In the previous post in this series, we examined how the four methodological shifts are reshaping retail customer research — an industry advantaged by direct access to population-scale transactional data and challenged by the volume of unstructured customer expression it accumulates. This post turns to fast-moving consumer goods, an industry that faces an almost exact mirror image of the retail data situation. The shifts still apply, but their configuration is fundamentally different, and so is the practical work of integrating them.

Fast-moving consumer goods is, for the purposes of this discussion, the world of branded packaged products — food and beverages, household care, personal care, the categories that move through retail at high frequency and that depend on brand-building for their margin structure. It is also the industry in which classical market research was, in a sense, invented; the syndicated panel methodologies of NielsenIQ, GfK, and Circana grew up alongside the postwar FMCG industry and have been its primary external customer-evidence infrastructure for decades.

The structural problem: FMCG does not own the transaction

The defining feature of the FMCG data situation is that the manufacturer does not own the transaction. When a customer buys a packaged-food product at a supermarket, the supermarket records the transaction; the manufacturer learns about it only through aggregated retailer data, syndicated panel estimates, or bespoke research with consumers recruited outside any actual purchase context.

This is the mirror image of the retail situation. Where retailers have rich behavioral trace data about specific customers and limited insight into brand-level positioning dynamics, FMCG manufacturers have rich brand and category data and limited insight into the specific customers who buy them. The first methodological shift — from sample to trace — applies differently here. Manufacturer trace data is concentrated where the manufacturer does own a direct interaction: e-commerce on its own site, app usage, directly-run loyalty programs, samples-and-rewards programs. These are typically a small fraction of total category transactions.

The traditional response to this gap was the syndicated panel. By recruiting representative panels of households and tracking their purchasing in detail, panel providers offered FMCG manufacturers an approximation of population-level transactional data they could not access directly. This worked, more or less, for the past sixty years. Two things have changed.

First, the same panel-recruitment challenges that have weakened classical survey research have begun to affect syndicated panels as well: declining response rates, declining panel persistence, rising costs, and a granularity demanded by modern decision-making that classical panels struggle to support.

Second, alternative data sources have proliferated. Direct-to-consumer channels, loyalty partnerships with retailers, embedded sensors and packaging-based digital interfaces, social-listening pipelines, and direct app interactions all provide manufacturers with new windows into customer behavior that did not exist a decade ago.

The first shift in FMCG: composite trace data

The result is that the first methodological shift takes a specific FMCG form. Manufacturers are not moving from a single sample-based method to a single trace-based method; they are assembling composite trace data from multiple partial sources, each of which covers a slice of customer behavior and none of which covers the whole.

This is harder than what retailers do, and it requires capabilities the classical FMCG research function did not need: data integration across sources with different identifiers, methodological judgment about how to combine evidence of different reliability, and explicit modeling of what each source can and cannot see. Some manufacturers have built these capabilities; others have not, and the gap between them is becoming a significant competitive variable.

Notably, syndicated panels continue to play a role in the composite — both as the most consistent longitudinal view of category dynamics, and as a ground-truth anchor against which the other partial sources can be calibrated. The shift in FMCG is not from panels to trace; it is from panels-alone to panels-plus-composite-trace.

The second shift in FMCG: brand-level customer expression

The second methodological shift — from asking to reading — operates with particular force in FMCG, because the unstructured customer expression about packaged products is, in many cases, extraordinarily rich.

People talk about food and beverages, household products, and personal-care items continuously. They post about new product launches, review specific variants, complain when formulations change, advocate for products they love, discuss categories on forums and in social communities, and record cooking, cleaning, and grooming routines on video platforms. Unlike retail-store reviews — which concentrate around the immediate purchase moment — FMCG customer expression spans the entire usage lifecycle.

For most of the past century, manufacturers' access to this expression was limited by analytical cost. Brand-tracking studies, focus groups, and ethnographic research gave them deliberate, structured access to small samples; the much larger volume of unprompted expression was, beyond the most prominent mentions, effectively invisible. Large language models change this economics, and three distinctive FMCG use cases follow.

The first is brand-perception monitoring at category granularity. Where classical brand-tracking studies measure perception through abstract survey questions, reading allows monitoring of how brands are actually discussed in their categories — which attributes are credited to which brands, where positioning is shifting, where competitive narratives are succeeding or failing.

The second is usage and occasion intelligence. FMCG marketing depends critically on understanding how products are actually used — at what occasions, in combination with what other products, by what kinds of customers, with what surrounding rituals. Customers describe their actual usage extensively in social and review channels. Reading this expression at scale produces an understanding of usage and occasion that was previously the exclusive domain of expensive ethnographic studies with small samples.

The third is innovation signal. When customers complain that a product does not exist that they wish existed, describe workarounds they have developed for product shortcomings, or articulate dissatisfaction with category norms, they are providing innovation signal that surveys structurally cannot capture. Reading systematically across the relevant expression channels surfaces this signal in a form that connects to specific product-development decisions.

The third shift in FMCG: synthetic audiences as concept-screening at scale

The third methodological shift — from asking to simulating — has perhaps its clearest natural home in FMCG.

FMCG innovation pipelines produce more product concepts, packaging designs, advertising variants, and positioning ideas than even the most well-resourced research function can field-test through real fieldwork. Concept-screening systems built around synthetic audiences allow manufacturers to evaluate hundreds of variants at the cost of evaluating ten, with the understanding that the synthetic step is screening and not validation. Real fieldwork remains essential for the candidates that survive screening, but the screening itself can now operate at a scale that classical methods could not support.

The methodological caveats apply with particular force in FMCG: claims about specific real-world products require careful validation against the manufacturer's actual customer base, and high-stakes commitments — major launches, repositioning of established brands, significant pricing moves — should never rest on synthetic evidence alone. But within these constraints, the FMCG use case is robust.

The fourth shift in FMCG: from quarterly trackers to continuous brand intelligence

The fourth methodological shift — from insights report to insights infrastructure — has a specific FMCG dimension. FMCG brand tracking has been one of the most ritualized practices in classical market research. Quarterly or annual brand-equity studies have served as the primary mechanism through which manufacturers monitor the health of their brands. These studies are valuable and will continue to have a role. But they share with all periodic reports the problem that they describe a quarter that has already ended, and they answer only the questions the study designers thought to include.

The emerging FMCG insights infrastructure complements rather than replaces brand-tracking. A composite trace-data layer, a customer-expression analysis layer running continuously against social and review channels, a synthetic-audience layer for hypothetical questions, and a question interface that lets brand teams query the combined evidence — together these provide a continuous brand-intelligence capability that quarterly trackers alone cannot. The quarterly tracker remains useful as an executive-synthesis instrument; the continuous infrastructure handles the day-to-day operational questions that the quarterly cadence could never address.

What this looks like in practice

A consumer-goods manufacturer that has integrated the four shifts looks different from one that has not — but differently than the retailer we described in the previous post.

It has assembled a composite trace-data layer combining direct interactions, retailer partnerships, syndicated panel data, and selected third-party sources, with explicit methodological treatment of what each can and cannot see. It runs continuous reading against customer expression in social, review, and community channels, with the qualitative signal connecting back to brand, category, and innovation decisions. It uses synthetic-audience methods systematically at the concept-screening layer of its innovation pipeline. And it operates a continuous brand-intelligence infrastructure that augments — without replacing — its classical brand-tracking apparatus.

The manufacturer that has not made these moves still operates the brand-tracking and panel-purchasing infrastructure that the industry built in the postwar decades. That infrastructure continues to produce useful evidence. But its share of the manufacturer's effective customer understanding is shrinking, because the evidence available outside of it is growing so much faster.

What this does not mean

This is not the death of the syndicated panel or of classical brand research. Both remain valuable, and the manufacturer that abandons them in pursuit of pure AI methods will discover quickly that the foundation they provided is harder to reconstitute than to maintain. The shift is from these methods alone to these methods plus the new capabilities — augmentation, not replacement. It is also not a claim that the new capabilities are easy to integrate: composite trace-data assembly, continuous expression analysis, calibrated synthetic-audience operation, and a continuous brand-intelligence infrastructure are each substantial undertakings. The manufacturers that benefit from them are the ones that have invested in the underlying capabilities over years.

In the final industry post of this series, we will turn to financial services — where the data situation differs again, the regulatory environment imposes constraints that retail and FMCG do not face in the same form, and the shifts take yet a different configuration.