In the first five posts in this series, we mapped the methodological reorganization of customer research — the three classical pillars under pressure, and the four shifts that are emerging in response. This post is the first of three in which we leave the methodological frame and look at how these shifts are playing out concretely in specific industries. We start with retail, for a reason: of the three industries we cover, retail is the one in which all four shifts apply most directly, and in which the asymmetry between organizations that have adapted and those that have not is becoming most visible.

Retail customer research has a distinctive structural advantage and a distinctive structural problem, and AI is reshaping both at once.

The retail advantage: trace data at population scale

The structural advantage is that retailers own the transaction. When a customer buys something, the retailer records what they bought, when, at what price, through which channel, in combination with what other items, and — if the customer is identifiable through a loyalty program, a customer account, or a payment-method link — over what trajectory of prior and subsequent purchases.

This is the richest behavioral trace data of any consumer-facing industry. Manufacturers know what their products did at retailers' shelves but not which customers bought them; service providers know what their customers did with their services but not what those customers did anywhere else; media platforms know what their users consumed but not what they bought. Retailers know what specific identifiable customers actually purchased, across categories, over years. The first methodological shift we described — from sample to trace — is, for retailers, not a future state but a current reality. The question is not whether the trace data exists but how completely it is being used.

In practice, the answer varies enormously. Leading retailers have built customer-data infrastructures that integrate loyalty, e-commerce, and store-level transactional data into unified customer views, with the capability to query that data in close to real time. Other retailers have the same data but in fragmented systems, queryable only through long IT projects, with the result that decisions about pricing, assortment, and merchandising are made on aggregate views that obscure most of what the data could show.

This unevenness is the first AI-driven competitive divergence in retail customer research. It is not really about AI as such; it is about whether the foundational data infrastructure that AI methods depend on has been built. But AI raises the stakes of the underlying infrastructure choice, because the value that can be extracted from well-integrated retail trace data has grown substantially with what large language models and modern analytics now make possible against it.

The retail problem: customer expression at scale, much of it noise

The structural problem is that retailers also receive enormous volumes of unstructured customer expression — product reviews, support tickets, social mentions, return-reason notes, in-app feedback — and have historically struggled to do anything systematic with most of it.

A large retailer accumulates millions of product reviews per year across its assortment. Each review contains, in principle, actionable signal: which products are disappointing customers in which specific ways, which products are exceeding expectations and why, where the sizing runs small, where the delivery experience is failing, where the description on the site is misleading. In practice, before large language models made unstructured analysis tractable at scale, almost none of this was systematically harvested. Reviews were used reactively (to triage individual customer complaints) and aggregatively (to show star ratings), but the qualitative content sat unread by anyone with the authority to act on it.

The second methodological shift — from asking to reading — addresses exactly this. Modern systems can process retail customer expression at scale, surface themes by product and by category, distinguish genuine product issues from generic dissatisfaction, and connect linguistic patterns to specific operational decisions: which products to reposition, which descriptions to rewrite, which supplier relationships to renegotiate, which fulfillment partners to evaluate. The shift is from a retailer that had customer expression to one that understood it.

Here, too, the unevenness is substantial. Some retailers have integrated systematic review and customer-language analysis into their merchandising and operations workflows; others continue to treat customer expression as a customer-service inbox. The competitive consequences of the difference are growing.

The retail use case for synthetic audiences

The third shift — from asking to simulating — has a specific retail application that is worth being explicit about, because it differs from the canonical synthetic-audience use case in consumer-goods research.

Retailers face a continuous flow of decisions about assortment, pricing, promotion, and merchandising that have to be made faster than classical research can support. Should this product be promoted at this price point next week? Should this category receive expanded shelf space at the cost of an adjacent category? Should this private-label launch be positioned as premium or value? Each of these decisions, in isolation, has historically been made on a combination of historical trace data, category-manager judgment, and ad-hoc qualitative input. Real customer research has rarely been feasible at the cadence at which the decisions are made.

Synthetic audiences offer retailers a way to introduce structured customer evidence into decisions that would previously have been made on judgment alone — not as the sole basis for those decisions, but as one signal among several, available at the cadence of the decisions themselves. A category manager weighing two private-label positioning options can, in a synthetic-audience workflow, see directional customer response to both within hours, validated against the retailer's actual customer base through calibration against historical patterns. This is not a replacement for the broader assortment research the retailer also conducts; it is a way to bring evidence into decisions that previously had none.

The methodological caveats from the synthetic-audience post apply: directional rather than definitive, validated against ground truth where possible, never the sole basis for high-stakes commitments. But the use case is genuine, and retail is one of the industries in which it is most clearly productive.

The retail challenge: continuous decision rhythms meet continuous infrastructure

The fourth shift — from insights report to insights infrastructure — is also most pronounced in retail.

Retail decision cadence is uniquely fast among consumer-facing industries. Pricing is reviewed weekly or daily; promotional plans run on monthly cycles; assortment decisions on seasonal cycles; merchandising on continuous in-store and on-site operational rhythms. None of these match the quarterly cadence of classical customer research. The mismatch was tolerated for decades because no faster customer-evidence layer existed; retailers compensated by relying heavily on category-manager intuition for decisions between research waves.

The continuous insights infrastructure described in the previous post is, for retail, not an organizational nice-to-have but a competitive necessity. Retailers that build infrastructure capable of delivering customer evidence at the cadence of retail decisions gain advantages over those that do not — in pricing precision, in assortment relevance, in promotional effectiveness, in private-label development. The shift from periodic to continuous is, in retail, also a shift from intuition-driven to evidence-driven for a substantial class of operational decisions that previously could not afford evidence at all.

What this looks like in practice

A retailer that has integrated all four shifts looks materially different from a retailer that has not.

It has a customer-data foundation that connects loyalty, e-commerce, and store-level transactional data into unified customer views with real-time queryability. It has systematic harvesting of customer expression from reviews, support interactions, and social channels, with the qualitative signal connecting back to specific products, suppliers, and operational decisions. It uses synthetic-audience methods to bring directional customer evidence into pricing, promotion, and assortment decisions that previously had to be made on judgment alone. And it operates a continuous insights capability that delivers evidence at the cadence of retail decisions rather than the cadence of quarterly research.

A retailer that has not integrated these shifts still has the same trace data, the same customer expression, and the same decision cadence; it simply leaves most of the available evidence unused, makes decisions on intuition where evidence could inform them, and waits for quarterly reports to confirm patterns the data showed months earlier.

The gap between the two is not subtle, and it compounds. The retailer that decides assortment with evidence beats the retailer that decides on intuition, in margin terms, in inventory-efficiency terms, and in customer-loyalty terms. The gap that exists at the start of any given year is, by the end of the year, larger.

What this does not mean

Two clarifications, because the discourse on AI in retail tends to reach for stronger claims than the evidence supports.

This is not the death of retail merchandising as a craft. Category management, buying judgment, and the intangible sense of what will work in a specific store for a specific customer base remain essential. The shift the four methodological moves enable is not the replacement of merchandising judgment by AI but the augmentation of merchandising judgment by evidence that was not previously available. Retailers that interpret the shift as a substitution miss what is actually happening.

And this is not a claim that the technology alone determines outcomes. The retailers that benefit from these shifts are the ones that have rebuilt their operational and decision processes around what the new methods make possible. Buying the technology without rebuilding the processes — which a number of retailers have tried — produces the same disappointing results that buying any other technology produces when the surrounding organization stays unchanged.

In the next post in this series, we will turn to fast-moving consumer goods, where the structural advantage that retailers have — direct access to transactional data — is precisely the structural disadvantage that consumer-goods manufacturers face. The shifts apply, but the configuration is different, and so are the use cases that work.