How GenAI Can and Can’t Help Manage Customer Insights

News

1. Leadership Transition at Jindal Steel as CEO Malhotra Exits

2. Meta Removes AI Image Feature After Instagram Privacy Backlash

3. Nadella warns of AI’s Reverse Information Paradox

4. TCS Expands AI Push With Client-Facing Engineers

5. Apple OpenAI Fight Signals AI’s Hardware Shift

6. OpenAI’s GPT-5.6 Puts Enterprise AI on a Cost and Speed Clock

To understand their customers and markets, a growing number of customer-oriented companies are using generative AI tools, alongside the language and reasoning capabilities of popular large language models (LLMs), to access and analyze their own internal content. These hybrid knowledge approaches, which typically employ a technique called retrieval-augmented generation (RAG), allow the integration of a company’s own customer insights with the general knowledge base on which an LLM was trained. Companies taking this approach to what was previously called “knowledge management” reap several benefits, such as enabling employees to access and summarize content in natural language. That capability is particularly important in large organizations — where employees searching for insights often have no idea where those insights originated or how they might be found.

Organizations gather and attend to insights about what their customers want, how they want to be sold to, and what products and services they have interest in. Customer insights typically originate from market research departments or external market research agencies, but they can also be found in sales interactions, customer letters and emails, website behaviors, social media interactions, customer service tickets, purchase patterns, focus groups, and other channels. It can add up to a lot of structured and unstructured information, so tools to help summarize, categorize, store, and access it certainly make sense.

However, companies that focus exclusively on storing and accessing knowledge are making the same error that many organizations made in the earlier generation of knowledge management: That focus is too narrow. Instead, companies should also address knowledge flows — how customer and market insights are created, analyzed, stored, and accessed. New technologies — using generative AI to at least some degree — can assist with all of those steps.

While earlier tools for knowledge management (such as Lotus Notes and Microsoft SharePoint) allowed broad access to customer and market insight content, many organizations did not experience a revolution in insights access and usage. The technology provided more access, but cultural challenges — difficulty in organizing and retrieving the knowledge, indifference to the content, organizational silos and overlaps, lack of collaboration with external agencies — often prevailed. Today, organizations still face those same cultural obstacles when tackling knowledge management.

To learn how generative AI can and can’t help leaders beat knowledge management obstacles, we spoke with customer and market insight specialists or leaders at eight different consumer-oriented companies. Some of these leaders had responsibility for creating and overseeing the generative AI system; others had broader responsibility for market research or insights-oriented technology. We also spoke with several vendors of GenAI-based technology for customer and market insights, and one market research agency that makes extensive use of GenAI-based qualitative research technology.

How Generative AI Tools Help

Companies can combine GenAI tools with customer and market insight content on their own (though it may be difficult), use external software vendors to simplify the work, or use a combination of both approaches. Procter & Gamble, for example, uses vendor-supplied software for knowledge storage and access but has created its own system for GenAI-based analysis and categorization of the content. As Kirti Singh, P&G’s chief analytics, insights and media officer, noted, this way they get “sharp, pointed answers from GenAI” rather than just links to documents.

Among the tool vendors, focus differs. Some primarily focus on the storage of and access to insights, but the tools’ value goes beyond providing simple automated virtual filing cabinets. Their functions include automated curation of documents, integration of diverse content types, on-demand analysis done in response to queries, and synthesized answers to prompts. This type of tool, however, assumes that data analysis has already been done and that insights are waiting to be found. Analysis of quantitative data is done by analytical software, and the capacity to analyze qualitative data is limited.

Other vendors concentrate on the analysis of qualitative customer data and documents for customer insights. Still other vendors emphasize rapid testing of consumer responses to advertising. Today, there is overlap among these categories, and most vendors are attempting to address the broad process of identifying or creating customer insights, curating and categorizing them, and making them available for later access. We expect that at some point, broad customer insights platforms will emerge, employing generative AI and other capabilities to address the entire process.

One customer insights leader told us, “AI is only as useful as the data it learns from.”

If the vendor’s primary function is insight storage and access, the approach typically involves adding the customer’s custom and proprietary content on top of large language models. The content stored can include structured data (quantitative market research results, spreadsheets, or customer satisfaction ratings, for example) and unstructured data (such as transcripts of interviews, social media comments, or focus group results) from both internal and external sources.

In most cases, organizations pursuing this route centralize as much customer and market content as possible in one system. Curation is required to reduce content overlaps, eliminate obsolete/irrelevant documents, and generally maintain quality. As one customer insights leader told us, “AI is only as useful as the data it learns from.” To make sense of insights, the GenAI tools typically must tackle categorization, summarization, and content tagging. Tagging the content makes it more likely to be retrieved later. Some companies use manual tagging, while some systems employ GenAI-based tagging using a predefined taxonomy.

Novartis offers an example of a company successfully revamping insight storage and access using GenAI tools. Working with an external vendor, the company developed a customer and market insights system, called Sherlock, for its consumer business. After users pose questions, the system gives answers by pointing to a specific line of text or a time stamp in a video. Sherlock also incorporates expert-curated microsites, known as Knowledge Zones, on particular topics, such as packaging. Users who add content to the system must adhere to strict governance guidelines about document formats and quality. Novartis’s research vendors can upload project deliverables directly into Sherlock.

The system helps Novartis avoid spending on redundant insights services across its business and helps employees find relevant insights quickly, without overgeneralization. (For example, it could flag results that were based on patient data from Europe only, using a feature called WatchOut.) The results have added up: Novartis saved more than $29 million in primary market research costs in just one year. Such use of GenAI to enhance insight storage and access can facilitate the democratization of information by helping employees find both knowledge and knowledgeable people.

Qualitative Data Analysis: A Special Problem

For a long time, the ability to do qualitative data analysis — a messy business — has not been included in off-the-shelf analytical tools. Specialized software for this purpose has mostly been used in academic qualitative research; historically, most market researchers have conducted semi-manual analyses, using spreadsheets. GenAI tools offer an alternative to this extremely time-consuming qualitative analysis work.

A recent academic article argued that GenAI is unsuited to qualitative analysis, but our analysis suggests that this is true only of generic AI chatbots. More specialized tools have other capabilities that make them suited to effective qualitative data analysis.

Tracy Tuten, who leads qualitative research at market research agency Illuminas (now part of Radius Insights), became an early adopter of a vendor’s generative AI software in order to mine customer insights. Tuten, who has taught market research at several universities, refers to this approach as “conversational qualitative data analysis.”

Via GenAI-based software, Tuten uses natural language prompts to analyze qualitative data from interviews and focus groups. The system lets her upload audio and video files for automatic transcription, generate summaries, surface themes, and compare them across audience segments. A large-scale qualitative project such as a global study with 30-plus interviews might have taken six weeks to analyze in the past but can now be synthesized in a day, Tuten said. The tool also lets her surface secondary insights that she might have missed in the unstructured data. Tuten often uses the software collaboratively with clients in workshops, enabling faster, more participatory insight discovery.

Given that many qualitative researchers previously relied on spreadsheets and manual cut-and-paste coding to analyze data, this AI-based approach represents a major advance in efficiency and rigor. However, uncritical use of generative AI or other forms of AI may have significant shortcomings. Conversational qualitative data analysis does not replace the researchers; it only augments their performance.

PepsiCo makes extensive use of software for creating customer and market knowledge, including both structured and unstructured data. The company has particularly focused on determining how customers respond to specific advertising campaign and brand messages. But that isn’t the only application the company has employed. In an interview with us and in The Consumer Insights Revolution, a book describing a transformation of customer insights at PepsiCo, Stephan Gans, senior vice president and chief customer insights and analytics officer, described the company’s “platform” for marketing research, called Ask Ada. It includes:

• The ability to test new creative content on real or synthetic customers.
• A data repository on the results from advertising, influencer, and other types of campaigns.
• Social listening capabilities.
• Predictive modeling.
• Knowledge management of customer insight lessons, present and past, including meta-learning.
• A conversational interface to all Ask Ada content.
• Gans also credited Ask Ada with reducing PepsiCo’s dependence on external agencies and consultants.

Why AI Isn’t Enough: Four Challenging Factors

Despite these new capabilities, a GenAI tool cannot replace a marketing leader for strategy work. As Gans noted, “Raising the bar on marketing and innovation effectiveness to fuel commercial excellence will become increasingly automated. Leading the understanding of consumer demand is much more strategic and still requires humans.”

As we discovered in our research, several important issues inhibit the ability of AI technology to transform customer and market insights. These issues predate knowledge management and generative AI and will cause problems if not addressed. Let’s examine four of these factors, along with examples of organizations that have encountered and overcome them.

1. Geographical and business units lack common approaches.

One global consumer goods company where we conducted interviews had acquired a GenAI-based customer and market knowledge tool from a vendor but thus far had made little progress improving access to global knowledge. The problem was that the company does business in over 100 countries, and country-based units have a large degree of autonomy. There’s no companywide consensus on names for brands, categories, and distribution approaches. “We have pockets of knowledge, but they are very incoherent,” the head of knowledge management for insights told us. “Different people are investing in different things. We have contradicting numbers and outputs — they are all contextualized differently.” No senior executive has attempted to create greater commonality of information and knowledge formats across units; it would be viewed as counter to the company culture. As a result, the generative AI tool is used by only a few geographical units, and the company’s marketers and product developers are unable to learn from each other. The company is attempting to develop a strategy to create a more centralized, global approach.

At PepsiCo, the story is different. When Gans was first named chief consumer insights and analytics officer in 2017, the company had a diverse set of approaches to customer and market insights. But Gans wanted to create “one nation” of market research so that the company’s marketers could learn from one another and share relevant insights. With strong support from the CMO, Gans created the Global Insights Council, which comprised 15 insights leaders representing all regions and central/global capabilities. Today, customer insights are tightly integrated into PepsiCo’s innovation work.

2. Customer and market insights aren’t part of strategy and culture.

Even the best technologies won’t succeed if the organization’s decision makers aren’t ardent consumers of that type of knowledge. Without those passionate consumers of information, a company’s market research efforts will fall flat, our research showed. So culture and change management work will often be required of leaders.

Without passionate consumers of information, a company’s market research efforts will fall flat.

Passion for data runs high at P&G, which is known for its long-term focus on being customer- and market-driven. Indeed, P&G recently celebrated its 100th year of market research; in 1924, the then-CEO asked a researcher to determine why customers were buying Ivory soap. P&G’s Singh told us, “At the heart of everything we do is the consumer. … Our strategy is to provide a superior product experience to our consumers. We employ experimental science, human and behavioral science, data science, and technology platform knowledge to understand our consumers.”

3. Agency relationships introduce data ownership complexity.

Many companies use external advertising and marketing agencies for consumer research. The client/agency relationship may lead to uncertainties and dysfunction involving analysis strategies, interpretations of analyses, and ongoing ownership of the data and results. If agencies end up owning all or most customer and market insights, a company’s employees will be unable to meet customer needs without external help. PepsiCo’s Gans argued strongly that the client company has to own all research results and insights created by agencies on the client’s behalf. He added that it’s not a good idea to own the data but then outsource the learning from it, because employees should apply lessons learned from market research in future campaigns.

However, for consumer-oriented companies that continue to work with agencies, some vendor software can facilitate collaboration between clients and agencies. Both parties can view, edit, and query customer research and produce a variety of outputs.

4. Analytics professionals may be seen as low-status “order takers.”

If that is the case in your organization, that reputation needs to change. At one of the consumer products companies where we conducted interviews, the insights and analytics function always had a library-like focus. Previously, internal customers who were interested in insights had to consult with a researcher or “librarian.” With the advent of a vendor-supplied GenAI tool, the function has been democratized.

However, users of the company’s system still treat it as a library; they don’t contribute much to the stock of insights. Enabling internal customers to serve themselves hasn’t substantially increased demand for the content and analysis. Users also don’t always supply high-quality prompts; they might ask, “What do we know about back to school?” not realizing that the company has decades of market research on the topic. As at Novartis, experts created micro-sites of curated content within the platform, to address particular information-access issues for certain areas. Still, budgets and head counts in the insights and analytics function have been cut in recent years. Similar functions outside the U.S. don’t want to pay for the GenAI-enabled tool, so they take different approaches to customer and market knowledge.

PepsiCo previously had something of an “order-taking” mentality for market research, and researchers were rarely asked to collaborate with the internal customers or help to shape the requests for insight. Research team members had little respect for their roles, and the function was asked to cut its budget several times. When Gans arrived in the leadership role for the function, he and the CMO concluded that PepsiCo was spending hundreds of millions of dollars per year on consumer insights and that it made little sense to do that unless the company were to become more customer-centric. He made a series of changes — including the implementation of a new AI and insights software system and the Ask Ada platform — that eventually made the customer insights and analytics organization well respected and well funded.

 

---

Overall, any AI tool should not be considered a replacement for what the organization is already doing well. As P&G’s Singh told us, “We brought together our tradition of being focused on understanding the customer with the latest AI solutions — not replacing, for example, customer home visits, but augmenting them with AI.”

Our study also suggests that the AI-enabled software for managing customer and market insights is evolving rapidly. Leaders are often interested in different features and functions that fit their company’s specific situation. But leaders should be aware that today’s software has limits. In poorly integrated global organizations, humans have created different names for customers, brands, and marketing approaches across geographies; even AI can’t pull together a unified set of customer and market insights in that case. Companies need humans to integrate and standardize that data to analyze it and act on it effectively.

Finally, if a company’s workforce isn’t actually interested in gathering and acting on customer and market insights, no software is likely to change that situation. These are problems that need to be addressed by humans, not AI.