Use synthetic personas to surface disagreement in your innovation workshop

Choose one opportunity and one uncertainty. “Explore our innovation strategy” is too broad. “What could prevent a service team from sharing scheduling information with a customer?” gives participants something concrete to examine.

Keep the offering and situation consistent while exploring different roles. Define the geography, industry and company size, then clarify each role’s responsibilities. Participants should be able to explain why that perspective matters to the learning question.

The workshop team and the modeled customers are different groups. Your colleagues interpret the output and make decisions; the synthetic personas provide simulated perspectives to interrogate.

Ask participants to write their assumptions independently before the group discusses them. This makes it easier to see where the team already agrees or disagrees, rather than allowing the first AI answer to set the agenda.

For each assumption, record:

• What the team currently believes about the customer situation.
• What evidence, if any, supports that belief.
• What would be different if the belief were wrong.

Select assumptions that could change the next investigation. A workshop does not need to resolve every uncertainty. It needs to identify which ones deserve attention before the team commits further effort.

Use a consistent core question in each role conversation. Ask about responsibilities, dependencies and the consequences of a proposed change. Follow up when a response suggests an alternative explanation or an overlooked constraint.

Bring selected passages back to the human workshop and label them as simulated. The facilitator can place different explanations side by side and ask which assumptions each one depends on.

This does not require a live meeting between AI personas or an automated consensus feature. Your team runs the conversations and makes the comparison. Agreement between simulated responses is not independent evidence, and repetition is not a measure of how common an opinion is among actual customers.

Illustrative scenario, not a customer result.

An innovation team is exploring a scheduling service for mid-sized industrial service providers in the Netherlands. The initial idea lets a provider and its customer see the same planned maintenance visits.

Participants assume that shared visibility will reduce coordination work. They explore an operations lead at the provider and a coordinator at a customer site, keeping the service concept and operating situation consistent.

The facilitator introduces two possible explanations for the group to examine. Perhaps coordination is difficult because information is scattered. Or perhaps the information is available, but each organization uses a different definition of a confirmed appointment.

The team asks questions that explore both possibilities. It does not report either explanation as a discovery about real customers. Its next research task is to walk through a recently rescheduled visit with people on both sides and compare the records, decisions and handoffs.

End each useful discussion with a short ticket written by the team:

1. Uncertainty: The question that remains unresolved.
2. Competing explanations: At least two plausible accounts of what is happening.
3. Customer context: Who could help investigate it and in which situation.
4. Evidence to seek: A recent example, observation, record or experiment.
5. Decision affected: What the team would reconsider based on the finding.

Assign an owner for the next investigation. If the group cannot identify a decision the ticket could affect, refine the question or set it aside. Do not convert simulated enthusiasm into an opportunity score.

Save the assumptions, relevant context and questions alongside the team’s interpretation. Mark real research separately from simulated material so that someone reading the notes later can see where each conclusion came from.

“Synthetic persona” is a method label, not a guarantee of fidelity. If a proposal uses “digital twin persona,” ask what actual person or population is modeled, what inputs and updates support it, and how its intended use is evaluated. That label should not be applied to Virtual Customer as a claim that it copies a named individual.

A successful session leaves the team with clearer uncertainties and an actionable research plan. Whether its hypotheses describe actual customer work still needs to be established outside the workshop.