prices of firms declined on average by 1.77%
(Albrecht, Rausch, and Derra 2021). While usage rates of on the day of the announcement. This finding suggests the
AI-related technologies are less than 5% among the 583,000 low confidence investors have in the ability of the firms
US businesses surveyed in 2018, the pace of AI-related to successfully implement AI efforts. Further, Mishra,
investments is picking up among the leading fortune-1000 Ewing, and Cooper (2022) confirmed with their findings
companies with 92% of the sample surveyed reporting that US-listed firms are in a state of impending transfor-
accelerated investments in data and AI (Bean 2021; Zolas mation with regard to AI. Despite growing enthusiasm
et al. 2021). However, more than 90% of the same sam- for AI’s potential, executives in a variety of industries
ple report that they are struggling to derive value from suggest that the technology is tricky to deploy and the
This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium,
provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.
© 2023 The Authors. AI Magazine published by Wiley Periodicals LLC on behalf of the Association for the Advancement of Artificial Intelligence
AI Magazine. 2023;44:67–84. wileyonlinelibrary.com/journal/aaai 67

23719621, 2023, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/aaai.12076 by Bbw Hochschule, Wiley Online Library on [12/04/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
68 AI MAGAZINE
initial payoffs are equivocal (Bergstein 2019; Haenlein and Organizations can use our framework to make deci-
Kaplan 2021; Lui, Lee, and Ngai 2022). Even within the sions on identifying and choosing relevant AI projects,
business functional areas where AI is expected to make implementing them, and continuously measuring the
a huge impact, a majority of the companies have not yet value they create to finetune these efforts.
moved beyond the AI evaluation stage, and fewer than
one in five have implemented AI in any particular core
process (IBM 2021). RATIONALE FOR THE AI
Prior research speculates many reasons why AI invest- DECISION-MAKING FRAMEWORK
ments are not yielding value due in part to a lack of
expertise, inadequate technological infrastructure, high There is a wide-spread recognition among businesses that
levels of perceived complexity, and dubious outcomes AI will be integrated into most of the enterprise applica-
(Bean 2021; Bergstein 2019; Meakin et al. 2021). In fact, tions in the next three years, but the competitive advantage
these factors may lead to satisficing—sub-optimal decision from AI may only depend on how creatively and respon-
making—that can impede AI-enabled strategic cognition sibly organizations use AI (Hupfer 2020; Lui, Lee, and
and capability learning in organizations (Simon 1957; Win- Ngai 2022). Given this context, a strategic decision-making
ter 2000). Despite the slowness of AI implementation suc- framework for effective and optimal AI is important for
cess, many organizations continue to see a great promise in many reasons. First, it is not the AI technologies them-
AI. The availability of big data, coupled with advancements selves, but how they are applied to solve the organizational
in computer processing and AI research, have afforded business problems that makes a difference to value cre-
organizations a unique opportunity to use AI to infuse ation. The strategic decision-making framework for AI
their products, lower product and service costs, optimize investments and initiatives provides a clarity of purpose
their business processes, and serve their customers bet- concerning how AI can be used to maximize value. Second,
ter (Lee and Shin 2020). Firms developing specialized AI what problems an organization solves is dictated largely by
capability are beginning to see the positive financial per- their business strategy and the desired business outcomes.
formance (Mikalef, Fjørtoft, and Torvatn 2019). Mikalef Therefore, there must be a tight integration between the
and Gupta (2021) surveyed 143 firms and found that AI business strategy and AI initiatives in achieving busi-
capability is positively associated with both organizational ness goals. Finally, not every company’s capabilities are
creativity and financial performance. At this intersection the same. Therefore, if a company’s strategy reflects a
of the promise of AI and the current trends showing the vision/goals that is/are beyond the reach of their current
challenges involved with the realization of AI investments capabilities to implement the necessary AI technologies,
lies an important question for organizations: how can then the company must consider several alternatives to
AI be used effectively and optimally to improve business acquiring or building these capabilities to reach their
outcomes? business outcomes. Therefore, a strategic decision-making
The question this article addresses is, why are businesses framework for AI investments and initiatives—in addi-
failing to derive the full value from their AI investments? tion to providing a roadmap for identifying, implementing,
In this article, we argue that to derive value out of AI and evaluating AI projects which are tightly connected
technologies, organizations must be effective and optimal with their business strategy—offers clues for the trade-offs
with their decision making on AI. This is only possible necessary to acquire/build AI-related capabilities.
through good understanding of AI technology and clarity Our review of prior research on AI frameworks suggests
concerning how AI can solve business problems. The goal that the research so far has been limited to organiza-
of this paper is to provide a prescriptive, decision-making tional decision-making structures, ethics frameworks, AI
framework for value maximization of AI investments and principles for society, and AI frameworks for account-
initiatives. Our framework provides a helpful roadmap to ability. Shrestha, Ben-Menahem, and Von Krogh (2019)
conceptualize the role of AI technologies in supporting conceptualized a framework which provides insights on
organizational goals. Our framework suggests that a how to optimally combine human decision-making with
firm’s business strategy must drive AI-driven business AI decision-making to improve the quality of organiza-
outcomes and measurements, which in turn should drive tional decision-making. The three structures they suggest
the implementation decisions. We also propose using a are: full human to AI delegation, hybrid between human
performance management system such as objectives and and AI, and an aggregated human-AI decision-making.
key results (OKRs) to ensure that the business and AI Lysaght et al. (2019) propose an ethics framework for
goals and objectives are tightly aligned and transparent big data in healthcare, in which they identify relevant
across the company, and the AI efforts are approached values to be used by clinical decision-makers to bal-
in an integrated manner by the different parts of a firm. ance between the potential harms created by algorithmic
AI MAGAZINE 69
FIGURE 1 A strategic decision-making framework for artificial intelligence (AI) investments and initiatives.
bias and using AI to generate public health benefits. framework for AI investments and initiatives in Figure 1.
Floridi and Cowls (2019) conceptualized a unified AI The goal of our framework is to help organizations max-
framework consisting of five principles for AI in society. imize value from their AI. To achieve this objective, our
Their framework aims to establish ethical principles for the framework suggests that organizations must have clarity
adoption of AI for benefiting society. The five principles about their business strategy. Once clear, the business
they suggest are beneficence, non-maleficence, autonomy, strategy drives the development of AI-driven business
justice, and explicability. Berscheid and Roewer-Despres outcomes and measurements, which in turn direct AI
(2019) propose a framework for accountability in AI-based model implementation decisions. Finally, rigorous focus
decision-making systems. Their framework aims to make on measuring the value from AI initiatives will help orga-
developers and clients of AI systems jointly responsible nizations evaluate the effectiveness of AI deployments
for the creation and use of AI systems. They suggest the and offer feedback on improvements needed. In this
creation of AI-validation document by developers and a framework, the decisions a firm has to take flow linearly
deployment disclosure document (DDD) by clients. Their from top-down and are boundary-spanning and interde-
rationale is that the creation of such documents repre- pendent. We describe the elements of our AI framework
sents a formal regulation system for the development in more detail below.
and use of AI systems. However, none of these frame-
works focuses on how to make decisions to maximize
investments from AI and our paper contributes in this Business strategy
area.
Within the framework, a company’s first decision is to clar-
ify their business strategy. Competitive business strategy is
THE STRATEGIC DECISION-MAKING about being different which means deliberately choosing a
FRAMEWORK FOR AI different set of activities to deliver a unique mix of values
(Porter 1996). An effective business strategy intends to take
For this paper, we define AI as “a system’s ability to a firm from where they are currently to where they intend
interpret external data correctly, to learn from such data, to go in a certain timeframe.
and to use those learnings to achieve specific goals and A well-defined strategy has three components—goal,
tasks through flexible adaptation” (Haenlein and Kaplan scope, and a firm’s competitive advantages. For instance,
2019, p. 5). Scholars have variously described AI as either Edward Jones, a St. Louis-based brokerage firm has a suc-
disruptive or revolutionary because of its immense poten- cinct strategy statement: Jones aims to “grow to 17,000
tial to alter society’s way of functioning. AI is considered financial advisers by 2012 [from about 10,000 today]
a disruptive technology because it has “. . . the potential by offering trusted and convenient face-to-face financial
of creating vast changes in the way we behave and go advice to conservative individual investors who delegate
about our daily activities. Disruption is a fact of life their financial decisions, through a national network of
that continually occurs whenever new inventions and one-financial-adviser offices” (Collins and Rukstad 2008).
processes enter the marketplace” (Girasa 2020, p. 4). Prior In contrast, Google appears to change their strategy from
researchers have placed AI in the category of “revolution- being the world’s information organizer to becoming a
ary” because of their ability to change the dynamics of provider of tools to help individuals increase their knowl-
the techno-economic paradigm of a society (Coccia 2021). edge, health, happiness, and success (Rosoff 2019). It is
AI is a broader domain that includes machine learning possible that some businesses have well-defined strategies
(ML) and deep learning as a subset, and the scope of and others do not, but in general, the executives are driving
this paper mostly concerns systems that implement this their firms’ efforts in a certain direction. These larger goals
subset (termed hereafter as AI/ML technologies) (Lee form the locus for AI-related decision-making (Collins and
and Shin 2020). We present our strategic decision-making Rukstad 2008).
70 AI MAGAZINE
Why is a well-crafted business strategy important for AI to increase their efficiency by 20X compared to manual
initiatives? A strategy enables a company to identify its processing and saved them approximately 20,000 euros
path and clarify its competitive differentiated positioning, per month in terms of recycling efficiencies (Zenrobotics
make trade-offs, and create a fit among activities (Porter 2022).
1996). For instance, if a company chooses to pursue growth,
profitability is prioritized less. Similarly, if a firm decides to
focus on institutional clients, they may ignore retail cus- AI-driven business outcomes and
tomers (Collins and Rukstad 2008). In addition, a firm measurements
can identify and build the necessary resources to develop
competitive advantages (Barney 1991). The next step in our framework concerns AI-driven busi-
Knowing the strategy and understanding the trade-offs ness outcomes and measurements. In this step, it is
are important for an organization to identify and imple- important for businesses to ask: What business outcomes
ment a well-focused inventory of AI initiatives that are are important? How can we measure them in the context
well-aligned for actualizing the business strategy. Prior of an AI initiative? Business outcomes flow from the busi-
research organized AI systems into seven types (Martínez- ness strategy. A well-defined business strategy usually has
Plumed, Gómez, and Hernández-Orallo 2021). The first clearly established goals. The subsidiary goals are derived
AI type is based on knowledge representation and reason- from the major goals. Achieving business value means
ing. Also known as rule-based AI, these systems embed achieving these business outcomes (Fountaine, McCarthy,
human expertise with a knowledge database to automate and Saleh 2019; Hunter and Westerman 2009). Having a
decision processes. An example of this AI is IBM Watson clear idea of business outcomes is essential for three rea-
Health. The second type of AI is learning-based, in which sons. First, knowing business outcomes lets stakeholders
computers learn in a statistical or numeric fashion. Predic- know what success looks like in business terms. Sec-
tive models using ML to understand human behavior (e.g., ond, it helps in identifying the appropriate AI projects to
fraud, loan risk, etc.) represent one example of learning- achieve these outcomes. Third, after the AI projects are
based AI. The third type of AI is focused on communication implemented, a company can measure and evaluate if the
between humans and machines through natural language. intended benefits are achieved.
An example of this AI type is a natural language-based sys- Organizations have experimented with many goal set-
tem (natural language processing [NLP]) to monitor social ting frameworks to facilitate the execution of a business
media to analyze customer reviews. The fourth type of AI strategy into implementation by keeping the focus on goals
is related to perception, wherein computer systems inter- consistent throughout all levels of a company. Manage-
pret data from sensors to perceive the environment around ment by objectives (MBO) and OKRs are a few examples
them. An example of this type of AI is a facial recognition of the frameworks used by organizations in this regard
system. The fifth type of AI specializes in planning wherein (Greenwood 1981; Doerr 2018). It is beyond the scope
the system is concerned with the realization of strategies or of this paper to discuss the merits and disadvantages of
optimization solutions to be executed by agents. Planning these frameworks, but for illustration purposes, we use the
algorithms in logistics have been around for some time OKR framework. This framework is used by many com-
whereas similar algorithms have achieved much success panies such as Google, Spotify, Walmart, The Guardian,
in chess, Go, and poker. The sixth type of AI specialization etc. (Castro 2018). OKR stands for objectives and key
is in physical interaction or robotics. Driverless cars and results. Objectives describe what is to be achieved. They
robotic vacuum cleaners fall in this area. The seventh type intend to be significant, concrete, and action-oriented,
of AI focuses on social abilities wherein collective intelli- but not necessarily measurable. The key results specify
gence emerges out of cooperation from a decentralized but how to achieve the objectives. They should be designed
self-organized group of agents. An example of such AI is a to be quantifiable, verifiable, time-bound, and realistic
recommender system wherein group intelligence emerges (Doerr 2018). For instance, consider a fictitious firm whose
from collaboration. business strategy’s objective is to gain the largest market
A well-articulated business strategy can help a company share in a region. To measure the objective, the com-
make an appropriate decision on which type(s) of AI pany could define some key results as: (1) launch a new
systems to invest in. When Lundstams, a Swedish-based product, and (2) acquire 100,000 new customers for prod-
waste management company articulated its strategy to uct X (Castro 2018). This company level OKR can next
be focused on sustainability, it chose to use robotic-based cascade to functional level OKRs. For instance, the sales
AI systems to automate and advance its waste sorting and customer service function could take key result 2
operations. The use of an AI system enabled Lundstams (related to acquiring new customers) as their objective
AI MAGAZINE 71
FIGURE 2 An exemplar of OKRs as a performance management system for AI initiatives. AI, artificial intelligence; OKR, objectives and
key result.
and define multiple key results under that objective as the to identifying the appropriate AI algorithms and technol-
following: ogy to implement. Following up on our earlier example,
Key result 1: Improve net promoter score (NPS) from 8 for the objective “improve net promoter score from 8 to
to 9. 9,” some helpful AI questions include: “how can a net pro-
Key result 2: Increase e-commerce conversion rate from 4% moter score be predicted?” and “what factors best predict
to 5%. net promoter score?” Some business problems are easier
Key result 3: Maintain customer acquisition costs under to translate into AI questions than others. In this context,
$20. a close collaboration between sales & marketing domain
Similarly, other functional areas such as marketing, experts and AI/ML engineers becomes essential. The key
human resources, operations, data science, and IT could decisions in this regard are: What business problem(s) can
similarly define their OKRs in line with the company’s be solved by AI? How can a business problem be translated
OKRs. While the process of OKR building is nuanced, into AI questions?
the idea here is to establish a system of goals and mea-
sures (also known as a performance management sys- ∙ Key performance indicators (KPIs)
tem) that aligns all AI activities within the company
with their business strategy. Such a system creates an A KPI is a “high-level measure of system output, traf-
ideal support to the implementation of our AI decision- fic or other usage, simplified for gathering and review on
making framework for extracting value from AI initiatives. a weekly, monthly or quarterly basis” (Gartner 2021). KPIs
Please see Figure 2 which could be used as an exemplar are important for several reasons. First, KPIs allow firms
of OKRs as a performance management system for AI to track progress toward goals set by frameworks such as
initiatives. OKRs (Kohavi, Tang, and Xu 2020). In fact, prior research
suggests that companies that make use of KPIs earn a 2.95%
∙ AI questions higher return on assets and a 5.14% higher return on equity,
yet less than 25% of firms have such processes in place
Given the OKRs a company intends to achieve, the AI (Proctor and Smith. 2017). Second, the KPIs themselves
initiatives should be focused on how best to help a firm may be used sometimes as variables to be optimized within
reach its goals. As a company gets to work on achieving the AI/ML models. Interested readers can refer to Gart-
key results, many business problems unfold. Some of these ner’s business value model, which defines hundreds of key
problems may be best addressed by AI/ML models. How- metrics used by businesses (Kandaswamy and Furlonger
ever, these business problems must be first translated into 2021).
AI questions. This is important because not every busi- From our earlier example, if a company has an objec-
ness problem may be solved by AI. It is also important to tive to improve the NPS from 8 to 9, naturally the company
understand what questions AI can handle and what it can- must keep track of this metric. Also, to create an AI/ML
not (Ng 2018). Second, having clear AI questions can lead model, the NPS variable can serve as a response variable
72 AI MAGAZINE
(or traditionally referred to as a dependent variable in technologies allows companies to quickly process and
regression models). Also, NPS can be potentially used in make sense of vast amounts of data (Hagiu and Wright
modeling in other relevant AI contexts such as customer 2020). For instance, when Netflix entered the content
retention, sales opportunity index, customer interaction business, it made use of its data on twenty-seven million
index, etc. (Proctor and Smith 2017). In addition, advanced U.S. subscribers and 33 million subscribers worldwide. By
AI/ML models which are sufficiently trained may uncover using subscribers’ viewing history, searches, and ratings,
novel KPIs that might be undiscoverable by the AI teams Netflix decided to create a successful U.S. adaptation
on their own (Kiron and Schrage 2019). of the British show House of Cards (Laney 2020). The
Furthermore, it is possible to evaluate business out- AI applications can use large volumes of data, but they
comes with AI metrics. For instance, when building a increasingly are able to use both the structured inputs such
model to predict the NPS, in addition to simply predicting as sales data, inventory data, web browsing metrics, etc.,
an NPS number, we may measure the relative importance and the unstructured inputs such as audio, video, images,
of different predictors of NPS that may be estimated as text data coming from social media, blogs, IoT devices, etc.
their regression coefficients in a regression model. Using (Canhoto and Clear 2020; Paschen, Pitt, and Kietzmann
variable selection methods, we could also ascertain the rel- 2020). While big data and AI offer a unique promise, the
ative impact of each predictor on the response variable and possibility of creating a unique advantage increases when
consequently zeroing on a fewer number of factors that the data is valuable and proprietary. For instance, the
businesses may be able to act on. Similarly, if a business is firm Mobileye sells its advanced driver-assistance (ADAS)
concerned about the retention rate as the key business out- systems aimed at providing collision-prevention and lane
come, in addition to using a classification model to predict warning systems to many car manufacturers. Because their
probability, we can use measures such as the AUC score systems are used by a large number of users, the system
(area under the curve) for evaluating different models and raised its accuracy to 99.99% giving the company a unique
lift ratios for performing ranking analysis. Such analyses advantage (Hagiu and Wright 2020). Therefore, it is impor-
using AI metrics can provide insights into the interplay tant for businesses to think critically about how data used
between various business factors and business outcomes, in AI applications can create and maximize unique value.
in addition to helping a firm predict a variable of interest To get the data component of the AI initiatives right,
and selecting a subset of customers who are most likely organizations must pay attention to a few important
to contribute to a business outcome (Shmueli, Bruce, and aspects. First, the data needed by AI/ML teams often exist
Patel 2016). in different IT systems in several formats making it dif-
For the purposes of our framework, the important ficult for AI/ML teams to access the necessary data in
decisions businesses need to make are: What KPIs can con- time (Campbell et al. 2020). Second, many AI/ML engi-
tribute to answering an AI question? What business and AI neers exhibit an inherent preference for doing model work,
metrics can track the performance of the AI model? but not the data work. This may lead to data cascades—
compounding events leading to a negative downstream AI
system’s performance from data issues (Sambasivan et al.
Implementation decisions 2021). For example, poor data practices at the upstream
processes led to IBM’s cancer treatment AI’s reduced data
The key decisions in implementation involve data and accuracy and a similar issue led to the inaccurate predic-
governance, models and their implementation, technology tions of Google flu trends’ AI system (Sambasivan et al.
tools, and people skills. We discuss each of these below. 2021). Third, consumers and regulators are increasingly
showing concerns about data privacy and accountability
∙ Data and governance issues. Both General Data Protection Regulation (GDPR)
and California Consumer Privacy Act (CCPA), in addition
The implementation of AI/ML models need data. Once to a slew of local state laws, make it mandatory for firms
a business decides what AI question(s) to answer and what to maintain high standards regarding privacy (Hamilton
metrics to use, the next set of decisions concerns: What and Sodeman 2020). Given these imperatives, organiza-
constitutes the relevant data for this problem? How can tions must make important decisions about the processes
this data be acquired and made available? How can a com- and governance mechanisms related to data. Based on our
pany ensure the data used is devoid of bias and meets the AI decision-making framework, identification of impor-
personal data protection standards imposed by the relevant tant business problems and the associated KPIs and AI
regulatory frameworks? metrics is critical for identifying the needed data.
Using customer information for making decisions is Continuing with our earlier example, to create an
not a new strategy, but the emergence of cloud and new AI/ML model to predict the NPS, data scientists first need
AI MAGAZINE 73
Dashboards, used for display of visualization, are not only

an essential part of business intelligence used for report-
ing purposes, but are also used to display results from
ML models for explainability purposes. Techniques such
as basic text mining can be used for unstructured data
with a small data size. Particularly, applications related
to labeling a document as belonging to a class are suit-
able for this type of analysis (Shmueli, Bruce, and Patel
2016). When the data size is big (e.g., millions of rows or
more) and when the data is structured, many ML models,
such as linear regression, support vector machines, logistic
regression, including deep learning methods such as arti-
ficial neural nets can be employed. Finally, when the data
size is big and the data is unstructured, models related to
advanced deep learning, NLP, and natural language under-
standing (NLU) can be employed (Wilson and Daugherty
2020). While working with dashboards and basic statisti-
F I G U R E 3 Data versus AI/ML methods typology (adapted
cal analysis requires moderate expertise, applying AI/ML
from Ramesh 2017). AI, artificial intelligence; ML, machine
methods which work with unstructured and/or big data
learning.
requires advanced skills.
To implement an AI/ML model, a business needs to
to identify relevant data. While domain knowledge and the make a few key decisions. First, how accurate must an
context in which a company operates is important, data AI/ML model be before it is considered for implemen-
scientists can also look at the previous research done in tation? Accuracy measures how well the AI/ML model
this area to identify relevant variables. For instance, for predicts the output variable, and different AI contexts
modeling NPS in a mobile telecom industry, researchers may require different AI/ML accuracy metrics. Consumers
identified nine customer experience attributes related to demand higher prediction accuracies for cancer detection
tariff plan, network voice data, network plan data, billing, and object detection for self-driving cars, while lower pre-
etc. (Markoulidakis et al. 2020). Once relevant data is iden- diction accuracies may be tolerable for a voice-controlled
tified in this manner, firms can turn their attention to the virtual assistant (Agarwal, Gans, and Goldfarb 2018). Sec-
next step of our framework. ond, is it important for businesses to interpret how an
AI/ML algorithm works for users and regulators? For
∙ Models and implementation instance, financial firms are held accountable under the
Equal Credit Opportunity Act (ECOA) to provide a reason
Once a business identifies the AI question, KPIs, and to consumers for denying a credit-card application. Use of
the appropriate data, the next decision is: What AI/ML AI/ML models which are not interpretable in such con-
model(s) needs to be developed? The AI question, the texts could be problematic. On the other hand, consumers
type of data and the structure of data serve as criti- may not care about the inner workings of a chatbot as long
cal inputs to selecting an appropriate AI/ML model. Lee it serves their needs well (Han 2021; Lee and Shin 2020).
and Shin (2020) categorize machine-learning algorithms Third, how can the loop be closed between an AI/ML
into supervised, unsupervised, semi-supervised, and rein- model’s predictions and decisions? For instance, the diag-
forcement learning methods. We limit our discussion to nostic measures provided by the cancer detection AI/ML
supervised methods, where a model is built for an out- models for a tumor are often probabilistic such as: “66%
put variable (Y) using a set of input variables (X’s). Some benign, 30% malignant, 4% no disease.” After an AI/ML
of the popular supervised methods include linear regres- model generates its predictions, additional judgment may
sion, logistic regression, naïve Bayes, K-nearest neighbors, be necessary to make a decision on whether or not to
neural networks, etc. (Shmueli, Bruce, and Patel 2016). send the patient for a biopsy. Should the AI/ML system
We provide a typology in Figure 3 for selecting an ML make that decision or should humans be involved? (Agar-
method based on data type and data structure (Ramesh wal, Gans, and Goldfarb 2018). Delegating these decisions
2017). When an organization has structured data and the solely to the AI/ML teams may not be a good idea for
data size is small (e.g., a hundred to a few thousand businesses.
rows), they can make use of statistical analysis such as For the example presented in this paper on creating an
classical regression for creating basic prediction models. AI/ML model for predicting NPS, the AI/ML analysts may
74 AI MAGAZINE
not have to create a model from scratch. Often, related and difficulty in hiring are consistently showing up as bar-
research is available in published outlets. In some con- riers faced by organizations in their AI deployment efforts
texts, the AI/ML analysts could reuse the models to apply (Wiggers 2021). Therefore, managers need to be mindful of
it in their organizational context instead of trying to rein- using novel hiring practices to attract the best talent.
vent the wheel. For instance, Markoulidakis et al. (2020) Continuing with our prior example, if the company’s
published a logistic regression-based algorithm to improve AI/ML team decides to try out several AI/ML and deep
the NPS predictions, and their method showed significant learning models, then making decisions on what technol-
improvement over prior methods. Once the NPS model ogy tools are needed will become important. Decisions
is developed, the next question is how to use it. In sur- concerning which standard AI tool frameworks to use,
veys, only about 15%–20% of customers provide data on how frequently AI models need to be updated, how to track
the NPS measure. Therefore, the NPS prediction model AI-model performance, how to create production-ready
can be used to predict the NPS scores for the remaining pipelines, whether a high-performance computing cluster
customers. The model’s results can also aid in identify- is needed for AI workloads, etc. become important (Balakr-
ing factors that can be leveraged for improving customer ishnan et al. 2020). Additionally, how to structure teams
experience (Parthasarathy 2021). for developing the models, whether to involve members
from other functional areas, and how to explain the find-
∙ Technology tools and people skills ings to the internal and external stakeholders also become
pertinent.
The next implementation decisions in our framework
concern the type of technology tools needed for AI deploy- ∙ AI project evaluation
ment and the type of people skills needed. Implementation
of different types of AI/ML models may necessitate dif- The next decision in our framework requires organiza-
ferent technology tools and computing resources. While tions to evaluate AI projects for the benefits they confer
structured data management relied on traditional rela- to the organizations. A process for accurate measure-
tional DBMS and data warehouses, new digital platforms ment of benefits and communicating them to sponsors
are emerging to support AI/ML applications. Examples and stakeholders is important to realize how AI initiatives
of these platforms include open-source frameworks and contribute to business performance (Hunter and Wester-
libraries for ML models (e.g., Google TensorFlow), applica- man 2009). In a survey conducted by Mckinsey, 22% of
tion program interfaces (APIs) for services such as vision, respondents say that more than 5% of their revenues were
speech, language, search (e.g., Google’s Cloud Translation attributable to the use of AI, whereas 48% of them reported
API), drag-and-drop models to develop, test, and deploy revenues of less than 5% from AI. Revenue increases
AI models (e.g., Azure ML Studio), and pretrained models were reported to be more common than cost decreases
for users to deploy in their contexts (Rai, Constantinides, (Balakrishnan et al. 2020).
and Sarker 2019). However, surveys suggest that compa- In measuring benefits from AI projects, organizations
nies with a strong base in core digital technologies (cloud, must make an important distinction between hard and
mobile, and web) and advanced technologies (big data soft benefits. If a benefit can be measured in financial
and advanced analytics) are more likely to adopt and inte- terms in a way that relates to a profit and loss (P&L) state-
grate the next generation of AI platforms (Bughin and ment, then it is considered a hard benefit. However, a
van Zeebroeck 2018). In another survey by Brock and von soft benefit can only be measured in terms of operational
Wangenheim (2019), it was found that firms with stronger terms and may not contribute directly to the P&L state-
digital skills anticipated higher AI-induced impact com- ment (Hunter and Westerman 2009). Organizations must
pared to firms with weaker digital skills. Additionally, measure both the hard and soft benefits. Continuing with
leaders in AI deployment tend to develop AI/ML applica- our example, how to relate the NPS prediction model to
tions in-house as opposed to buying them, and they also improved customer experience and eventually to increased
employ more AI specific talent, such as data engineers, sales becomes an important question. Managers first need
data architects, and translators (Balakrishnan et al. 2020). to determine what the NPS model suggests as key vari-
For instance, companies such as Capital One planned to ables to manipulate to improve the NPS. If the NPS model
hire a large number of tech workers with many of them suggests “live chat support” as one of the key determi-
expected to specialize in AI. The company recently com- nants of NPS score, then the managers must examine the
pleted shifting its data, applications, and IT systems to current response times. For instance, if the current aver-
Amazon Web Services and wants to strengthen its ability age response time for live chat is 8 min, then through a
to leverage the integrated cloud data using AI-focused hir- data science A/B experiment, they could manipulate the
ing (Castellanos 2021). However, lack of competent people live chat times for a sample of customers to examine its
AI MAGAZINE 75
effect on NPS. In this manner, managers could improve 2018) focused on three D’s: (1) enhancing digital technolo-
certain business processes to increase NPS. Also, using pre- gies to elevate customer experience, (2) making delivery
dictive models, managers could link NPS score to sales. For more convenient, and (3) making customer experience
instance, a regression model developed by the firm shows faster at drive-thru (Markoulidakis et al. 2020).
that an increase of 10% in NPS is associated with a 3% Although enhancing digital technologies is essential for
increase in sales. Using such model predictions, managers any company, McDonald’s kept their focus firmly on con-
could translate the soft benefits into financials (Hunter sumer experience. Daniel Henry, the CIO of McDonald’s
and Westerman 2009). Also, it is important to tie the mea- suggested, “when you’re a product-obsessed organization,
surement of AI benefits to the key results identified in you think about the consumer before the technology.
section “AI-driven business outcomes and measurements” Technology by itself means nothing” (McKinsey 2020).
(improve NPS from 8 to 9). At the end of the year, man- McDonald’s decided to spend $6 billion dollars on the
agers can report to what extent the key results from AI “experience of future” project which included store remod-
are achieved. Table 1 summarizes the decisions a com- eling to include touch screen ordering systems, digital
pany must make using our AI framework and why such menu boards, and a global app and mobile ordering sys-
decisions are important for an AI initiative. tem (Klein 2018). These digital innovations, in addition
to providing conveniences for quick ordering and deliv-
ery for customers, offer another important advantage for
A USE CASE ON McDonald’s USE OF AI AI: customer-specific order data. This data is essential
for offering personalized recommendations for customers.
In this section, we discuss McDonald’s use of AI in line When Apple and Google started allowing people to opt out
with the framework presented in this paper. We con- of tracking when using apps, it threatened companies’ abil-
ducted a strengths, weaknesses, opportunities, and threats ity to get the first party data. Therefore, in August 2021,
(SWOT) analysis on McDonald’s to understand the reason- McDonald’s rolled out a customer loyalty rewards program
ings behind McDonald’s business strategy and the use of AI with an eye to get customer data directly (Sebastian 2021).
to exploit business opportunities. Formed in Illinois in 1956 In line with their business outcomes, one of the AI
and incorporated in Delaware in 1965, McDonald’s is one of projects McDonald’s recently started addressed the ques-
the globally recognized brands with operations in 119 coun- tion: “how do we offer the best personalized recommen-
tries. Approximately 36,000 out of their 39,000 restaurants dations for our customers?” In 2019, McDonald’s acquired
are franchised. The core strengths of McDonald’s lie in Dynamic Yield, an Israeli-based company with expertise
the distinct capabilities that facilitate/build high customer in AI. Dynamic Yield specializes in using AI/ML tech-
satisfaction scores on their menu items and their high nologies to increase personalization and improve customer
brand value (Mergent 2022). The weaknesses in McDon- service. As opposed to static menu boards, McDonald’s
ald’s operations lie in the challenge of lacking/weaker started using digital menu boards powered by AI/ML algo-
capabilities for navigating the complexity of operating a rithms, which display food items based on the time of
huge network of franchises, quickly responding to con- the day, weather, current restaurant traffic, and trending
sumer preferences in a wide variety of geographic areas, menu items. They can also suggest additional items to
and coordinating their supply chains (Mergent 2022). The a customer based on the initial selections (McDonald’s
opportunities lie in growing their customer numbers. As Corporation n.d.). In terms of business outcomes, only
of 2017, in their nine largest markets, McDonald’s had a McDonald’s is privy to internal data concerning the value
15% share of the 89 billion annual visits to Quick Service generated by these projects, but the initial data appears
Restaurants (Mergent 2022). The threats lie in the intense promising. McDonald’s former CEO, Steve Easterbrook,
competition that the Quick Service Restaurant industry suggested that the use of recommendation algorithms
faces and the changes in consumer preferences. The SWOT built into the drive-thru menu boards have resulted in
analysis clearly suggests capabilities that McDonald’s pos- larger orders (Yaffe-Bellany 2019). McDonald’s first quarter
sesses or needs to build to match for opportunities for results for 2021 also beat analysts’ expectations with a rev-
growth and the use of AI as a tool to accomplish such goals. enue of $5.12 billion as opposed to an expected $5.03 billion
In 2017, McDonald’s launched its new business strategy (Lucas 2021).
called the Velocity Growth Plan, which aimed to achieve While KPIs such as order size, customer visits, rev-
growth through increased guest counts. The strategy is enue, etc. can be reliably measured, it is not always easy
customer-focused in that its outcomes include retaining to measure some strategic KPIs McDonald’s intends to
existing customers, regaining lost customers, and con- use such as, “guest satisfaction” and “whether McDon-
verting casual customers into committed customers. To ald’s is a family destination which appeals to parents.”
operationalize this new strategy, (McDonald’s corportion McDonald’s uses a lagging indicator such as “visits by
76 AI MAGAZINE
TA B L E 1 Decisions to make in the AI framework.

Decision to make Why is the decision important for an AI initiative?
What is our business strategy? Strategy provides a direction for a destination a company wants
to reach, and AI can help a company get there on a timely
basis.
What are our AI-driven business outcomes and Business outcomes: knowing business outcomes lets
measurements? stakeholders know how success looks like in business terms.
AI questions: AI questions translate business problems into AI
problems.
KPIs and AI measures: KPIs help translate business goals into
quantifiable definitions and metrics. They also help firms keep
track of their success or failures. AI metrics such as AUC,
accuracy rates, lift ratios, R-squared, and MAPE are helpful in
evaluating predictive models and ascertaining which of the
model’s performance measures is helpful in meeting business
goals.
What are our implementation decisions? What data is needed? What governance mechanism must be in
place?
Organizations carry vast amounts of data. Organizations should
make available the data that is purposeful in answering the AI
questions. The governance mechanisms help provide quality
data for AI/ML projects and help meet regulatory and privacy
guidelines.
What AI/ML models to build? How are they to be implemented?
Not every model is equal in terms of their difficulty or the value it
adds. Careful selection of AI/ML models is important for
solving problems and at times for explaining how the model
works.
Also, implementation of models requires a deep understanding
of how they must be integrated into business process and how
the employees must adapt to the new changes.
What technology tools to use?
Different AI/ML algorithms may need different technology tools.
Knowing the possible options helps decide between buy versus
outsource choices.
What skills we need?
Different types of AI projects need different types of people skills.
The skills could be both technical and business-related in
nature. Hiring competent AI resources is critical to AI project
success.
How do we measure value from AI projects? Establishing a process with a focus on measuring value from AI
projects, communicating results to the internal and external
stakeholders, and tying the results to actions is important.
AI, artificial intelligence; KPI, key performance indicator; ML, machine learning.
families with kids under the age of 13” and a leading indi- service-restaurants (QSRs) using Twitter data suggests that
cator “a place I’m happy to bring my children” to get McDonald’s outperformed the other restaurants in terms
a pulse on their family destination outcome (Kiron and of positive sentiment (Nawaz et al. 2018).
Schrage 2019). While customer surveys shed some light Another AI project piloted by McDonald’s intends to
on these outcomes, they may be prone to selection bias. answer the question: “Can we use machines to understand
The use of ML methods such as sentiment analysis, which and converse with customers to take orders?” McDon-
uses online customer reviews data (e.g., Twitter posts), ald’s is testing AI-enabled automated voice order-taking at
offers an additional source of insight. Recent research 10 drive-thrus in the Chicago area. This system is aided
which performed sentiment analysis on six major quick- by the technology provided by Apprente, a voice-based
AI MAGAZINE 77
technology company acquired by McDonald’s in 2019. Second, the challenge of developing a clear business
Chris Kempczinski, the CEO of McDonald’s, suggests that strategy is often primarily an organizational one and
the system is about 85% accurate. Kempczinski mentioned depends on effective leadership. With so many forces at
that they had to concurrently train human assistants work making choices and tradeoffs difficult in organiza-
because as soon as there was a question or an issue with tions, a clear intellectual framework to guide business
the AI system, the employees immediately stepped in. This strategy is a necessary counterweight. Therefore, the
caused the system to get less overall data points. The CEO top leadership must be actively involved with the AI
believes that it might take about 5 years for this system to projects to set the goals, identify business problems, and
be ready to scale up. As for the technologies and skills, to assess the value from such efforts. Third, even though
Kempczinski suggested that instead of building internal technology plays an important role, AI projects require
capabilities for these advanced technologies, they may rely closer collaboration across functional areas. When the
on acquisitions and partners to implement and scale such models identify key predictors, and suggest interventions,
types of technologies (Ruggless 2021). In fact, McDonald’s implementing such changes may require close collab-
created an internal division called McD Tech Labs, and oration among different business units. Finally, we see
appointed the co-founder of Apprente as the VP and the that the big tech firms are spearheading the AI efforts.
firm’s employees as its core team. McDonald’s plans to Many of these firms deal mostly with unstructured data,
expand Tech Labs by hiring additional engineers, data and by using their strengths in software development and
scientists, and other technologists (Dallke 2019). As for rapid innovation, they have created many AI tools and
data governance issues, although McDonald’s suggested frameworks. The other businesses do not necessarily have
that they take data security seriously by paying attention to invent AI tools, but they must find ways to strategically
to the GDPR and the CCPA (McKinsey 2020), they were apply these systems to their businesses. This is where
recently hit with a lawsuit by an Illinois customer who firms must be ready with their business strategy and
alleged that McDonald’s had collected his voice data with- outcomes and identify important business problems to be
out consent when he ordered through the voice-enabled AI solved by AI. Then through effective and optimal strategic
system. The lawsuit alleges that McDonald’s collection of decision-making on AI investments and initiatives, they
voiceprint information without customer consent is in vio- are better prepared to reap the benefits of AI. Firms with
lation of the Illinois’ Biometric Information Privacy Act, no prior exposure to AI can use our framework and start
or BIPA. This lawsuit is a stark reminder that any short- with a pilot project. Once they start realizing the value
comings in data governance efforts of organizations will from AI, they can ramp up their AI investments and
be quickly exposed as they move towards implementing AI initiatives.
technologies (Maze 2021). We summarize how the differ- One of the important questions concerning the manage-
ent elements of our AI framework have been incorporated ment of innovations, such as AI projects discussed in the
in McDonald’s use case in Table 2. article, is how firms sustain their competitive advantage in
markets. In this context, the study of sources of technolog-
ical breakthroughs is important. Past research by Coccia
DISCUSSION (2017) suggests that the recognition of unsolved problems
and a strong drive to find solutions induce the emergence
In this paper, we make a case for organizations to adopt and development of radical innovations. We agree with
a strategic decision-making framework for AI investments this view and suggest when AI, applied to challenging and
and initiatives to successfully deploy their AI projects. unsolved problems, has the potential to innovate and cre-
We provide the framework in terms of a flow chart in ate competitive advantages for firms. The second source of
Figure 4 as an aid for managerial decision-making. As our innovation could come from radical innovations adopted
framework shows, AI initiatives cannot be isolated from in research labs by high-skilled scientists who experiment
a firm’s business strategy, and a tight alignment is impor- and innovate with freedom (Coccia 2016). In the context
tant between the business and AI efforts. In the absence of of AI, this could mean an AI pilot project by an inter-
such a cohesive strategy, AI projects may lead to satisficing disciplinary team to freely explore and innovate to find
(looking for a needle sharp enough to sow with) instead of solutions to important strategic problems. Such freedom to
optimizing (looking for the sharpest needle in the stack) explore, we believe, will aid the insight discovery process
outcomes toward value maximization from AI to achieve in AI contexts and may positively contribute to organiza-
business goals. When companies focus on the operational tional performance. A related question in this context is,
effectiveness of AI without a well-thought and well-crafted why should organizations persist with disruptive technolo-
competitive business strategy, AI implementations and gies such as AI? Prior research by Coccia (2020) suggests
results are highly likely to fail (Porter 1996). that disruptive technologies, similar to the AI efforts
78 AI MAGAZINE
TA B L E 2 Mapping AI decision-making framework to McDonald’s case.

Element of AI framework Example from McDonald’s case
Business strategy Velocity Growth Plan represents McDonald’s business strategy. It
aims to achieve growth through increased guest count,
convenience, and satisfaction.
AI-driven business outcomes and measurements Business outcomes
Higher guest count, larger order size, easier delivery, family
destination which appeals to parents.
AI questions
• How to personalize customer orders?
• How to recommend products based on what customers already
added to their order?
• How to dynamically modify digital menu board based on the
time of the day, weather, current restaurant traffic, trending menu
items, etc.?
• How to automate conversational chatting with customers to take
their orders?
KPIs
Guest count, order size, order amount, revenue, guest satisfaction,
visits by families with kids under the age of 13, family friendliness,
accuracy of orders processed by AI voice-agent, etc.
AI measures
Accuracy rates of automated order-taking, Relative importance of
different predictors in predicting what customers order on a
drive-thru menu, etc.
Implementation decisions Data and governance
McDonald’s in-store kiosks, global mobile apps, online orders collect
individual information. McDonald’s suggests they pay close
attention to data regulation frameworks such as the GDPR and
CCPA. They are planning on using customer loyalty program to
compensate for possible lack of first party data when Google and
Apple allow users to block providing data to apps.
Technology tools
McDonald’s has invested heavily in digital technologies, mainly in
kiosks, global app, digital menu boards incorporating
recommendation systems, and a pilot project on AI-enabled
speech recognition system for order taking.
Modeling and implementation
McDonald’s use of recommender systems in their digital dashboards
and the pilot test of AI-enabled voice assistant for order taking
rely on advanced machine learning methods. Methods such as
natural language processing, deep neural networks, recommender
algorithms, etc. are needed to successfully implement these
projects.
McDonald’s implementation of digital menu boards appears to have
gone well, but they seem to be experiencing some initial issues
with the voice AI-system for order processing. At least one lawsuit
alleges that McDonald’s has not obtained consent for collecting
voice data when ordering through their AI-enabled drive-thru
agent.
People skills
McDonald’s has acquired two companies specializing in AI. Their
McD-Tech labs is found with the members of the AI company
they recently acquired. Their strategy for gaining skilled AI
personnel is to use strategic acquisitions, partners, and new hires.
(Continues)
AI MAGAZINE 79
TA B L E 2 (Continued)
Element of AI framework Example from McDonald’s case
Value from AI projects Only McDonald’s can be privy to this knowledge, but available data
suggests the use of recommendation technology within the digital
menu boards has increased the order size. Their revenues are also
showing growth. Also, the AI voice-assistant is currently reported
to be showing 85% accuracy in correctly identifying orders. Being
one of the first entrants into using AI in the QSR industry affords
them with competitive advantages.
AI, artificial intelligence; CCPA, California Consumer Privacy Act; GDPR, General Data Protection Regulation; KPI, key performance indicator; QSR, quick-
service-restaurant.
FIGURE 4 A flowchart of artificial intelligence (AI) decision-making framework for managers.
suggested in this work, tend to aid disproportionate growth framework to suggest how technological factors (i.e.,
in some markets compared to established technologies, compatibility between needs and technology, a relative
and that the technological cycle of disruptive technologies advantage over prior innovation), organizational factors
tends to have an up-wave phase longer than down wave (i.e., top management support, data, culture, resources),
phase, suggesting an asymmetric shape of a technological and environmental factors (i.e., competitive pressures
cycle. The past research, thus, emphatically supports inno- and government regulations) are important for AI adop-
vation with AI efforts and the current paper contributes tion (AlSheibani, Cheung, and Messom 2018; Bremser,
by suggesting a decision-making framework to maximize Piller, and Rothlauf 2017; Pumplun, Tauchert, and Heidt
value from such investments. 2019). Our framework shares some similarities and dif-
One of the prerequisites for AI value creation is suc- ferences with this framework. While we draw on many
cessful AI Adoption. Prior research used a modified TOE factors from this framework (e.g., data quality, government
80 AI MAGAZINE
regulations, compatibility between needs and technology), We also discuss conditions in which data becomes a
our framework differs in conceptualizing how organi- source of competitive advantages. For instance, if a firm
zational strategy drives the development of AI-driven creates an AI application which continues to gather big
business outcomes, which in turn guide the implementa- data, which then continuously improves the AI program’s
tion decisions. In addition, we pay specific attention to the accuracy, such data becomes a source of competitive
importance of using a performance management system, advantage. We also discuss how connecting predictions
such as an OKRs, in establishing a set of outcomes and to decisions is important within a business process and
including a series of measures that indicate the progress across the functional areas. The different ways in which an
on AI project performance. organization can create a network of AI systems wherein
We believe that a business strategy-based approach to the predictions from one system become inputs into other
AI initiatives is important because while it is reasonable system, and thereby, facilitating organizational efficiencies
for each department within a company to execute its own will be a fertile area of research in the future.
pilot AI project to test the potential of AI, over a period However, our work is not without its limitations. First,
of time, it has the unintended effect of quickly turning our work omitted some human-computer interaction
into a siloed and disintegrated environment. Therefore, aspects of AI. For instance, work by Kellogg and Valen-
tying AI initiatives to business strategy, outcomes, and tine (2022) suggests five common mistakes managers do
departmental goals is important. However, without a per- when introducing AI. As an example, one of the mistakes
formance management system to set goals at the enterprise managers do is to quickly bring in AI to address areas
and departmental levels, and to keep track of AI initia- where it might have the most impact, but if they do not
tives, it is hard to effectively manage the portfolio of AI get support from employees, it will result in potential fail-
projects. In this context, the OKR approach we suggested ure. Rather, if AI’s efforts are focused on aspects of work
will be very helpful. The benefits of OKRs could be signif- that employees do not like, then there is more likelihood
icant in the sense they can be shared across the enterprise of gaining employees’ support for the implementation of
providing visibility to the AI efforts. Moreover, accurate AI. Future work could look at how AI implementation
predictions in one part of the system have rippling effects must go together with employees’ psychological readi-
on other parts, and better coordination also allows deci- ness to embrace AI. Second, companies are using AI to
sions to be optimized over a wider network (Agarwal, innovate concerning employee productivity, but the results
Gans, and Goldfarb 2022). are not often promising. For example, using AI to pro-
vide real-time feedback to improve performance helped
workers making men’s pants improve productivity, but the
CONCLUSION employees making jackets, which involved more complex-
ity in the task and craftsmanship, complained that AI was
Our work provides firms with a strategic decision-making demotivating for them as it failed to adequately capture
framework for AI initiatives to be used with a perfor- the quality of their work (Kellogg and Valentine 2022).
mance management system such as an OKR system to Future research could investigate areas where AI could be
extract value out of AI investments. We provided the effective in boosting innovation. Third, our work did not
use case of McDonald’s as an exemplar for companies focus extensively on model explainability, a key obstacle
to implement our framework. We argue in this work to the implementation of AI systems. Prior research sug-
that it is not the AI technology, but how it is applied gests that a lack of model transparency and explainability
to solve the “important business problems” that creates could hurt the success of AI efforts (Adadi and Berrada
value. We further argue that the decisions on AI imple- 2018). However, research is emerging to suggest different
mentation must be guided by the overall business strategy explainability techniques to enhance the applicability of
and AI-driven business outcomes. Then the next question AI systems. We refer interested readers to the work by
becomes how to translate the overall strategy and out- Adadi and Berrada (2018) in this regard. We believe the
comes into AI initiatives. To bridge this gap, we propose revolution in AI is undoubtedly here; the only question for
an OKR performance management system, which con- firms is whether or not they are ready. The intention of our
sists of objectives and key measures (OKRs) at the firm work is to prepare firms and managers for the compelling
and departments levels. The OKRs then become the basis opportunities inherent when maximizing the value of their
for recognizing relevant AI projects, making them trans- AI.
parent across the business, and integrating the results of
various AI projects across the businesses to derive business CONFLICT OF INTEREST
value. The authors declare that there is no conflict.
AI MAGAZINE 81
ORCID Campbell, C., S. Sands, C. Ferraro, H. Y. J. Tsao, and A. Mavrommatis.

Naveen Gudigantala https://orcid.org/0000-0002-5231- 2020. “From Data to Action: How Marketers Can Leverage AI.”
8980 Business Horizons 63(2): 227–43.
Canhoto, A. I., and F. Clear. 2020. “Artificial Intelligence and
Machine Learning as Business Tools: A Framework for Diag-
REFERENCES
nosing Value Destruction Potential.” Business Horizons 63(2):
Aaser, M., J. Woetzel, and K. Russell, 2021. Five Insights about 183–93.
Harnessing Data and AI from Leaders at the Frontier. McKinsey Castellanos, S. 2021. Capital One Steps Up Hiring Software Engi-
Global Institute. Retrieved from: https://www.mckinsey.com/ neers After Cloud Migration. The Wall Street Journal. https://
capabilities/quantumblack/our-insights/five-insights-about- www.wsj.com/articles/capital-one-steps-up-hiring-software-
harnessing-data-and-ai-from-leaders-at-the-frontier. engineers-after-cloud-migration-11626730997?mod=djemAIPro
Adadi, A., and M. Berrada. 2018. “Peeking Inside the Black-Box: A (last accessed: 19 July 2021).
Survey on Explainable Artificial Intelligence (XAI).” IEEE Access Castro, F. 2018. The Beginner’s Guide to OKR. What is OKR?.
6: 52138–60. Retrieved from: https://felipecastro.com/en/okr/what-is-okr/.
Agarwal, A., J. Gans, and A. Goldfarb. 2018. Prediction Machines. The Coccia, M. 2016. “Radical Innovations as Drivers of Breakthroughs:
Simple Economics of Artificial Intelligence. Boston, MA. Harvard Characteristics and Properties of the Management of Technology
Business Review Press. Leading to Superior Organizational Performance in the Dis-
Agarwal, A., J. Gans, and A. Goldfarb. 2022. From Prediction to Trans- covery Process of R&D Labs.” Technology Analysis & Strategic
formation. Boston, MA. Harvard Business Review Press. Retrieved Management 28(4): 381–95.
from: https://hbr.org/2022/11/from-prediction-to-transformation. Coccia, M. 2017. “Sources of Technological Innovation: Radical
Albrecht, T., T. M. Rausch, and N. D. Derra. 2021. “Call Me Maybe: and Incremental Innovation Problem-Driven to Support Com-
Methods and Practical Implementation of Artificial Intelligence petitive Advantage of Firms.” Technology Analysis & Strategic
in Call Center Arrivals’ Forecasting.” Journal of Business Research Management 29(9): 1048–61.
123: 267–78. Coccia, M. 2020. “Asymmetry of the Technological Cycle of Disrup-
AlSheibani, S., Y. Cheung, and C. Messom. 2018. “Artificial Intelli- tive Innovations.” Technology Analysis & Strategic Management
gence Adoption: AI-Readiness at Firm-Level.” In Proceedings of 32(12): 1462–77.
the 22nd Pacific Asia Conference on Information Systems – Oppor- Coccia, M. 2021. “Technological Innovation.” In The Blackwell Ency-
tunities and Challenges for the Digitized Society: Are We Ready? clopedia of Sociology, edited by G. Ritzer and C. Rojek. Hoboken,
PACIS 2018, Yokohama, Japan. New Jersy. John Wiley & Sons, Ltd. https://doi.org/10.1002/
Balakrishnan, T., M. Chui, B. Hall, and N. Henke. 2020. “The State 9781405165518.wbeost011.pub2.
of AI in 2020.” Mckinsey.Com. https://www.mckinsey.com/ Collins, D. J., and M. G. Rukstad. 2008. “Can You Say What Your Strat-
business-functions/mckinsey-analytics/our-insights/global- egy Is?” Harvard Business Review. https://hbr.org/2008/04/can-
survey-the-state-of-ai-in-2020 (last accessed: 17 November you-say-what-your-strategy-is.
2020). Dallke, J. 2019. “McDonald’s Launches Tech Lab in Silicon Val-
Barney, J. 1991. “Firm Resources and Sustained Competitive Advan- ley after Buying Drive-Thru Tech Startup.” Chicago Business
tage.” Journal of Management 17(1): 99–120. Journal. https://www.bizjournals.com/chicago/news/2019/09/
Bean, R. 2021. “Why Is It So Hard to Become a Data-Driven Com- 10/mcdonald-s-launches-tech-lab-in-silicon-valley.html (last
pany?” Harvard Business Review. https://hbr.org/2021/02/why- accessed: 10 September 2019).
is-it-so-hard-to-become-a-data-driven-company (last accessed: 5 Doerr, J. 2018. Measure What Matters: How Google, Bono, and the
February 2021). Gates Foundation Rock the World with OKRs. New york:NY.
Bergstein, B. 2019. “This Is Why AI Has Yet to Reshape Most Busi- Penguin.
nesses.” MIT Technology Review. https://www.technologyreview. Floridi, L., and J. Cowls. 2019. A Unified Framework of Five
com/2019/02/13/137047/this-is-why-ai-has-yet-to-reshape-most- Principles for AI in Society. Available at SSRN 3831321.
businesses/ (last accessed: 13 February 2019). Fountaine, T., B. McCarthy, and T. Saleh. 2019. “Building the
Berscheid, J., and F. Roewer-Despres. 2019. “Beyond Transparency: AI-Powered Organization.” Harvard Business Review 97(4): 62–73.
A Proposed Framework for Accountability in Decision-Making AI Gartner. 2021. “Key Performance Indicator. Gartner Glos-
Systems.” AI Matters 5(2): 13–22. sary.” Garter.com. https://www.gartner.com/en/information-
Bremser, C., G. Piller, and F. Rothlauf. 2017. “Strategies and influ- technology/glossary/kpi-key-performance-indicator.
encing factors for big data exploration.” AMCIS 2017 Proceedings. Girasa, R. 2020. “AI as a Disruptive Technology.” In Artificial Intel-
4. Boston, MA. https://aisel.aisnet.org/amcis2017/AdoptionIT/ ligence as a Disruptive Technology (pp. 3–21). Cham: Palgrave
Presentations/4 Macmillan.
Brock, J. K. U., and F. Von Wangenheim. 2019. “Demystifying AI: Greenwood, R. C. 1981. “Management by Objectives: As Devel-
What digital transformation leaders can teach you about realis- oped by Peter Drucker, Assisted by Harold Smiddy.” Academy of
tic artificial intelligence.” California Management Review 61(4): Management Review 6(2): 225–30.
110–134. Haenlein, M., and A. Kaplan. 2019. “A Brief History of Artifi-
Bughin, J., and N. van Zeebroeck. 2018. “Artificial Intelligence: Why cial Intelligence: On the Past, Present, and Future of Artificial
a Digital Base is Critical.” McKinsey Quarterly. Retrieved from: Intelligence.” California Management Review 61(4): 5–14.
https://www.mckinsey.com/capabilities/quantamblack/our- Haenlein, M., and A. Kaplan. 2021. “Artificial Intelligence and
insights/artificial-intelligence-why-a-digital-base-is-critical (last Robotics: Shaking Up the Business World and Society at Large.”
accessed: 26 July 2018). Journal of Business Research 124: 405–7.
82 AI MAGAZINE
Hagiu, A., and J. Wright. 2020. “When Data Creates Competi- Method for Customer Experience Survey Analysis.” Technologies
tive Advantage.” Harvard Business Review. https://hbr.org/2020/ 8(4): 76.
01/when-data-creates-competitive-advantage (last accessed: Jan– Martínez-Plumed, F., E. Gómez, and J. Hernández-Orallo. 2021.
Feb 2020). “Futures of Artificial Intelligence through Technology Readiness
Hamilton, R. H., and W. A. Sodeman. 2020. “The Questions We Levels.” Telematics and Informatics 58: 101525.
Ask: Opportunities and Challenges for Using Big Data Analyt- Maze, J. 2021. “McDonald’s Faces Lawsuit over Its Voice
ics to Strategically Manage Human Capital Resources.” Business Recognition Technology.” Restaurant Business. https://www.
Horizons 63(1): 85–95. restaurantbusinessonline.com/technology/mcdonalds-faces-
Han, M. C. 2021. “The Impact of Anthropomorphism on Con- lawsuit-over-its-voice-recognition-technology (last accessed: 7
sumers’ Purchase Decision in Chatbot Commerce.” Journal of June 2021).
Internet Commerce 20(1): 46–65. https://doi.org/10.1080/15332861. McDonald’s Corporation. 2018. “McDonald’s Unveils New Global
2020.1863022. Growth Plan.” Prnewswire.Com. https://www.prnewswire.com/
Hunter, R., and G. Westerman. 2009. Real Business of IT: How CIOs news-releases/mcdonalds-unveils-new-global-growth-plan-
Create and Communicate Value. Boston, MA. Harvard Business 300415634.html (last accessed: 1 March 2018).
Review Press. McDonald’s Corporation. n.d. “McDonald’s to Acquire Dynamic
Hupfer, S. 2020. “The Window for AI Competitive Advantage Is Yield, Will Use Decision Technology to Increase Personaliza-
Narrowing.” Deloitte Insights. https://www2.deloitte.com/us/ tion and Improve Customer Experience.” Mcdonalds.Com.
en/insights/industry/technology/ai-competitive-advantage- https://corporate.mcdonalds.com/corpmcd/en-us/our-
narrowing.html (last accessed: 11 September 2020). stories/article/ourstories.dynamic_yield_1164112100.html.
IBM Institute for Business Value. 2021. Turn Input into Impact McKinsey. 2020. Modernizing Technology in the Service of the
with an AI-Led Marketing Strategy, IBM. https://www.ibm. Customer. McKinsey & Company. https://www.mckinsey.com/
com/thought-leadership/institute-business-value/report/ai- business-functions/mckinsey-digital/our-insights/modernizing-
marketing-strategy (last accessed: 8 November 2020). technology-in-the-service-of-the-customer (last accessed: 21 May
Kandaswamy, R., and D. Furlonger. 2021. The Gartner Business Value 2020).
Model: A Framework for Measurin Business Performance. Gartner. Meakin, T., J. Palmer, V. Sartori, and J. Vickers. 2021. Winning with
New York, NY. https://www.gartner.com/en/doc/3713917-the- AI Is a State of Mind. McKinsey & Company. Retrived from:
gartner-business-value-model-a-framework-for-measuring- https://www.mckinsey.com/capabilities/quantumblack/our-
business-performance. insights/wining-with-ai-is-a-state-of-mind (last accessed: 30
Kellogg, K. C., and M. A. Valentine. 2022. “Five Mistakes Man- April 2021).
agers Make When Introducing AI—and How to Fix Them.” Mergent, Inc. 2022. Mergent Online. University of Portland Library.
The Wall Street Journal. Retrieved from: https://www.wsj.com/ Retrieved from: http://www.mergentonline.com.
articles/managers-mistakes-artificial-intelligence-11667596940 Mikalef, P., S. O. Fjørtoft, and H.Y. Torvatn. 2019. “Developing an
(last accessed: 5 November 2022). Artificial Intelligence Capability: A Theoretical Framework for
Kiron, D., and M. Schrage. 2019. “Strategy for and with AI.” MIT Business Value.” Lecture Notes in Business Information Processing.
Sloan Management Review 60(4): 30–35. 373 LNBIP, pp. 409–16.
Klein, D. 2018. “McDonald’s Spending $6 Billion on Nationwide Mikalef, P., and M. Gupta. 2021. “Artificial Intelligence Capabil-
Remodels.” QSR Magazine. https://www.qsrmagazine.com/fast- ity: Conceptualization, Measurement Calibration, and Empirical
food/mcdonalds-spending-6-billion-nationwide-remodels (last Study on Its Impact on Organizational Creativity and Firm
accessed: August 2018). Performance.” Information & Management 58(3): 103434.
Kohavi, R., D. Tang, and Y. Xu. 2020. Trustworthy Online Controlled Mishra, S., M. T. Ewing, and H. B. Cooper. 2022. “Artificial Intel-
Experiments: A Practical Guide to a/b Testing. Cambridge England: ligence Focus and Firm Performance.” Journal of the Academy
Cambridge University Press. Marketing Science 50: 1176–97.
Laney, D. 2020. “Infonomics: The New Economics of Information.” Nawaz, H., T. Ali, A. Al-laith, I. Ahmad, S. Tharanidharan, and S.
Big Data and Advanced Analytics (6-3): 432–8. K. A. Nazar. 2018. “Sentimental Analysis of Social Media to Find
Lee, I., and Y. J. Shin. 2020. “Machine Learning for Enterprises: Out Customer Opinion.” In International Conference on Intelligent
Applications, Algorithm Selection, and Challenges.” Business Technologies and Applications (pp. 110–115). Singapore: Springer .
Horizons 63(2): 157–70. Ng, A. 2018. AI for Everyone. Coursera. https://www.coursera.org/
Lucas, A. 2021. McDonald’s Revenue Tops Pre-pandemic Levels, Fueled learn/ai-for-everyone.
by the Strong U.S. Recovery. CNBC. https://www.cnbc.com/2021/ Parthasarathy, A. 2021. Improving Net Promoter Score Using
04/29/mcdonalds-mcd-q1-2021-earnings.html (last accessed: 29 Machine Learning. TM Forum Inform. https://inform.tmforum.
April 2021). org/insights/2021/04/improving-net-promoter-score-using-
Lui, A. K., M. Lee, and E. W. Ngai. 2022. “Impact of Artificial Intel- machine-learning/.
ligence Investment on firm Value.” Annals of Operations Research Paschen, U., C. Pitt, and J. Kietzmann. 2020. “Artificial Intelligence:
308(1): 373–88. Building Blocks and an Innovation Typology.” Business Horizons
Lysaght, T., H. Y. Lim, V. Xafis, and K. Y. Ngiam. 2019. “AI-Assisted 63(2): 147–55.
Decision-Making in Healthcare.” Asian Bioethics Review 11(3): Porter, M. 1996. “What Is Strategy?” Harvard Business Review. https://
299–314. hbr.org/1996/11/what-is-strategy (last accessed: Nov–Dec 1996).
Markoulidakis, I., I. Rallis, I. Georgoulas, G. Kopsiaftis, A. Doulamis, Proctor, P., and M. Smith. 2017. “The Gartner Business Value
and N. Doulamis. 2020. “A Machine Learning Based Classification Model: A Framework for Measuring Business Performance.”
AI MAGAZINE 83
Gartner.com. https://www.gartner.com/en/doc/3713917-the- Zolas, N., Z. Kroff, E. Brynjolfsson, K. McElheran, D. N. Beede, C.

gartner-business-value-model-a-framework-for-measuring- Buffington, N. Goldschlag, L. S. Foster, and E. Dinlersoz. 2021.
business-performance. Advanced Technologies Adoption and Use by US Firms: Evidence
Pumplun, L., C. Tauchert, and M. Heidt. 2019. “A New Organiza- from the Annual Business Survey (No. w28290). National Bureau of
tional Chassis for Artificial Intelligence-Exploring Organizational Economic Research. http://doi.org/10.3386/w28290
Readiness Factors.” In 27th European Conference on Information
Systems – ECIS 2019. Stockholm and Uppsala, Sweden.
PWC. 2019. “Sizing the Prize.” PWC.com.Retrieved from: AU T H O R B I O G R A P H I E S
https://www.pwc.com/gx/en/issues/analytics/assets/pwc-ai-
analysis-sizing-the-prize-report.pdf. Naveen Gudigantala is an Associate Professor of
Rai, A., P. Constantinides, and S. Sarker. 2019. “Next Generation Dig- Information Systems and Silicon Valley Distinguished
ital Platforms: Toward Human-AI Hybrids.” MIS Quarterly 43(1):
Professor at the Dr. Robert B. Pamplin, Jr. School
iii–ix.
of Business at University of Portland. His research
Ramesh, R. 2017. How to Hire a Data Scientist? https://www.
drrajramesh.com/. interests include the business value of artificial intel-
Rosoff, M. 2019. Google CEO’s Sunny New Message: We Want to be ligence (AI) systems, data analytics, healthcare analyt-
Helpful. CNBC. https://www.cnbc.com/2019/06/20/sundar- ics, decision-support systems, ethical issues in supply
pichai-alphabet-founders-letter-2018-focuses-on-being- chains, and information systems education. His jour-
helpful.html (last accessed: 19 June 2019). nal articles have appeared in internationally reputed
Ruggless, R. 2021. McDonald’s Tests Automated Drive-Thru journals in information systems (IS) and in reference
Order-Taking in Chicago Area. Nation’s Restaurant News.
disciplines of marketing and management (e.g., Com-
https://www.nrn.com/quick-service/mcdonald-s-tests-
automated-drive-thru-order-taking-chicago-area (last accessed:
munications of AIS, Decision Support Systems, Interna-
3 June 2021). tional Journal of Information Management, Information
Sambasivan, N., S. Kapania, H. Highfill, D. Akrong, P. Paritosh, and L. Systems Management, Journal of Information Systems
M. Aroyo. 2021. “Everyone Wants to Do the Model Work, Not the Education, Management Research Review, etc.).
Data Work”: Data Cascades in High-Stakes AI. In Proceedings of
the 2021 CHI Conference on Human Factors in Computing Systems
Sreedhar Madhavaram is the Alumni Professor of
(pp. 1–15). Yokohama, Japan.
Marketing and the United Supermarkets, Inc. Professor
Sebastian, D. 2021. “McDonald’s Loyalty Program Goes
National.” The Wall Street Journal. https://www.wsj.com/ of Marketing, and the Area Coordinator for Marketing
articles/mcdonalds-introduces-customer-loyalty-program- and Supply Chain Management at the Rawls College of
11624376506#:%7E:text=Users%20will%20get%201%2C500% Business at Texas Tech University. He earned his PhD
20points,cheese%20biscuit%2C%20the%20company%20said (last in Marketing at the Rawls College of Business at Texas
accessed: 22 June 2021). Tech University. His research interests include theory
Shmueli, G., P. C. Bruce, and N. R. Patel. 2016. Data Mining for development, marketing strategy, capability building,
Business Analytics: Concepts, Techniques, and Applications with
firm competitiveness, information systems/technology,
XLMiner. Hoboken, NJ. John Wiley & Sons.
Shrestha, Y. R., S. M. Ben-Menahem, and G. Von Krogh. 2019. “Orga-
and strategic issues at the intersection of marketing and
nizational Decision-Making Structures in the Age of Artificial information technology. His research has appeared in
Intelligence.” California Management Review 61(4): 66–83. several journals, including The Data Base for Advances
Simon, H. A. 1957. Models of Man. New York: Wiley. in Information Systems, the Journal of the Academy of
Wiggers, K. 2021. Survey Finds Talent Gap Is Slowing Enterprise Marketing Science, the Journal of Business Ethics, the
AI Adoption. VentureBeat. https://venturebeat.com/2021/04/19/ Journal of Business Logistics, Industrial Marketing Man-
survey-finds-talent-gap-is-slowing-enterprise-ai-adoption/ (last
agement, Information & Management, the Journal of
accessed: 19 April 2021).
Business Research, and AMS Review.
Wilson, H. J., and P. R. Daugherty. 2020. “The Next Big Break-
through in AI Will Be Around Language.” Harvard Busi-
ness Review. https://hbr.org/2020/09/the-next-big-breakthrough- Pelin Bicen (PhD, Texas Tech University) is an Asso-
in-ai-will-be-around-language (last accessed: 23 September 2020). ciate Professor of Marketing at the Suffolk University’s
Winter, S. 2000. “The Satisficing Principle in Capability Learning.” Sawyer Business School. Previously, she worked in
Strategic Management Journal: 981–96. marketing strategy and marketing analytics for BBDO
Yaffe-Bellany, D. 2019. “Would You Like Fries with That? McDon-
Advertising Agency, managing both national and inter-
ald’s Already Knows the Answer.” The New York Times.
https://www.nytimes.com/2019/10/22/business/mcdonalds-tech-
national brands. Her research focuses on intra- and
artificial-intelligence-machine-learning-fast-food.html (last interfirm marketing and innovation strategy, resource
accessed: 22 October 2019). constrained innovation, global supply chain strategies,
Zenrobotics. 2022. Lundstams. making better use of material. Terex. and interorganizational governance. She has authored
com. Retrived from https://zenrobotics.com/clients/lundstams. several articles in marketing and innovation journals
84 AI MAGAZINE
some of which are Journal of Business Research, Creativ-

ity and Innovation Management, Journal of Marketing How to cite this article: Gudigantala, Naveen,
Management, Journal of Marketing Theory and Practice, Sreedhar Madhavaram, and Pelin Bicen. 2023. “An
Journal of Business and Industrial Marketing, Innova- AI decision-making framework for business value
tion: Management, Policy, and Organization, Manage- maximization.” AI Magazine 44: 67–84.
ment Research Review, Journal of Marketing Education, https://doi.org/10.1002/aaai.12076
Journal of Global Scholars of Marketing Science, and
Journal of Research in Marketing and Entrepreneur-
ship among others. She has presented her research
at national and international marketing strategy and
innovation conferences.

AI Magazine - 2023 - Gudigantala - An AI Decision Making Framework For Business Value Maximization

Uploaded by

Copyright:

Available Formats

AI Magazine - 2023 - Gudigantala - An AI Decision Making Framework For Business Value Maximization

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

AI Magazine - 2023 - Gudigantala - An AI Decision Making Framework For Business Value Maximization

Uploaded by

Copyright:

Available Formats

Received: 23 June 2022 Revised: 20 November 2022 Accepted: 2 January 2023

An AI decision-making framework for business value

Naveen Gudigantala1 Sreedhar Madhavaram2 Pelin Bicen3

1 RobertB. Pamplin School of Business

INTRODUCTION their AI investments (Bean 2021). Indeed, most non-tech

AI Magazine. 2023;44:67–84. wileyonlinelibrary.com/journal/aaai 67

Dashboards, used for display of visualization, are not only

TA B L E 1 Decisions to make in the AI framework.

TA B L E 2 Mapping AI decision-making framework to McDonald’s case.

FIGURE 4 A flowchart of artificial intelligence (AI) decision-making framework for managers.

ORCID Campbell, C., S. Sands, C. Ferraro, H. Y. J. Tsao, and A. Mavrommatis.

Gartner.com. https://www.gartner.com/en/doc/3713917-the- Zolas, N., Z. Kroff, E. Brynjolfsson, K. McElheran, D. N. Beede, C.

some of which are Journal of Business Research, Creativ-

You might also like