Let AI play the sceptic before the CFO does

Run this method in IdeaHone
Creates a space with the entities and scoring criteria of this method, ready for your next session.
The CFO's first question in the decision meeting is always the one nobody prepared for. This method fills the hour between the workshop and the decision: an AI assistant attacks each shortlisted concept from three roles, the team records the objections as concerns and risks, and answers them before the sponsor asks.
When to use it
Use it after the shortlist exists and before the decision workshop. You need three to five concepts that fit in two sentences, plus the facts the team already knows about each; without facts you get generic objections you could have guessed yourself. Plan an hour for four concepts.
Why it works, and where it stops
A role-played objection is cheap. Nobody has to say "this will never pay back" to the colleague who proposed it, and a concern raised by a machine is easier to accept than one raised by a colleague: no hierarchy, no history, no face to save. People answer the objection instead of defending themselves.
The limits are just as plain. The AI knows nothing about your company unless you tell it, it invents numbers if you let it, and it is exactly as polite or as harsh as you ask. A soft prompt produces soft objections.
How to run it
- Pick the three to five concepts. Fewer and the team defends the one it loves; more and the objections blur.
- Give the AI the concept in two sentences plus the facts you know. What it is, who it is for, rough cost, what exists today, what has been tried before.
- Ask for objections from three roles, one at a time. Finance director, then the customer who tried something similar, then the operations lead. All three at once gives a mixed list nobody owns.
- Decide on each objection in one minute. Someone reads it out loud, the team sorts it: answered (we have the fact), open (becomes a question with an owner), or valid (becomes a risk or an assumption to test).
- Write the answers into the concept, with who answered. "Payback within 14 months, says Procurement" is an answer. "Payback should be fine" is not.
- Repeat for the next concept. Same roles, same order, same minute per objection, so the concepts stay comparable.
- Bring the open list to the sponsor instead of hiding it. Three open questions and two named risks on page one read as diligence. The same five items found in the meeting read as surprise.
The three roles and what they ask
| Role | Typical questions |
|---|---|
| Finance director | What does it cost before the first euro comes back? When is payback, and what if it takes twice as long? What else could the same money do? |
| Customer who tried something similar | Why did I stop using the last thing like this? What would make me switch, and what would make me leave again? Who has to say yes before I can buy? |
| Operations lead | What breaks on day one? Who runs it when the project team is gone? Which process, system or supplier does this quietly depend on? |
Rules
- The AI never scores and never decides. It produces objections, the team produces answers, the sponsor decides.
- Every objection is labelled as AI-generated in the record. Tools like IdeaHone mark AI content as such and keep concerns and risks as items next to the idea; a spreadsheet column works too.
- A human reads every objection out loud. Reading silently lets the team skip the uncomfortable ones.
- An objection that needs a fact gets a fact, not an argument. "Sales has 40 open requests for this" answers. "We think the market is big enough" does not.
Ask each role for the single objection it would consider fatal. One fatal objection teaches more than ten mild ones, and its answer often becomes the first line of the concept.
The AI will happily invent a competitor, a regulation or a cost figure to make its objection sound stronger. Treat every fact it offers as a question to check, never as evidence.
What takes the edge off the objections
- Running it without the facts. A vague sentence gets vague objections. Spend the first ten minutes collecting what the team knows.
- Letting the AI's politeness soften the objections. Most assistants hedge by default. Ask for the objection as the role would say it in a bad mood, and strip the "however" endings.
- Defending instead of recording. The minute is for sorting, not for winning. If someone starts arguing, the facilitator writes "open" and moves on.
- Skipping the step with the sponsor. The polished concept goes up without the open list, the CFO's question still arrives, and the team looks as if it hid something.
- Using it as a substitute for talking to a real customer. The role-played customer is a rehearsal; its valid items are questions for the next real conversation, not answers. Assumption mapping tells you which to test first.
AI that suggests, people who decide
IdeaHone turns what is said into cards, scores ideas against your criteria with written reasons and drafts the summary. Every suggestion can be edited; no decision is made by a machine.
Can AI stress-test a business idea?
Yes, within limits. It is good at raising the objections a cautious person would raise and the questions a team has stopped seeing. It is bad at knowing your facts, so give it the facts and have a human check every number it mentions.
What roles should an AI play to challenge an idea?
The finance director (cost, payback, alternatives for the money), a customer who tried something similar (why they stopped) and the operations lead (what breaks on day one). Ask for one role at a time.
Should AI objections go into the decision record?
Yes, labelled as AI-generated and with the team's answer next to each one. An unanswered objection is a question or a risk, and both belong in front of the sponsor.