A digital twin of a consumer is a computational model that simulates how a real person would respond to new products, prices, or marketing messages. The accuracy of a digital twin depends entirely on the accuracy of the data used to build it — and different approaches use fundamentally different data.
Why Digital Twins Matter Now
Consumer research has always relied on asking people what they think, what they prefer, and what they would do. Surveys, focus groups, and interviews are essential — they capture perception, emotion, and reasoning in ways that behavioral data alone cannot.
But there is a well-documented gap between what consumers say and what they actually do.
When researchers compared stated purchase intent to real willingness to pay, 83% of consumers said they would buy a product, but only 42% followed through when real money was involved.1 Broader analyses estimate that stated preference predicts actual purchase behavior with roughly 34% accuracy.2 This does not mean surveys are unreliable. It means they measure something different from behavior. Surveys capture how people think about decisions. Purchase data captures what they actually decided.
At the same time, the infrastructure underlying traditional research is under pressure. An analysis of 4.1 billion survey attempts found that 33% were fraudulent.3 AI-generated fraudulent responses have risen 43% year over year.4 Researchers now discard an average of 38% of collected data before analysis begins.5
Despite billions spent on consumer research, 85% of new CPG products fail within two years of launch — products that passed concept tests, scored well in purchase intent studies, and cleared internal stage-gate processes.6
The market for digital twins and simulation-based research is growing from $1.8 billion to a projected $8.2 billion by 2029.7 The appeal is clear: a digital twin can be queried instantly, does not fatigue, and does not need to be recruited. The question is what data these twins should be built on.
How Digital Twins Are Built Today
Most digital twins of consumers are built from one of three data foundations.
Surveys and Interviews
The most common approach trains digital twins on stated preferences — survey responses, interview transcripts, or extended conversations with AI moderators. The resulting twins replicate how individuals describe their reasoning, preferences, and intentions. This approach is well suited for questions about perception, brand sentiment, and messaging — situations where understanding how people think is the goal.
The limitation is that these models inherit the say-do gap. They learn what people report about their behavior, which research consistently shows diverges from what people actually do.
Demographic and Population Data
A second approach builds twins from census records, socioeconomic data, and browsing patterns. These models can simulate population-level trends quickly — market sizing, directional research, trend forecasting — without fielding a study.
The limitation is that these are statistical composites, not individual behavioral records. The twins represent how a person like this might behave, based on demographic probability, rather than how a specific person actually behaved.
Verified Purchase Data
A third approach builds twins from actual transaction records — what a consumer bought, where, when, at what price, and in what combination with other products. Unlike generic models that estimate how an average person might behave, these twins reflect specific, real people and their demonstrated choices. The preferences are derived from revealed behavior — decisions backed by real money — rather than stated intent.
This approach is strongest for behavioral prediction: pricing sensitivity, brand switching, competitive share of wallet, and category purchasing. It captures patterns that consumers may not consciously recognize or accurately report — like gradually shifting spend toward a competitor over six months, or consistently buying premium in one category and value in another.
Why the Data Foundation Matters
Each of these approaches produces useful results. The question is whether the training data matches the question being asked.
For attitudinal questions — how do consumers feel about this brand? what language resonates? — traditional survey or AI twins built on top of survey data is the right input. The say-do gap is not a problem when the research question is about perception.
For behavioral questions — will consumers actually buy this product? at what price? from which competitor? — transaction data provides a more direct signal.
Research by 84.51°, Kroger’s data science division, illustrates why this distinction matters. In a study comparing self-reported category behavior to actual purchase records, 60% of consumers who identified themselves as buyers of a category were placed in the wrong behavioral segment. And 75% had never actually purchased in the categories they claimed to buy.8
These were not dishonest respondents. They were ordinary people doing what people do — misremembering, rounding up, conflating intention with action. This is precisely the kind of data that survey-trained digital twins learn from. A twin that accurately replicates these responses is accurately replicating the gap between reported and actual behavior.
A digital twin built on verified purchase records does not have this problem. It learns from what happened, not from what was reported.
The accuracy of a digital twin is bounded by the accuracy of its training data. There is no algorithmic shortcut around this constraint.
What Transaction-Grounded Digital Twins Require
Building digital twins from purchase data requires solving a data infrastructure problem first. The twin is only as complete as the behavioral record underneath it.
Consent-based collection. Consumers share their purchase data in exchange for tangible value — cashback, spending insights, loyalty benefits. This is first-party data with explicit consent.
Cross-retailer visibility. Most data sources see one retailer. A loyalty program knows what a shopper buys at that chain but nothing about their purchases everywhere else. Meaningful behavioral modeling requires visibility across the consumer’s full shopping footprint.
SKU-level granularity. Category-level data is too coarse for product-level decisions. Transaction-grounded twins are built on individual item records — distinguishing organic whole milk from conventional 2%, a full-price purchase from a promoted one.
Use Cases
Pricing research. Traditional willingness-to-pay studies ask consumers to state their price threshold. Transaction-grounded twins model revealed price sensitivity — how consumers actually responded to price changes across retailers and over time.
Competitive intelligence. Purchase data reveals the actual competitive set: what the consumer bought before, what they switched to, and what they bought alongside it. This often surfaces competitors that brand teams did not know they were losing share to.
Product launch simulation. Transaction-grounded twins can model adoption likelihood against what consumers in a category actually buy today — adding a behavioral check to attitudinal concept testing.
Marketing optimization. Testing messages against confirmed category buyers, validated through purchase data, ensures the test audience reflects real behavior rather than self-reported category membership.
Frequently Asked Questions
What is a digital twin of a consumer?
A digital twin of a consumer is a computational model calibrated to an individual person’s preferences, attitudes, or behaviors. It can be queried to predict how that person would respond to a new product, price, or experience. The fidelity of the twin depends on the data used to build it: some are trained on survey responses and interviews, others on demographic patterns, and others on verified purchase records.
How is a digital twin different from a customer persona?
A persona is a composite archetype that represents a segment average — useful for aligning teams around a shared picture of the customer. A digital twin represents a specific individual and can generate predictions about that individual’s likely behavior in untested scenarios. Both are useful; they serve different purposes.
Are digital twins accurate?
Digital twins’ accuracy depends on the data the twin is trained on and what type of question is being asked.
For factual questions — what brands has this person purchased in the past three years? At what price points were those items purchased? — digital twins built on direct purchase data are significantly more reliable than self-reported answers. The data comes from verified transaction records, not recall. And because consumers connect their actual accounts electronically, the fraud rate is effectively zero. There is no way to fabricate a purchase history when the data is pulled directly from a real retailer account.
For lifestyle and preference questions, twins grounded in purchase data can generate surprisingly strong inferences. The depth of detail in a consumer’s purchase history — across categories, brands, price tiers, and timing patterns — reveals preferences, priorities, and life circumstances that are difficult to capture completely in a survey. Purchase patterns surface health priorities, household composition, brand loyalty, price sensitivity, and category exploration without asking a single question.
What is the say-do gap?
The say-do gap is the divergence between what consumers report in research settings and what they do in the real world. Studies estimate that stated purchase intent predicts actual behavior with roughly 34% accuracy. This does not invalidate surveys — surveys measure perception, which is the right input for many research questions. But for predicting purchasing behavior, twins grounded in purchase data provide a more direct signal.
Should digital twins replace surveys?
No. Digital twins and surveys answer different questions. Twins built on behavioral data are strongest at predicting what consumers will do — what they will buy, at what price, from which competitor. Surveys and qualitative research are essential for understanding why consumers make choices, how they perceive brands, etc. The most effective research programs use both.
How does consent-based data collection work?
Consumers explicitly agree to share their purchase data, typically in exchange for clear value — cashback, spending analytics, or loyalty benefits. The consumer initiates the share, understands what is being collected, and can revoke access. This is first-party data with documented consent, distinct from inferred, scraped, or third-party data. Learn more →
Sources
- Veylinx, “How Big is the Gap Between What Consumers Say and Do”
- CloudArmy / Gartner CEB, “Why Stated Preferences Fail”
- Greenbook, “State of Survey Fraud 2025” (4.1B attempts analyzed)
- PMC, “AI-powered fraud and online survey integrity”
- Research Shield / Kantar, “Real Impact of Survey Fraud”
- Nielsen via Food Navigator, “Why do 85% of new CPG products fail within two years?”
- Fish.Dog, “Synthetic Research Platforms: The 2026 Market Map”
- 84.51° (Kroger), “Behaviorally-Verified Sampling vs. Self-Claimed Sampling”
Questions? enterprise@ariodata.com