What Are Systems for Reproducing Customer Service Knowledge Using Generative AI? Directions in Challenges and Solutions as Seen from the ABeam Survey on the State of Personnel Development on the Ground in Retail | Insights

At the same time as there has been a re-evaluation of the value of physical stores since the pandemic, labor shortages have become endemic in the retail sector, rendering the tasks of securing and developing the talent to sustain store operations a pressing task for companies. Within this space, the skills and service capabilities of in- store staff charged with customer service responsibilities directly impact the trust that customers have in the sales floor and the overall customer experience. Despite this, such skills have, to date, been individualized and passed on through OJT as implicit knowledge.
Wanting to more concretely understand how these challenges manifest on the ground, ABeam Consulting conducted the “Survey on the State of Personnel Development on the Ground in Retail” (the “ABeam Retail Survey”) in June 2025, targeting 1,200 staff serving in stores across different retail sectors (general supermarkets, specialty shops, drug stores, etc.). The ABeam Retail Survey visualized the challenges facing the sector in both qualitative and quantitative terms, centering on instances where staff had issues in serving customers, experiences of addressing inquiries regarding product knowledge, and the state of knowledge sharing on sales floors.

Based on the results of this survey, this Insight organizes the structural issues in personnel development on the ground in retail and goes in search of effective approaches to preparing and making use of individualized knowledge as a reproducible asset. In particular, we put the focus on transferring and leveraging customer service knowledge using generative AI (“customer service knowledge AI”). Personnel development means more than just “teaching.” It means “creating a state wherein specific outcomes can be reproduced no matter who is executing the work.”
What this Insight demonstrates is practical hints for incorporating the potential of generative AI technology into transformation of ground-up personnel development. It also presents one potential practical solution to the HR challenges faced by companies on the ground in retail.

About the Author

Masaharu Nagahara

Director

1. What Is Happening on the Ground in In-Store Personnel Development?

This section reveals the deepening labor shortages in the retail sector and the limits of traditional OJT staff development. In particular, it brings into relief the need for “reproducible development processes” amid the context of the increasing difficulty of developing ready-to-go personnel to work on high-specialization sales floors.

The Return to Physical Stores Has Heightened Interest in In-Store Staff Development

As the trend back to physical stores continues in the wake of the pandemic, the retail sector is once again experiencing growing labor shortages. This has made development of the personnel who underpin the operation of physical stores a pressing business challenge.
Of these, one of the challenges confronting companies is the issue that, under development schemes centered on OJT by senior employees on the sales floor, there is constantly “no time to teach” and “not enough people to teach” (see Figure 1). This results in a situation wherein the staff responsible for product categories requiring high levels of expertise are out on the sales floor while still lacking relevant knowledge and untrained in how to talk about their products. Scenarios where specialist knowledge is needed are particularly common in customer service in specialty stores, department stores and home centers (see Figure 2). In such circumstances, staff may be hesitant to proactively engage with customers, potentially lowering customer satisfaction and contributing to loss of sales opportunities.

Changes in Stores Rendering Personnel Development Difficult

We hear even from many clients that we support that they are unable to fill shifts due to labor shortages, and so have to expand the range on the sales floor filled per staff member. This is likely another factor adding to the difficulty of catching up to the necessary level of product knowledge. The most common situations where staff run into issues in customer service are handling complaints, followed by situations related to product knowledge (see Figure 3). While complaints are negative situations, depending on how they are handled, they can also become opportunities to win people over as fans of a brand. Product knowledge situation can also contribute to cross-selling and upselling, depending on how staff perform customer service. Junior staff may be wasting such opportunities. To overcome such conditions, companies need to adopt an approach of implementing “reproducible processes” as organizations, and not letting personnel development rely on the techniques and experiences of individuals.

2. Why Does Personnel Development Continue To Be So Individualized?

This section presents the structural background to why in-store staff development has tended to be so individualized. We will probe the circumstances that have led to a “watch and learn” culture and the systems of evaluation and promotion that underpin it preventing the consolidation of standardized development models.

Watch and Learn Culture

In-store personnel development depends in large part on OJT that passes on informal knowledge through “watch and learn,” “listen and learn” and “make mistakes and learn” practices. The ABeam Retail Survey has also shown that “watch and learn from how managers or seniors handle customers” and “search for yourself” development environments are predominant compared to other models (see Figure 4). At the end of the day, instruction of in-store staff in relation to customer service will ultimately require a process of building up experience through handling actual customers. But this unintentionally furthers individualization of operations and gets in the way of realizing development systems that produce consistent results, no matter who did the initial instruction. Furthermore, because structures, in which the personnel who “work their way up” by gaining experience on the ground are the ones who tend to get promoted, have become deeply rooted, personnel development ends up following a style that depends on implicit knowledge in the form of “watch and learn,” rather than being a systematic framework. Through our experience of providing support for in-store operations, we have frequently encountered situations where such traditional structures remain strong, making it more challenging to promote transformation.

How to Close the Gap Between the Desire to Learn on the Ground and Those in Charge of Development

The limits to such individualized approaches to personnel development are thus clear. Nevertheless, while there remains strong demand to learn skills and knowledge on the ground (see Figure 5), the proportion of survey responses stating that development opportunities were “largely” or “totally absent” climbed as high as some 40% (see Figure 6). Amid this context, we also believe there are a certain number of voices who want to expand personnel development opportunities on offer. Attention has fallen on the use of generative AI as a means of promoting efficient and reproducible personnel development amid limited staffing and time resources.

Behind the belief that generative AI could be effective here are three problems that have failed to be addressed under traditional development schemes.
The first of these is leveling the cost of development. In conventional OJT, significant variation in the quality of development arises depending on the experience and ability of the trainer. It is also a fact that companies struggle to secure enough trainer time for the process. Using generative AI allows companies to share knowledge of a set level of quality and use customer service scenarios across all stores with consistent quality, thus expanding opportunities for personnel development without incurring significantly more work hours.
The second problem is the high level of flexibility on the ground. In many cases performing planned training in stores is difficult due to factors such as the waves of business they experience and the shift system they work on. Generative AI can respond to inquiries when staff face issues and can implement “close-to-the-ground-model instant learning” of the kind that conventional training cannot provide enough of.
The third problem is that of eliminating individualization　in personnel development and ensuring reproducibility. Customer service by veteran in-store employees not only takes product knowledge, but also includes a great deal in terms of company-specific behavioral guidelines and decision-making criteria. Much of this, however, remains unverbalized, implicit knowledge. Generative AI structures such knowledge and functions as a “reproducible development platform” that allows anyone who uses it to arrive at the same way of thinking and the same processes for handling customers. These characteristics allow generative AI to serve as a means of stabilizing personnel development that had been difficult to achieve through conventional OJT and simultaneously driving enhancements, while meeting the demand for learning on the ground. In the next section, we will build on this background to go over in greater detail the structural characteristics of customer service knowledge and how generative AI can help transmit this.

3. Can Generative AI Transmit Veteran Customer Service Knowledge?

In this section, we analyze the structure of customer service knowledge held uniquely by veteran in-store employees and reveal the requirements and challenges involved in using generative AI to transmit this.

The “Company Character” Inherent to Veteran Customer Service

So, what is different about the customer service of veteran in-store employees compared to junior employees or part-time employees? Is it extensive product knowledge? Or an accumulation of know-how about how to quickly respond to customer questions? Naturally, these elements play a big part. With that said, beyond those elements, it is also likely that a certain “company character” that develops unconsciously over many years of helping customers is also an important part of veteran in-store employee customer service. In practice, customer service differs according to the culture of the given company, even when dealing with the same product or being asked the same question by a customer. Veteran employees no doubt make recommendations or proposals and handle complaints in line with company policies and behavioral guidelines. In some cases, veteran in-store employees who have worked at a company over long years will even have built up that culture.

Barriers to Analysis and Structuring of Implicit Knowledge

Veteran in-store employee customer service is more than just a concentration of simple product knowledge. Rather, it can be thought of as being composed of the following elements.

Company policies and behavioral guidelines
Customer service manuals and internal rules
Product knowledge
Personal experience and unformalized implicit knowledge

Such total “knowledge” can be thought of as contributing to the customer service ability of veteran in-store employees. We therefore need to reflect this structure in the design of inputs to AI.
Many companies have already prepared store manuals and established clear company policies and behavioral guidelines. However, what constitutes “specific customer service” in line with those policies and behavioral guidelines is something that people can picture, but that is often not formalized. For that reason, companies need to take stock of their customer service operations and organize a comprehensive knowledge base per scenario, making explicit “how to respond in such and such a situation.” Companies thus verbalize the intuition of veteran in-store employees and teach it to the AI. It is precisely this translation work that forms the single biggest barrier to building a knowledge AI.

Roles and Design Philosophies Per Methodology in Building Knowledge AIs

The second hurdle in this process is “capacity to build AIs.”
In building a knowledge AI, companies need to combine multiple approaches as appropriate depending on their goals. Firstly, the process of system prompt design is a method for adjusting the direction of responses by devising command texts for the AI. On the ground in retail, giving AI commands to “explain in a tone that gets close to the customer” and “organize points in terms of three criteria: price, characteristics and aim of use when comparing products” can allow us to make the AI learn patterns of speech and comparison perspectives that veteran in-store employees perform on a daily basis. Another potential application is, in handling complaints, commanding the AI to follow a method of “(1) apologizing -> (2) empathizing -> (3) organizing the facts -> (4) presenting an alternative,” so that the complaint handling flow intended by the company is reproduced. Devising such system prompts constitutes an approach that improves response quality and consistency without making the model itself learn anything.

Next, Retrieval-Augmented Generation (RAG) is a mechanism for enabling AI to reference existing internal data in the form of manuals, product information and FAQs. For example, by using RAG to refer to information that should not be learned internally to the model because it is updated frequently, such as new product specifications or campaign information, the AI can accurately incorporate the latest information into its responses. RAG thus fulfills the role of responding with “information you want to know right now” in store by searching the latest product catalogues and answering based on that, when, for example, an in-store staff member asks it a specific question such as “can you use this drill on concrete?”

Fine tuning, on the other hand, is a method aimed at getting the AI model itself to do additional learning so that it can more deeply understand company-specific knowledge and customer service policies. For example, by getting the AI to internalize typical company phrasings veterans have learned from experience or implicit rules such as “at our company, we propose products to customers after checking the purpose for which the customer wants to use them,” companies can improve the reproducibility and consistency of responses. Infrequent but essential knowledge that does not extend over the long term, such as specific customer service scenarios or decision-making criteria, in particular, are suitable targets for fine tuning.

System prompt design, RAG construction and fine tuning thus all differ in their properties and can be broadly categorized into “methods for adjusting the usage of the model” and “methods for training the model itself.” It is thus exceptionally important in constructing a knowledge AI to design the flow of how the AI is made to understand different information using different methods, and in what order it processes that information, based on the strengths of each approach.

On the Ground Actions to Ensure Successful Knowledge Transfer

To clear these hurdles, it is important for companies to verbalize and formalize customer service practices that tend to be individualized based on day-to-day behaviors. Creating mechanisms to absorb customer questions and responses to those questions day by day, and sharing those among team members on the sales floor, collecting common customer inquiries and methods for responding to them, and templatizing customer service phrasings and key decision-making points so that even inexperienced staff can apply them – all of these efforts contribute to the transmission of knowledge and the creation of a ready-to-go workforce.
Technical insight is required in order to have capacity to build AI. However, even in the retail industry, the discussion has moved past the stage of talking about “whether to use” AI to the stage of asking “how to use AI,” leaving companies no choice but to make use of the technology. While the utilization of specialist external resources is core to the implementation of AI, the technology has reached the point where companies should also incorporate highly motivated personnel from within as project members and handle projects by including internal personnel development within their scope.

4. Reframing Personnel Development Away From “Teaching” and Towards “Reproducing”

This section will discuss the need for companies to move on from conventional “teaching” personnel development towards a “reproducibility development” that allows all employees to perform the same customer service. New mechanisms of knowledge provision, skill acquisition and feedback leveraging generative AI are key to evolving in-store personnel development.

Realizing Reproducible Personnel Development Through Generative AI

The essence of personnel development in stores is not “teaching” itself, so much as it is “enabling anyone, anytime, anywhere to perform the customer service intended by the company.” This is the essence of the idea of “reproducibility.” Going forward, what will be important is not teaching and developing staff, but having ideas for how to create reproducible mechanisms.
Generative AI is a technology that is an order of magnitude more effective in applications that involve using stored up knowledge in a variety of forms. Rather than just having users ask it questions and get answers when they run into trouble, the technology allows companies to build mechanisms for enabling users to perform what veteran employees do, facilitating, for example, product knowledge acquisition in response form or skilling up through role plays of pitches or complaint handling. And more important than anything else is the design of feedback. In traditional OJT and e-learning, the focus tends to be on “whether something was taught,” with questions of “was it understood” and “can it be reproduced” often tending to become a black box. As part of knowledge acquisition, it is essential to build systems that let staff objectively understand their own degree of mastery and the quality of their own customer handling.
For example, when roleplaying the handling of a customer complaint, the AI can evaluate how the staff member handled the issue, offer comments suggesting improvements, such as “did you empathize with the customer enough” and “was the alternative proposal put forward clearly enough,” and give feedback on the outcome through a numerical score or rank. During product knowledge acquisition, the AI can also analyze the product categories a member of staff struggles with (e.g., specialist knowledge of white goods, explanations of the ingredients in cosmetics, etc.), present this as an area for concentrated study, and help them consolidate their knowledge in a short span of time. It is also important for the AI to extract information from the frequency and content of QAs being accessed, analyze what knowledge in-store staff are accessing most often, and connect this to further extending knowledge reinforcement.
Generative AI thus surfaces the learning and instruction processes that, under conventional methods, would have been influenced by each individual’s experience, and plays a core role in achieving “reproducible personnel development” that allows anyone to approach the thought processes of veteran employees.

5. The Future-Facing Extensibility of Customer Service Knowledge AI

This section surveys the potential for applications of customer service knowledge AI beyond personnel development applications. Structured customer service knowledge has the potential to do more than just support employees. Potentially, it could even be extended to the automation of customer handling and to improving customer experience (CX).

Customer Service Knowledge Has Strong Potential for Use Beyond Personnel Development

The “customer service knowledge AI” covered in this Insight sets its sights on structuring in a reproducible form the customer service skills and product knowledge currently built up by veteran employees in an individualized manner, and using generative AI to create mechanisms that allow in-store staff to learn and master this information anywhere. However, this is no more than a first step in the use of knowledge AI, and future extensibility is extremely important.
The applications of knowledge go beyond a company’s employees. If customer service knowledge and product knowledge can be structured with enough precision, the AI could be reapplied as is to being a customer service AI or a chatbot. This would allow it to contribute to automating and enhancing e-commerce sites, official apps and customer service channels, and give AI the potential to directly support the core business itself in the retail sector. In particular, integrating Q&A data built up through consumer chat histories and customer service knowledge AI would directly contribute to CX transformation.
Generative AI could thus reproduce the individualized knowledge of veteran employees and open the path towards a future where AI could be used as a corporate asset to improve organizational capacity and extend the business.

When Should Your Company Start Its Knowledge AI Initiatives?

Because AI is evolving at an extremely fast pace, we often hear clients say that they do not know when they should start with the technology. What impacts the success of utilizing AI, however, is inputs. Even if AI itself advances, without the data to put into it, or with only poor quality data, companies will not get the outcomes they are hoping for. Going forward, it will also be essential for companies to develop personnel who can come up with ideas for “what AI can do.” To avoid being in a situation where they are unable to make progress once they have started using AI, companies should take the first step of taking stock of and structuring their knowledge, preparing for the full use of generative AI and rebuilding their “foundations” for personnel development.

6. Conclusion

ABeam Consulting offers clients comprehensive support for not only implementing AI tools for personnel development, but also for structuring customer service knowledge, based on the restrictions, and organizational and operational design considerations involved in on-the-ground implementation. Implementing knowledge AI is not a one-and-done. Rather, sustained interlinking of data governance and HR and training departments, overall design, including the design of evaluation indicators, as well as data and AI optimization are all key to the process. We strive to continue supporting the development of knowledge AI, while working side by side with our clients, so that the utilization of knowledge AI does not “end at the PoC stage.”

Insights TOP