Decision Memo: Synthetic Users Are Research. Synthetic Customers Are a Different Standard.
Tuesday’s article made the case that synthetic user tools are useful and synthetic customers are the problem: the real leadership question is not whether to run AI-generated simulations, but whether the team knows what those simulations are and are not allowed to prove.
The classification problem most teams are skipping
Here is what most teams are getting wrong. Synthetic user tools do not just make research faster. They make weak validation look complete.
A team that used to lack customer evidence now produces an AI-generated persona panel, a synthetic buying journey, a simulated price test. The output looks thorough and the confidence looks earned. Nothing has actually changed.
The false signal is analytical output. The real signal is changed exposure. The right question is not whether the team produced a compelling synthetic user analysis, but whether that analysis changes what you are willing to risk.
That is the decision this memo is about.
The decision is not whether to use synthetic user tools
It should be used. That question is settled.
The real question is whether synthetic user output lowers or raises the standard for real-world evidence.
Most teams will be tempted to let it lower the standard. Faster, cheaper research makes the next step feel less dangerous: a small budget extension, a go-to-market plan that treats synthetic preference as demand, a pricing decision that treats simulated responses as willingness to pay. That is how commitment happens without a decision.
A 2001 meta-analysis of 29 experiments found that participants in hypothetical settings overstate their preferences by a factor of roughly three. Language model outputs introduce a second layer of separation from actual behavior. A 2024 study in Political Analysis found that LLM-generated responses compress variance and diverge from human benchmarks in ways that make them unreliable for statistical inference. AI does not remove the preference-behavior gap. It makes the gap harder to see.
Preference is cheap to simulate. Commitment still creates exposure.
What leaders usually misread
The most common mistake is treating synthetic output as a weaker version of real evidence, rather than a different category of evidence entirely.
It is an understandable mistake. A well-run synthetic user session surfaces objections, tests messaging, and produces responses that sound like real buyers. The output has the texture of discovery.
But a synthetic user output is a claim about what someone like this might say.
A funding or go-to-market decision is something different. It says: we are willing to assign resources, foreclose alternatives, and expose ourselves to the consequences of being wrong about this market.
A synthetic user can tell you what a person like this might say. It cannot tell you whether they will pay, return, recommend, or survive a B2B procurement process.
That distinction should be explicit before any output from a synthetic user session enters a decision. A synthetic output can say a segment finds the offer appealing. A decision asks whether that appeal converts into payment under real conditions, with real alternatives, in a real buying system.
In B2B, the buying system is the critical gap. Gartner puts the average buying group at 6 to 10 decision-makers for complex solutions. Forrester’s 2024 State of Business Buying report puts it at 13 stakeholders, with 89% of decisions crossing multiple departments. A synthetic user addresses one node. The deal survives or dies in the other twelve.
The better lens: assumption stress tests, not demand signals
The useful reframe is not asking what the synthetic output says about the market. It is asking which assumption the synthetic output was designed to test, and what real experiment that test should now trigger.
That shift changes the purpose of the tool entirely. Synthetic users are not a faster path to market evidence. They are a cheaper path to a sharper real experiment.
The decision is not whether AI personas are useful. The decision is whether your team knows what they are and are not allowed to prove.
The stronger use of synthetic user research is not generating more market insight. It is reducing the cost of reaching a cleaner commitment decision by identifying which assumption deserves the next real test, and by naming which assumptions cannot be tested synthetically at all.
The synthetic user funding rule
Here is the rule I would use:
Synthetic user research may lower the cost of mapping assumptions, but it should raise the standard for what counts as market evidence.
If a team can now generate customer hypotheses, segment responses, adoption scenarios, and objection maps faster than before, leadership should expect sharper gates before the next budget is approved. A team should not receive the next commitment because it produced a compelling synthetic user analysis. It should receive the next commitment because it reduced a specific uncertainty through real-world contact with the market.
Before synthetic user tools, teams could argue they needed more time to reach real customers. Sometimes that was true. Mostly it was a way to avoid a hard test.
With synthetic tools, that argument disappears earlier.
If the output is not attached to a specific assumption, and the assumption is not attached to a specific real test, the team is not doing research.
It is producing comfortable material.
