The question you thought of on the drive home
Every researcher has had this experience. The session ends, the participants collect their incentives and go, and somewhere between the venue and home the question you should have asked arrives fully formed. It is always a good question. It is always too late.
The gap between what a study answered and what you later needed it to answer is the ordinary condition of customer research. A study gets commissioned against the questions that feel urgent in one particular quarter. It is fielded, analyzed, presented, and filed. Then the quarter ends and the business keeps moving. A competitor launches something nobody modeled. Procurement changes a supplier and the formulation shifts. Finance proposes a price increase. A new marketing lead arrives and wants to know why the brand is positioned the way it is. Every one of those moments raises a question your research could have answered while it was still running.
By then the participants have gone home for good. Reaching them again means recruiting a new sample, which means a new budget, a new fielding cycle, and a new set of people who are not the people you originally learned from.
MNAI Digital Panel exists to close that gap. It takes customer research you have already completed and turns it into a panel that stays available.
What is actually sitting in your archive
When people picture their research archive, they picture the deliverable. The findings deck, the topline summary, the slide that went to the board. That is the part that circulates internally, and it is the least useful part of what was produced.
Underneath the deck is participant-level material. Recruitment screeners describing who each person was and why they qualified. Profiles capturing category behavior, purchase history, attitudes, and household context. Transcripts in which individual people explained their reasoning at length, in their own words, with the hesitations and self-contradictions left in. Survey responses tied to individuals instead of pooled into percentages.
That material describes specific human beings in considerable detail. Everything the findings deck compressed into an average was known at the time about particular people. Compression made the deck presentable. It also made the research stop being able to answer anything new.
What Digital Panel does
Digital Panel reads the participant-level material and reconstructs each participant as a Digital Twin, a model of one specific person that can respond to new questions the way that person plausibly would. Applied across a whole study, this produces a panel. The forty people who sat in your sessions in 2022 become forty models you can put a new question to today.
The panel persists. It does not disband when a project closes. When the pricing question arrives eight months later, the same panel is there, and the twins are the same individuals with the same characteristics they had before.
They answer one at a time, as individuals. Ask the panel whether a proposed subscription change is acceptable, and no verdict comes back. Forty separate answers come back, each carrying the reasoning behind it, and you examine them the way you would examine any qualitative dataset. Where does the distribution cluster? Which segments split? Who objects, and on what grounds? Which objection appears in eleven different responses phrased eleven different ways? Whether the agreement you are seeing is real or whether one loud group is masking a minority position that matters commercially.
The output is the shape of a room full of people who disagree with each other in specific and traceable ways.
Why you should not simply believe that this works
Every vendor selling simulated customers claims their simulation is accurate. The claim is cheap to make and difficult for a buyer to check, and the research profession is right to be skeptical about it.
Two things make this one checkable.
The first is published. The Mnemonic Digital Twin was tested against 2,058 real people who had each answered a very large battery of questions. Twins were constructed from part of what each person had said, then asked the questions that had been withheld, and their answers were compared with what those people actually answered. The twins reached 89% of the human accuracy ceiling. That ceiling is the level of agreement a person reaches with their own earlier answers when asked the same thing again, which is well short of perfect, because people are not perfectly consistent with themselves. Measuring against the human ceiling instead of against perfection is the honest way to state the result. The methodology, the sample, and the comparison to previously published work are all available to read.
The second is specific to you, and it matters more. Accuracy on a public benchmark tells you how well the underlying method models one person. It does not tell you how faithfully a reconstructed version of your customers will behave. That depends on who your participants were, how your sessions were moderated, what category you are in, and how much of the original material still exists. No vendor can give you that number in advance, and any vendor who offers one is guessing.
What can be done is to measure it on your own study, because you are in the unusual position of already knowing the answers. Your archive contains what your participants actually said. So the twins are built from the earlier material, a portion of the original responses is held back, the panel is asked those same questions, and its output is scored against the real transcript. You receive a fidelity assessment for your specific study before the panel is used to explore anything you do not already know.
Every engagement starts there.
What you can actually do with the Digital Panel
- Ask the question you thought of afterward. The most common use is also the simplest. A study answered what it was scoped to answer. You have since thought of four more things you wish had been on the guide. The panel takes them.
- Test concepts against the people who told you what they wanted. Concept testing usually happens with a fresh sample who arrive with no history. The people in your archive already explained what frustrated them about the current product and what they were willing to trade away. Putting a new concept in front of them means putting it in front of an audience whose expectations you documented.
- Understand what a price change does to the customers who are currently paying. Pricing research is expensive and slow, and it is often skipped for exactly that reason. A validated panel drawn from your own customer base lets you look at a proposed structure, an annual-only tier, a change to a promotional mechanic, or a bundle reconfiguration, and see which individuals accept it, which grumble and stay, and which describe leaving. The reasoning attached to each response usually matters more than the split.
- Pressure-test language before it is expensive to change. Claims, campaign lines, packaging copy, and onboarding emails can go to the panel while they are still cheap to revise. What you are looking for is the specific misreading. One participant hearing a sustainability claim as an admission that the previous formula was worse is worth more than a favorability score.
- Look at an experience change from inside it. Onboarding flows, returns policies, service tiers, and subscription management all tend to be evaluated by the people who designed them. A panel that includes the customers who already complained about your returns process will tell you where the redesigned version loses them.
- Decide what deserves new fieldwork. Digital Panel is often most valuable as a filter. When nine concepts are competing for a research budget that covers two, running all nine past a validated panel gives you a defensible basis for choosing which two go in front of real people. The cost of the fieldwork does not change. What changes is the quality of what you spend it on.
- Give new people a way to interrogate the past. Institutional memory in most insights functions lives in the heads of people who eventually leave. A new brand manager can read a five-year-old findings deck and learn conclusions. With a panel they can ask the underlying customers why, and follow up, which is a different kind of learning entirely.
- Reopen a decision that has since been questioned. When a launch underperforms, the postmortem usually consists of people arguing from memory about what the research said. Going back to the panel and asking it directly about the thing that went wrong produces a more useful conversation than the argument.
Where Digital Panel stops
There are questions it cannot help with, and being clear about them is part of the method.
It only models people you have already studied. If you need to understand a market you have never researched, a demographic you have never recruited, or a country you have never entered, there is nothing in the archive to build from. That work still requires human fieldwork.
It models your customers specifically. It is not a general simulator of public opinion, and it should not be used as one.
It depends on what survived. Studies that exist only as a findings deck can seed a panel, but with no participant-level record to score against, the validation step has nothing to work with. We establish which case you are in before anything is committed.
And it does not remove the need for human research when a decision is irreversible and expensive. What it changes is the quality of the questions you bring to that research and the number of options you still have to test by the time you get there.
Where to start with the Digital Panel
An engagement begins with an assessment of your archive rather than a purchase. You tell us which study, what year it ran, and what material still exists. We establish whether it supports validation and what a pilot would look like. If the archive cannot support a fidelity assessment, we say so.
Most companies are sitting on years of qualitative research that stopped being useful the day it was presented. The people are still in there, described in more detail than anyone remembers.
