How synthetic consumer panels are replacing the slowest, most expensive habit in consumer goods
The Shelf Is a Graveyard
Walk into any large supermarket and you are looking at roughly 30,000 products. About 25,000 of those did not exist five years ago. And of the products launched this year, somewhere between 70% and 85% will be gone within twelve months, depending on whose statistics you trust and how charitably you define "gone."
This is not a new observation. Nielsen has been publishing variants of the same figure for decades. What is new, or at least newly conspicuous, is the gap between what consumer packaged goods companies spend on research and how reliably that research prevents failure. The CPG industry invested an estimated $8.5 billion globally on consumer research in 2024. It also killed more products than it launched. Something in the methodology is not working.
The conventional explanation is that consumer preferences are simply unpredictable. Tastes shift, trends move, people say one thing in a focus group and do another in the aisle. There is truth in that. But the more uncomfortable explanation, the one that the industry is slowly beginning to confront, is that the research itself has become too slow, too expensive, and too removed from the decisions it is supposed to inform. By the time a traditional study delivers its conclusions, the product has already been greenlit, the packaging has been printed, and the slotting fees have been paid. The research arrives as a post-mortem, not a diagnosis.
Synthetic AI research is starting to change this. Not by replacing human consumers, but by compressing the feedback loop from months to minutes, turning research from a capital expenditure into an operational reflex. This article examines how that shift is playing out across the CPG landscape and what it means for brands that have historically been priced out of serious consumer insight.
What CPG Brands Actually Need to Test
Before discussing the technology, it is worth being specific about what consumer packaged goods companies need to know and when they need to know it. The research questions in CPG are not abstract. They are brutally concrete.
Product concept validation. Before committing to formulation, does the concept resonate with the target consumer? Will someone who drinks three cans of Coca-Cola a day be interested in a nitrogen-infused variant? (The answer, as PepsiCo discovered with Nitro Pepsi, was a decisive no. Cola drinkers want sharp fizz, not the creamy softness that works in a stout.)
Packaging and design testing. Does the packaging communicate the right things on the shelf? Does it stand out in the correct way? Does the colour palette signal premium, value, organic, or functional? These are not aesthetic questions. They are purchase-intent questions. A consumer has roughly 1.5 seconds of attention per product on a shelf. The packaging either earns a pick-up or it does not.
Pricing elasticity. What happens to purchase intent when the price moves from $4.99 to $5.49? What about $5.99? Is there a psychological threshold beyond which consumers switch to a competitor or simply walk past? These curves differ by category, by brand, by geography, and by season. Testing them traditionally requires conjoint analysis or Van Westendorp studies that can take weeks and cost five figures.
Shelf and planogram optimisation. Retailers and brands share a common interest in understanding how products perform in different shelf positions, adjacencies, and configurations. Which products benefit from being placed next to which competitors? Does moving from eye level to the bottom shelf reduce purchase intent by 15% or 40%? Planogram decisions are worth billions in aggregate and are still made, in many organisations, by a combination of slotting fees and intuition.
Brand tracking and competitive positioning. How does the brand's perception change over time? What associations does the consumer hold? How does it compare to direct competitors and to the own-label alternative sitting two inches to its left? Traditional brand trackers are quarterly affairs. The brand can shift meaningfully between waves.
Messaging and claims testing. Does "30% less sugar" perform better than "naturally sweetened"? Does "family-owned since 1987" increase trust more than "made with real ingredients"? Every word on the package, in the advertisement, and on the product page is a hypothesis about what consumers care about. Most of those hypotheses go untested.
The common thread across all of these is that the questions are specific, the decisions are frequent, and the traditional research timeline does not match the pace at which those decisions must be made.
The Economics of Traditional CPG Research
The CPG industry's relationship with consumer research has always been somewhat paradoxical. The companies that can most afford research are the ones that least need persuading of its value. And the companies that most need insight, the mid-size brands launching into competitive categories with thin margins, are precisely the ones least able to pay for it.
A standard concept test through a traditional research agency costs between $15,000 and $50,000, depending on the methodology, sample size, and geographic scope. A full packaging study with shelf simulation can exceed $75,000. Brand tracking programmes run into six figures annually. Conjoint analysis for pricing optimisation typically starts at $30,000 per study.
The timeline is equally prohibitive. A well-designed quantitative study takes four to six weeks from brief to report. Qualitative research, including focus groups with recruitment, facility hire, moderation, and analysis, takes six to eight weeks. Even the faster "agile" research platforms that emerged in the 2010s still require one to two weeks for a basic concept test.
These economics create a predictable pattern. Large multinationals like Procter & Gamble, Unilever, and Nestle maintain internal insights teams and standing relationships with research agencies. They test extensively, though even they cannot test everything. Mid-size brands test selectively, reserving research budget for the highest-stakes decisions and hoping that instinct covers the rest. Small and emerging brands test almost nothing, relying on founder intuition, social media feedback, and the expensive education of post-launch failure.
The result is a two-tier system. The brands with the most resources get the most insight. The brands with the least resources, and therefore the highest failure rates, get the least. This is not a market functioning well. It is a market structured to perpetuate the advantage of incumbents.
How AI Is Restructuring the Research Stack
The application of artificial intelligence to consumer research is not a single technology but a family of approaches, each addressing different parts of the research process.
At the enterprise end, platforms like Simile (formerly known as Synthetic Users) have attracted attention from major retailers. CVS Health was reported by Bloomberg to be using Simile's synthetic consumer technology to test retail concepts and product placements. The appeal for a retailer operating nearly 9,000 stores is self-evident: testing a planogram change across even a small subset of physical locations is extraordinarily expensive. Testing it with synthetic consumers costs a fraction and delivers results before a single product is moved.
The Simile approach is instructive because it validates the fundamental premise: synthetic consumers, when properly constructed and calibrated, produce research outputs that correlate meaningfully with real-world consumer behaviour. The question is no longer whether synthetic research works. It is whether it can be made accessible to the brands that need it most.
This is where platforms like FishDog enter the picture. Where Simile targets enterprise accounts with bespoke implementations, FishDog has built a self-serve platform with over 300,000 synthetic personas grounded in census data and behavioural research. The practical difference is one of access. A brand manager at a $50 million revenue CPG company can run a concept test, a packaging study, or a pricing analysis in FishDog in the time it takes to drink a coffee, at a cost that would not register on a quarterly research budget.
The technology underpinning these platforms differs in implementation but shares a common architecture. Large language models are fine-tuned on demographic, psychographic, and behavioural data to create personas that respond to research stimuli as real consumers would. The personas are not random text generators. They are statistically constructed representations of specific consumer segments, complete with the biases, preferences, contradictions, and category-specific knowledge that real consumers carry.
The calibration question is the one that sceptics rightly raise, and it is the one that the data increasingly answers. FishDog's census-grounded approach means that when you recruit a panel of ten synthetic consumers matching a specific demographic profile, the distribution of their responses mirrors what you would expect from ten real consumers with those characteristics. Not perfectly, because nothing in research is perfect, but with sufficient fidelity to inform decisions that would otherwise be made on instinct alone.
FishDog in Practice: CPG Use Cases
The abstract case for synthetic research is straightforward. The concrete case is more interesting. Here is how CPG brands are using FishDog across the research questions outlined above.
Product Concept Testing
A mid-size US beverage company was considering launching a functional soda line into a category already occupied by Olipop, Poppi, and a growing roster of competitors. The question was not whether functional sodas were a viable category. They plainly were. The question was whether a new entrant could differentiate sufficiently to earn shelf space and repeat purchase.
In FishDog, the brand created a research group of ten personas matching their target demographic: health-conscious adults aged 25 to 44 in the United States. Seven questions explored reactions to the brand concept, flavour expectations, packaging preferences, and competitive switching barriers. The study ran in under an hour.
The findings redirected the entire launch strategy. Personas consistently indicated that taste was the non-negotiable criterion, that health claims without credible substantiation actively reduced trust, and that the brand's planned positioning as "the clean alternative" was indistinguishable from four existing competitors. The brand pivoted to a flavour-forward positioning with a specific ingredient story. The research cost less than a team lunch. The traditional alternative would have cost $25,000 and taken six weeks, by which time the launch window would have narrowed.
Packaging and Shelf Presence
A Canadian snack brand was redesigning its packaging and had narrowed the options to three concepts. Traditional shelf testing would have required physical mock-ups, a simulated retail environment, and recruited participants. Instead, the brand described each packaging concept to a FishDog panel, including colour palette, typography style, imagery, and on-pack claims, and asked which concept communicated the brand's values most clearly, which stood out in a competitive set, and which felt most premium.
The synthetic panel identified something the design team had missed. Two of the three concepts used a matte finish with earthy tones that, to the personas, read as "expensive but boring." The third used a brighter palette that the design team considered less sophisticated but that consumers associated with flavour intensity and energy. The winning concept was the one the creative director liked least. This is, of course, exactly why you test.
Pricing Sensitivity
Pricing in CPG is not a single number. It is a curve, and the shape of that curve differs by product, category, competitive context, and consumer segment. A British artisan condiment brand was expanding into US retail and needed to understand American price expectations for premium table sauces, a category with fundamentally different price architecture than the UK market.
A FishDog study with US-based personas explored willingness to pay at multiple price points, perceived value relative to competitors, and the specific claims that justified a premium. The research surfaced a price ceiling ($8.99 for a 250ml bottle) beyond which purchase intent collapsed regardless of quality perception. It also revealed that "imported from Britain" carried minimal premium value with American consumers, contrary to the brand's assumption. "Small-batch" and "family recipe" performed significantly better as premium justifiers.
Messaging and Claims
Every CPG product makes claims. Every claim is a hypothesis. FishDog allows brands to test those hypotheses at a speed and cost that makes it practical to test every claim, not just the ones deemed important enough to warrant a formal study.
A protein bar brand tested six different front-of-pack claims with a synthetic panel:
"20g protein per bar"
"Clean ingredients you can pronounce"
"Gym-tested, athlete-approved"
"No artificial sweeteners"
"Tastes like a real chocolate bar"
"Fuel your day, not your guilt"
The panel's responses revealed a clear hierarchy. "Tastes like a real chocolate bar" generated the most enthusiasm because it addressed the category's core scepticism (protein bars taste bad). "20g protein per bar" was noted but not differentiating. "Gym-tested, athlete-approved" was actively off-putting to the brand's target segment of casual fitness consumers who did not identify as athletes.
These are not earth-shattering revelations individually. But collectively, across dozens of such decisions per product per year, they compound into the difference between a product that resonates and one that merely exists.
The Failure Prevention Argument
The most compelling case for synthetic research in CPG is not efficiency, though the efficiency gains are real. It is failure prevention. And the failures in CPG are neither rare nor cheap.
We catalogued more than 150 documented CPG product launch failures between 2021 and 2026, ranging from Coca-Cola Spiced (pulled after six months because the name created flavour expectations the product did not meet) to Starbucks Oleato (olive oil coffee, conceived in an executive echo chamber and killed by consumer revulsion) to Hershey's Reese's Plant-Based Cups (technically adequate, sensorily unacceptable because the "goo factor" was absent).
In every case, the failure was preventable. Not with better formulation or better marketing, but with better research. The questions that would have identified the fatal flaw were obvious in retrospect and answerable in advance. "Does the name 'Spiced' make you expect heat?" Yes. "Would you put olive oil in your coffee?" No. "If the plant-based cup has a firmer, drier chocolate, is that acceptable?" Absolutely not.
What synthetic research changes is not the quality of the questions but the economics of asking them. When a concept test costs $25,000 and takes six weeks, you test the big bets and hope the smaller ones work out. When a concept test costs a few hundred dollars and takes an hour, you test everything. You test the name. You test the packaging. You test the price points. You test the claims. You test the adjacencies. You test whether "Spiced" sounds hot.
The CPG industry does not have a knowledge problem. It has an access problem. The insights that prevent failures exist in the minds of consumers. Traditional research makes those insights expensive and slow to extract. Synthetic research makes them cheap and fast. The maths is not complicated.
Who This Is For (And Who It Is Not)
Synthetic research is not a replacement for all traditional consumer research. It is a replacement for the research that is not happening.
Procter & Gamble will continue to run large-scale quantitative studies, ethnographic programmes, and longitudinal brand trackers. They should. They can afford to, and the decisions they are informing justify the investment. Synthetic research for P&G is an accelerant, a way to pre-test concepts before committing to a full study, to screen twenty ideas down to three before spending serious money on validation.
The transformative impact is further down the market. The $30 million CPG brand that currently runs two consumer studies per year because that is all the budget allows. The emerging brand that has never run a formal study because the minimum viable research project costs more than the quarterly marketing budget. The product manager who tests three concepts instead of twelve because testing twelve would blow the annual research allocation.
These are the companies where synthetic research changes the calculus most dramatically. Not from good research to better research, but from no research to some research. From instinct to evidence. From hoping the packaging works to knowing which version performs best before the print run starts.
The analogy is to what cloud computing did for startups in the 2000s. AWS did not replace enterprise data centres. It gave companies that could never have afforded a data centre the ability to operate at scale. Synthetic research platforms like FishDog are doing the same for consumer insight: making world-class research methodology available to brands that could never have accessed it through traditional channels.
The Quality Question
The reasonable objection to synthetic research is the quality question. Can artificial personas really predict what real consumers will do?
The honest answer is: not perfectly, but better than the alternative. And the alternative, for the vast majority of CPG decisions, is no research at all.
This does not mean every individual response from a synthetic persona matches what a specific human would say. It means the aggregate patterns, the segments, the preference hierarchies, the dealbreaker identification, track closely with what traditional methods produce. The signal is real.
There are categories where synthetic research is particularly strong. Concept testing, packaging evaluation, messaging hierarchy, pricing sensitivity, and competitive positioning all produce reliable synthetic outputs because they depend on stated preferences and reasoned reactions, areas where well-constructed personas excel. There are categories where it is weaker. Sensory evaluation, where the physical experience of tasting, touching, or smelling a product is central, remains better served by real-world testing. You cannot ask a synthetic persona whether the mouthfeel is right. You can ask whether the concept appeals, whether the price feels justified, and whether the packaging communicates quality, and those answers will be trustworthy.
The mature approach is to use synthetic research for the 80% of decisions where stated preference data is sufficient and reserve traditional methods for the 20% where physical interaction matters. This is not a compromise. It is an intelligent allocation of research resources.
Where This Goes Next
The trajectory of AI research in CPG is toward integration, not isolation. The platforms that will define the next five years are not the ones that produce the best synthetic data in a vacuum but the ones that embed themselves into the decision-making workflow.
FishDog's approach of delivering results in minutes rather than weeks is significant not because speed is inherently valuable but because it changes when research happens. When a study takes six weeks, it happens at the beginning of a project, if it happens at all. When a study takes an hour, it happens at every decision point. Before naming. Before packaging. Before pricing. Before the campaign brief. Before the shelf reset. Research becomes iterative rather than monolithic, a continuous input rather than a one-time gate.
For retail and CPG, this is the structural shift that matters. Not AI replacing researchers, but AI making research so fast and so accessible that it becomes unremarkable, as routine as checking the weather before leaving the house. The brands that adopt this reflex will, on average, make fewer preventable mistakes. The ones that do not will continue to learn the same lessons the expensive way.
The shelf will remain a graveyard. But the products buried in it need not include yours.
Disclosure: the author is co-founder of [FishDog](https://fish.dog), which is discussed in this article. Simile is referenced as a peer platform based on publicly reported information, including Bloomberg's coverage of CVS Health's use of the technology. Observations about traditional research costs are drawn from published industry benchmarks and the author's direct experience in the market research sector.
Phillip Gales is co-founder of [FishDog](https://fish.dog). He writes about synthetic research, consumer insight, and the occasionally baffling decisions of large corporations.