A shiny new tool for your insights toolkit
Qual and quant take weeks to complete, and budgets are getting cut back. That’s why we built an AI-driven “synthetic panel” or “digital twin.” You get fast learnings for five core metrics, helping you prioritize ideas and focus resources where you’ll get the most impact.
And it costs under $1000.
Shopper graph foundation
What exactly is a synthetic panel?
We trained an AI model to make choices like human shoppers would so it will fill out your survey just like an online consumer panel. We did this by training our AI with the records of millions of real-world decisions by people —basket receipts, loyalty data, recipes saved online, surveys, product reviews, and social posts.
The data is all anonymous for privacy reasons. But you can screen by age, income, demographics, responsibility for purchase decisions, and category purchase history just like you would with a real panel.
Why would you use a synthetic panel/digital twin?
First off, this doesn’t replace primary research. We are big believers in both qual and quant; qual to probe unmet needs deeply and quant to provide rigor and confidence.
This is for quick, inexpensive learning where teams ordinarily would just make guesses. Be the hero who provides the organization real data, fast, to replace guesswork:
- Changes made to claims language after it went through extensive testing. Instead of hoping for the best, A/B testing with a digital twin tells you whether the revision performs as well as the original.
- Marketing wants to test digital ads online. Running them through the synthetic panel first saves time and money because you’ll identify the best ones to rotate into the online test.
- Choosing among claims for the package front panel. A quick Max-Diff test for all the callouts will pinpoint the one with the most resonance.
- No budget to test ideas at the fuzzy front end of innovation. A quick screen of dozens of early concepts will show you exactly where to focus, for less than $1,000.
Top management suggests a new direction. Test it and tell them the results today, and you’ll always have a seat at the table.
Generating the synthetic panel
Submit your concept, and our system assembles a panel of virtual shoppers (N=500) aligned to the screening criteria of your brief. The survey can have both written text and images for the virtual shopper to read/view and respond to (yes, it can discern all kinds of different images!)
Each one fills out the survey independent of the other virtual shoppers, producing raw scores for each metric.
Scoring and adjustment
You get back a complete report just like with a real consumer panel. With a concept test this includes:
- Purchase Interest: trial measure
- New & Unique: novelty compared to known products
- Solves a Need: problem-fit rating
- Virality: likelihood that users, after trial, would share the product/idea with other people
- Overall Score: weighted average of the five metrics
We compare the first three metrics to norms from our (human) quant results from thousands of products.
Accuracy
We trained the model to make it accurate. We did this by running identical tests with the synthetic panel and a real consumer panel.
At first it got stuff wrong, so then we showed it the errors and it learned what to adjust in order to score things closer to human behavior. This was repeated across dozens of product categories — lather, rinse, repeat. Because it’s a learning model, it learned.
To validate accuracy, we conduct R² (coefficient of determination) analysis by comparing the rank order and distribution of scores between the synthetic panel and the real consumer panel. In multiple studies, we’ve observed R² values consistently in the 0.68–0.73 range, indicating a solid to strong predictive correlation.
This gives product teams confidence that synthetic results reliably mirror consumer preferences — making it a credible, low-risk step before you invest in primary research, or when there’s no budget for traditional learning.
Key Takeaways
- Fast — get learnings today.
- Inexpensive — under $50.
- Cutting edge — digital twins are an important new tool. Start integrating it in your process before the competition does.
- Reliable — you don’t need perfection but you do need accuracy, which is proven.
