OPPORTUNITIES AND CHALLENGES IN ADOPTING AI CAPABILITIES: AN EMPIRICAL STUDY

ABSTRACT

In today's competitive business landscape, companies seek new opportunities and operational optimizations through innovative technologies. This article analyzes a real use case from a Fortune 500 company that leveraged Artificial Intelligence (AI) to enhance marketing strategies and partner engagements. The objective of this article is to explore how advanced capabilities can be applied in real business environments, diuscss business readiness for AI, and evaluate AI's adoption tailwinds and opportunities. Additionally, it examines organizational readiness aspects for emerging capabilities and the implications of change management in AI adoption. This paper is essential for understanding the practical applications of AI in business and its challenges of integration into daily operations. The findings are structured as insights, serving as a valuable resource for the practicioners exploring the way to leverage value driven operations.

АННОТАЦИЯ

В условиях современной конкурентной бизнес-среды компании стремятся находить новые возможности и оптимизировать свои операции с помощью инновационных технологий. В этой статье рассматривается реальный пример из компании, входящей в список Fortune 500, которая использовала искусственный интеллект (AI) для улучшения маркетинговых стратегий и взаимодействия с партнерами. Цель статьи – исследовать, как продвинутые технологии могут применяться в реальных бизнес-сценариях, обсудить готовность бизнеса к внедрению AI, а также оценить факторы, способствующие его принятию и выявить возможности для улучшения. Также рассматриваются аспекты организационной готовности к новым технологиям и влияние управления изменениями на процесс внедрения AI. Эта статья необходима для понимания практического применения AI в бизнесе и трудностей его интеграции в повседневные операции. Результаты представлены в виде выводов, которые могут стать ценным ресурсом для практиков, стремящихся использовать подходы, ориентированные на повышение эффективности бизнеса.

Keywords: Artificial Intelligence, insights, marketing transformation, value driven operations.

Ключевые слова: Искусственный интеллект, аналитические данные, трансформация маркетинга, эффективная операционная деятельность.

INTRODUCTION

According to MkKinsey [1] the interest to an AI across industries is undeniable today, and companies are experimenting with diferrent techniques and approaches like machine learning (ML) and statistical data science as subsets of AI. Some companies are industry leaders in landing AI based capabilities, they set standards and establish best practices, while others are still exploring the potential of AI tools and the organizational requirements for adopting emebring capabilities. As highlighted in Russell and Norvig's book "Artificial Intelligence - A Modern Approach" [2], AI and ML have become crucial technologies driving innovation across industries. AI broadly refers to machines simulating human intelligence to perform tasks like image and speech recognition and decision-making. ML, as a subset of AI, allows machines to learn from data and improve over time without explicit programming. For instance, AI powers virtual assistants like Amazon's Alexa, which understands and responds to natural language commands, while ML algorithms within Alexa learn from user interactions to enhance response accuracy. This distinction shows AI's broad scope in mimicking human cognitive functions, while ML focuses specifically on self-improving algorithms.

Despite the broad scope of data science, which requires scholars and IT professionals to understand its fundamentals, the practical applicability of AI-based use cases can be easily explained to anyone, enabling companies to benefit from them even without having in-house data science expertise. For example, a data correlation matrix or heatmap, which visually represents relationships between multiple variables, can be easily explained and assessed in the context of business scenarios. For example - in retail, a correlation heatmap can show how factors like pricing, marketing spend, and seasonal trends impact sales (see Figure 1). The correlations between these factors, such as how changes in price affect sales are straightforward to assess. Identifying strong correlations can, in turn, allow businesses to optimize their strategies, like increasing marketing budgets during high-impact seasons to maximize revenue. When dealing with numerous factors i.e. 20, 50, or even over 100, manual data processing to find correlations is impractical. ML can efficiently analyzes large datasets, uncovers hidden patterns, and provides actionable insights, significantly enhancing operational productivity and decision-making and correlation matrix is one of the examples.

Figure 1. Data correlation heatmap built around 4 factors

Another simple example of AI in business is regression analysis, which for example helps predict outcomes like sales based on factors such as marketing spend, customer satisfaction, and planogram compliance (see Figure 2). Companies analyze historical data to forecast future sales and optimize resources. For instance, if every additional $1,000 spent on marketing increases sales by $5,000, businesses can plan budgets to maximize revenue. Regression analysis can include multiple factors, such as marketing spend, trade reps coverage and even outside temperature by seasons. The more factors analyzed, the more detailed the prediction. Such data analysis, which is difficult to perform manually, can significantly enhance decision-making and strategic planning.

Figure 2. Regression analysis of two factors – sales as a function of marketing spend

According to a McKinsey survey [4] of more than 12,000 retail customers and around 60 executives across Europe, 10 opportunities in five key categories account for over 80 percent of the potential EBIT impact (see Figure 3). Notable use cases that are considered industry standards include price recommendations, assortment recommendations, and promotions optimization. These use cases fully rely on advanced analytics and AI/ML techniques.

Figure 3. Key use cases per category

Same McKinsey research indicates that order forecasting and price recommendation analytics have become industry standards. Companies are now starting to implement advanced capabilities like personalized promotions and marketing mix optimization (MROI) into their business operations.

CASE STUDY AND IMPLEMENTATION

Now, let's look at a real implementation of AI capabilities from my experience at a mature, globally present company. We will analyze its results, gains, challenges, lessons learned, and overall insights.

Business Context

This article focuses on real implementation case in a tobacco and nicotine product giant with billions in revenue and a global workforce. They offer a range of products from traditional cigarettes to modern nicotine alternatives.

It operates in heavily regulated environments and face unique challenges when it comes to technology deployment at scale. For such companies, implementing an emerging capabilities is a strategic move crucial for competitiveness and market adaptation and requires mobilization at all levels-from end markets to global teams.

Business Problem Statement

The retail business generates vast data from purchases, customer and consumer behaviors, trade promotions, and partnership products. However, there is a lack of combined analysis of marketing spend efficiency and buying patterns, no smart segmentation of retailer behavior, and underutilization of historical data for actionable insights

Business Goals & Objectives

Leverage historical data and ML techniques to enhance revenue and optimise marketing spend through targeted trade promotions and customer engagements.

Users

Trade Representatives, Sales & Delivery Representatives, Territory Managers (back office), Retailers. Capabilities Adoption Principles

Developed with extensive business involvement, followed by users driven testing and hyper-care phases, the project included comprehensive analytics dashboards on adoption, transition into support documents, and service lines to ensure effective adoption, exploitation, and data-driven decision-making. Resources needed for adoption (dedicated support, knowledge center, community of practice) also considered as part of the business case.

Use Case 1: Retailer Segmentation

Retailer segmentation is a common business challenge, but it is fundamental for understanding of customers behaviours, preferences and overal customers sentiment. In this use case, the k-means clustering technique of ML with five clusters was used, providing a balanced solution for grouping and understanding customer behavior without complicating data analysis. The simplified process for customer clustering based on RFM (recency, frequency, and monetary) analysis you may see as Figure 5 below. The RFM model was first proposed by Hughes of the American Database Institute in 1994 [5]. As a popular tool for customer value analysis, it has been widely used for measuring customer lifetime value [6] and in customer segmentation and behavior analysis [7]. In the following paragraphs, a brief description of the RFM model from the above literature is provided.

Figure 5. simplified process for customer segmentation

To segment customers based on their purchasing behavior I used the RFM model, which evaluates customers based on: Recency (R) - the number of days since the customer's last purchase, Frequency (F) - the total number of purchases made by the customer and Monetary (M) - the total amount spent by the customer. The RFM scores were normalized, ranging from 0 to 1, to ensure that each metric had the same weight. Next, each customer was assigned to the nearest centroid based on the Euclidean distance.

Where , and  are the RFM scores of the centroid c.

To achieve clustering, I recalculated the centroids as the mean of all customers assigned to each cluster and repeated the assignment and update steps until the centroids no longer changed significantly. Figure 5 below shows a simplified representation of k-means clustering for a group of customers (outlets) based on their RFM scores.

Figure 5. Outlets clustering using k-means

According to the RFM rules and segment definitions defined by the business, all outlet segments can be described as follows:

Cluster 1 (C1) – High Performers: Retailers who buy a lot, regularly.

Cluster 2 (C2) – New: Retailers who bought a lot recently but with a low frequency score, indicating they may be new outlets that require attention.

Cluster 3 (C3) – Regular: The majority of outlets with similar average performance.

Cluster 4 (C4) – Disengaged: Retailers who buy regularly but with significant delays between purchases, requiring additional engagement.

Cluster 5 (C5) – Fix the Basics: Outlets that do not demonstrate significant performance and need to be investigated.

The clustering analysis provided significant business insights by segmenting customers based on volume and purchase regularity. This automated analysis, which impacted thousands of customers, revealed valuable patterns that were previously impossible to mine manually. However, it also posed challenges for the business on how to effectively utilize the insights. We will revisit these challenges later in this article.

Use Case 2: Marketing Spend Efficiency Analysis

Another fundamental question in business is how efficient the money spent to promote products or services is. According to a Gartner survey conducted across CMOs [3], “Marketing budgets are climbing back.” Survey results show that budgets have recovered somewhat, with the average marketing spend increasing from 6.4% to 9.5% of company revenue across almost all industries. Thus, the question of effective spending has never been more relevant: How effective is my marketing spend, and how does it affect sales?

To reveal which promotional campaigns are performing well and which are not yielding the desired results, sales uplift was calculated by comparing historical sales without the promotion to sales with the promotion applied. For marketing spend analysis, I used a two-tailed test, which allows detecting effects in both directions—positive (increased sales) and negative (decreased sales)—based on a vast amount of historical data.

According to Butt, Moeen, and Baig [8], two-tailed tests are commonly used in statistical analysis in finance and marketing to help determine significant differences in data without direction bias. In finance, they evaluate investment returns, assess risk models, and analyze market reactions. In marketing, they measure campaign effectiveness and compare consumer preferences, ensuring unbiased, reliable decision-making.

I aimed to identify insights about both successful and unsuccessful promotions using almost two years of historical data. To achieve this, two-sample tests were used to determine any significant difference between the sample means in either direction (higher or lower).

 mean of sample 1, ,  

This formula will give us a t-value that quantifies the difference between the means of the two samples.

As next, two hypotheses were formulated:

Null Hypothesis (H0): The promotion has no effect on sales.

Alternative Hypothesis (H1): The promotion influences sales, either positively or negatively.

With the significance level set at 5%, it is split between the two tails of the normal distribution, with 2.5% (0.025) in each tail, as shown in Figure 6 below:

Figure 6. critical regions for hypothesis testing

By using the two-tailed test and setting the rejection regions at ±1.96, we account for the possibility that the trade promotion could either significantly increase or decrease sales. This approach ensures that we can detect any significant deviations from the expected historical sales performance due to the promotions.

RESULTS ANALYSIS AND DISCUSSION

Research on the adoption of AI capabilities shows that this work aligns with other studies. According to a study by the Journal of Economics and Management Strategy [9], AI adoption is concentrated in key segments of the economy but is also somewhat diffused. Very large firms and high-growth startups are leading the way in adopting emerging capabilities. While the average firm adoption in 2017 was only 5.8%, this figure rises to 18.2% when considering employment weighting. This gap suggests that understanding the impacts at the worker level will require more attention to firm dynamics, beyond just a skill or task-based view.

In my research, despite business discovery, technical design, implementation, and deployment were straightforward, real challenges were posed by organizational requirements for adoption. These included the major question - how to integrate AI capabilities into daily operations organically and what organizational changes (roles, responsibilities, organizational design) are needed. This goes beyond just technical knowledge and project scope.

Let's examine the key insights from the two use cases I worked on: segmentation and marketing spend efficiency conducted for a major business and what insights were discovered beyond just technical.

Business Involvement

Business users are normally enthusiastic about new opportunities. Even without strong senior management involvement, they actively participate in business discovery sessions, contributing to solution logic and UX despite having other daily responsibilities. Business involvement in solution design was agreed upon and committed to as part of the value case conversation. Business users proactive involvement may create a challenge on the later stages if senior stakeholders are not attached to this users involvement much, I will explore this challenge in more details later in this article.

Delivery Agility

Maintaining an MVP (Minimum Viable Product) scope for emerging AI/ML capabilities is crucial. This approach allows quick adaptation of innovative ideas and addresses challenges. For the mentioned use cases, a small agile delivery team of four developers and one scrum master was mobilized. To ensure implementation agility, a cadence of meetings was established, covering both technical and business aspects. Implementation took four months from initial engagement to product go-live at a national scale, involving business users with no prior knowledge of data science or AI/ML. Such an agility can create a bottle neck if a business is not able to cope with the technical delivery pace (due to other agendas, shortage of resources etc).

Customer Segmentation

The examined dataset had over 100,000 retail outlets of varying sizes. Traditional methods for outlet supply and promotion involved manual, biased selection of only limited, predefined retailers. Clustering complete dataset revealed overlooked patterns, such as high-potential outlets not regularly covered. Figure 7 shows the segmentation of the actual universe of outlets, forming a typical bell curve with a skew towards the disengaged cluster.

Figure 7. segmentation bell curve

Results of this analysis showed that there’s immense potential for further analysis and outlet supply balancing by introducing additional factors such as outlet size, type, and reach. One of the benefits segmentations unlocked was understanding how to organize trade coverage for sales reps based on potential value. This approach is fundamental to the concept I call “Value-based coverage,” which is a common challenge across industries reliant on customer engagements in the field. However, such an insight created a business dilemma of how to action all these insights as part of daily operations, this will be explored further.

Marketing Spend Efficiency

Figure 8 below shows the analysis of three different promotions over 12 months of historical data. Key patterns include sales plateaus and varying effects of the same promo. Here are the detailed insights:

Promo 1 - “Pay for Performance”: Sales quickly plateaued despite investment due to store capacity limits. This finding is crucial for setting marketing budgets per segment, balancing the reduction of loyal consumer attrition against deal-seeker attraction.

Promo 2 - “In-store Displays”: High-visibility promotions, like in-store displays, had mixed results. Investments showed a clear threshold—only reaching a certain investment level affected sales due to the competitive landscape in outlets. Investments below the threshold did not negatively affect sales, as competitors quickly filled the promotional capacity gaps in-store.

Promo 3 - “Bundling”: Product bundling resulted in an 8% to 11% sales uplift, highlighting that the effectiveness of bundles depends on perceived relevance.

Overall, just exploitation of these two use cases helped the business understand that there are plenty of potentilally actionable situations, which require rethinking of daily operations just due to scale of insights. In trade promotions area while they can significantly influence sales, their success varies. In-depth data analysis revealed key insights for marketers in trade promotions, such as sales plateaus, investment thresholds for in-store materials, and the importance of bundling relevance. Incorporating these findings into daily operations from one hand can significantly optimize customer engagement strategies and marketing investment allocation and overall ROI and from other hand has no direct answer of how to seamlessly adjust business operations to benefit from it.

Figure 8.  Marketing spend impact on sales

AI capabilities utilisation

As I mentioned in this article above, piloting AI capabilities in real business is less about technical excellence and more about finding the right momentum in organisation to enhance its productivity and how to support end users with emerging capabilities adoption. When researching broader technology adoption topic, I came to conclusion that in general adoption can be built around two possible approaches: to push (bring tools, convince users, proactively support) or to pull (explain benefits, gain internal support). For this use case, I used a combination of both, with a dominant push component enabling the business with decision-making.

From my experience with complex transformations, I observed that solution exploitation doesn’t equal its adoption. Solutions might be used sporadically out of curiosity rather than being an integral part of operations. Based on my exerience with other technology implementations, measured utilization (Figure 9) can show initial high user interest peaking post-implementation, followed by stabilization. Time spent in the solution gradually declined due to fewer logins and improved user competence, needing less time to get results. However, the drop in logins indicates barriers preventing frequent use. This ties into the challenges section, highlighting gains, headwinds, and overall insights and learnings.

According to an empirical article by Victoria Uren and John S. Edwards [10], at some point, the adoption of emerging capabilities can experience the "dead valley" phenomenon - struggling to adopt AI if the solution is not ready to be used without organizational changes. More complex technical capabilities require building bridges between developers and business stakeholders to foster mutual understanding and enhance data, technology, and organizational skills. For continuous high adoption, the support of top management is essential, not only for individual projects but as continuous behaviour-led culture, to develop social capital and promote knowledge sharing across the organization.

Figure 9. AI capabilities adoption curve impacted by processes, technology and people related challenges

CHALLENGES

A research article from the Department of Psychology at the Norwegian University of Science and Technology [11] highlights that organizational technology adoption and readiness for change are common challenges across industries. Companies seek ways to help employees use these solutions effectively by conducting training sessions and creating social capital around new technologies. AI adoption challenges are nothing different and challenges are like any other change management related and require a proper adoption framework. For this empirical study, I categorized the challenges into three areas: people, processes, and technology.

People

Implementation of emerging capabilities in businesses and effective injections of them into business operations require mature leadership, focus on strategy, clear benefits and structured assumptions and risks view. IT is expected to “drive this bus” by showing the way to the business and leading them through the change. This however require equal partnership and strong commitment from all sides. In the mentioned use cases a push approach was used where IT led a role of a bus driver setting the stage and agenda for implementation, which at certain point created a pace-related challenge for the business and led to a lack of business ownership at the end. Technical teams' agility created challenges for business users to keep pace, leading to declining participation. What was obvious for technical teams required knowledge and change management for the business users. Despite IT was driving the project, collaboration with marketing was pivotal to define the right direction and lack of participation posed a delivery challenges.

Another challenge was the lack of senior stakeholder support, as the project was initiated from the bottom, reaching cross functional integration limits without senior management's awareness and support.

Additionally, there was a gap in understanding between technical teams and marketing driven by lack of business saviness from the technical teams and vise versa - marketing lacked technical details, highlighting the need for leadership that can bridge this gap.

And finally, the delivered products and insights themselves formulated a non-trivial question for the business users – how I can handle all these insights and not impact my other duties, which by nature of the challenge in between of people and processes (organisational) – related.

Processes

Establishing processes is crucial for any project success. Without defined processes, tools can become toys with initial high interest but quick disengagement from users. Following established standards and procedures helps employees maintain productivity on a high level post implementation. Processes help define roles, knowledge, and tools needed for effective technology adoption. Absence of this pillar complicates solution adoption and further scaling potential. There are academic articles by other scholars that explain how shared services can help businesses change and grow [12], organizations with Global Business Services or Global Shared Services can support business users in adopting new solutions by providing resources, maintaining knowledge, and helping scale solutions across the organization. Technical teams should not play adoption role, instead adoption counterpart and its role should be defined and agreed upon.

From organisational readiness perspective, though not all the answers are available at the beginning of the project, this aspect should be discussed before departure, by aligning on the relevant risks and discussing potential mitigation strategies that could address processes (organisational) related challenges. And such strategies must be aligned with and supported by senior sponsors of the AI project.

Technology

Selecting the fit-for-purpose technology, understanding by the business its implication for the daily operations is a definite prerequisit prior to start. Technology selection is interconnected with people and processes and is integral to the adoption framework. Victoria Uren and John S. Edwards, in their empirical adoption study, describe these as “operational elements of socio-technical systems” and the “golden triangle” of people, processes, and technology [9]. Managers need to focus on the interactions between these elements, as the links between them are as important as the elements themselves (see Figure 10).

Figure 10. Golden triangl of socio-technical systems

Technical teams should carefully guide business users by familiarizing them with the pros and cons of different data science techniques and approaches, translating complex concepts into business-friendly language. Achieving this will not only empower business users to contribute more effectively to projects but also foster trust, increase commitment, and potentially create data science ambassadors within the business user community.

CONCLUSION

The adoption of AI capabilities in business holds transformative potential, offering significant opportunities in areas like customer segmentation and marketing spend efficiency analysis. Although such use cases involve data science techniques such as k-means clustering, regression analysis, and statistical analysis, they can still be explained and translated into business-friendly language. The two use cases presented were selected as the most common across industries, based on research of other scholars in the field [13] and surveys conducted by large consultancy firms. These examples demonstrate the tangible benefits AI can bring, including more precise targeting, personalized customer experiences, and optimized marketing budgets. However, realizing these benefits is not without challenges.

In the area of customer segmentation, AI enables businesses to move beyond traditional, often biased, geographic-based approaches by leveraging vast amounts of data to identify more nuanced customer segments. This allows for highly tailored marketing strategies that enhance customer satisfaction and loyalty. Similarly, in marketing spend efficiency analysis, AI-driven insights help allocate marketing resources more effectively, ensuring that investments yield the highest possible returns. When combined with segmentation analysis, these insights can create efficiencies that surpass traditional methods.

Despite these promising opportunities, businesses face challenges in the areas of people, processes, and technology that can slow down adoption. Successful AI integration requires skilled and adaptable personnel, the redesign or adjustment of business processes to seamlessly incorporate AI-driven insights into daily operations, and overcoming technical issues such as data quality and system integration. Addressing these challenges through training, process alignment, robust data governance, and strong leadership support is essential to unlocking AI's full potential.

In conclusion, while the journey to integrating AI into business operations presents challenges, the potential rewards make it a worthwhile endeavor. Overcoming these hurdles can unlock new opportunities, enhancing efficiency and effectiveness while providing a competitive edge in an increasingly data-driven marketplace. This article aims to guide IT practitioners and businesses through the complexities of AI implementation and adoption, offering insights to help harness its transformative power.

References:

1. The state of AI in 2023. McKinsey, 2023, page 5 https://www.mckinsey.com/~/media/mckinsey/business%20functions/quantumblack/our%20insights/the%20state%20of%20ai%20in%202023%20generative%20ais%20breakout%20year/the-state-of-ai-in-2023-generative-ais-breakout-year_vf.pdf#:~:text=URL%3A%20https%3A%2F%2Fwww.mckinsey.com%2F~%2Fmedia%2Fmckinsey%2Fbusiness%2520functions%2Fquantumblack%2Four%2520insights%2Fthe%2520state%2520of%2520ai%2520in%25202023%2520generative%2520ais%2520breakout%2520year%2Fthe
2. Russell and Norvig, Artificial Intelligence - A Modern Approach, 2021, pages 36-37
3. “What Marketing Budgets Look Like in 2022”, Gartner, 2022 https://www.gartner.com/en/articles/what-marketing-budgets-look-like-in-2022
4. “The State of Grocery Retail”, McKinsey, 2024 https://www.eurocommerce.eu/the-state-of-grocery-retail/
5. Hughes A. M., Strategic Database Marketing, 1994, Probus Publishing Company, Chicago, USA. Google Scholar
6. Cheng C.-H. and Chen Y.-S., Classifying the segmentation of customer value via RFM model and RS theory, Expert Systems with Applications. (2009) https://doi.org/10.1016/j.eswa.2008.04.003
7. Chen D., Sain S. L., and Guo K., Data mining for the online retail industry: a case study of RFM model-based customer segmentation using data mining, Journal of Database Marketing & Customer Strategy Management. (2012) https://doi.org/10.1057/dbm.2012.17
8. "Journal of Business Research - Assessing the Firm-Level Financial Consequences of Clustering", Butt Moeen, Baig Ahmed, 2024 https://ideas.repec.org/a/eee/jbrese/v178y2024ics0148296324001577.html
9. McElheran, Kristina, et al. "AI adoption in America: Who, what, and where." Journal of Economics & Management Strategy, 2024 https://onlinelibrary.wiley.com/doi/full/10.1111/jems.12576
10. Victoria Uren, John S. Edwards, Technology readiness and the organizational journey towards AI adoption: An empirical study, International Journal of Information Management, 2023 https://doi.org/10.1016/j.ijinfomgt.2022.102588.
11. Saghafian, Mina, Karin Laumann, and Martin Rasmussen Skogstad. "Stagewise overview of issues influencing organizational technology adoption and use." Frontiers in Psychology 12 (2021): 630145. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2021.630145/full
12. Chen, Xin, Qizhi Dai, and Chaohong Na. "How finance shared services affect profitability: an IT business value perspective." Information Technology and Management (2023): 1-16. https://link.springer.com/article/10.1007/s10799-023-00391-1
13. Alves Gomes, Miguel, and Tobias Meisen. "A review on customer segmentation methods for personalized customer targeting in e-commerce use cases." Information Systems and e-Business Management 21.3 (2023): 527-570. https://link.springer.com/article/10.1007/s10257-023-00640-4