Get an AI second opinion before an important decision

Start with the decision you need to make

“Evaluate my idea” leaves too much room for a pleasant overview that changes nothing. State the choice, deadline and criterion: start a pilot now or after checking the data; buy a service or extend an existing system; accept the full scope or split the work into stages.

Include a credible alternative, such as keeping the current process. The discussion can then examine whether action is necessary. Define criteria before reading the answer: available staff, required capabilities, spending limits and conditions that would stop the work.

Worked scenario: can you promise delivery in a month?

Imagine an agency considering a four-week CRM implementation for handling inbound requests. Its lead wants to accept: the team knows the system and there are few screens to build. Yet the brief does not establish the quality of existing data, access to integrations or the client’s availability for approvals. This is a hypothetical example, not a customer result.

A second opinion may reveal that those dependencies matter more than the screen design. The claim to test is whether the team can deliver with unprepared data. A useful outcome might be a discovery stage with a list of evidence needed to confirm the delivery date. The team remains responsible for the actual schedule and scope.

If every agent endorses the original deadline, ask for the supporting facts. Agreement does not resolve an unknown dependency. If an objection relies on an invented complication, request evidence or retain it as a hypothesis.

Give the participants the same brief

Place a source beside each important fact: a document title, email date or spreadsheet row. That makes the eventual check easier. Use information you are authorized to share with the services involved; anonymized conditions are often sufficient for an initial review.

• Decision: the options you are choosing between and the deadline.
• Established facts: requirements, agreed terms and available documents.
• Assumptions: expectations that have not been observed or agreed.
• Constraints: people, budget, dependencies and mandatory conditions.
• Stop condition: what would make an option unacceptable.

A prompt for a useful second opinion

“Help review this decision: [options and deadline]. Established facts: [facts and sources]. Assumptions: [list]. Constraints: [conditions]. Consider the strongest argument for each option, the strongest objection and the option of making no change yet. Do not treat my preference as evidence. Identify the missing fact most likely to change the conclusion. Finish with a conditional recommendation, preserve material disagreements and list the checks a person must complete. Do not invent numbers or supporting evidence.”

Replace the fields with your own brief. In Colay, choose Consensus for discussion and a shared conclusion. Use Ask separately if you want individual initial answers. For a subsequent synthesis request, supply the relevant excerpts and original brief; do not assume separate chats automatically become its context.

What a useful result should contain

For the CRM example, confirming access and inspecting the data would support the next decision. “All the models think it is feasible” is not enough to promise a date. Keep the original brief with the memo so you can identify which conditions change later.

When one answer is enough

Editing an email or summarizing your own text may only require one model and normal review. Discussion is more worth testing when there is a real alternative and an overlooked condition could change the decision. Colay subscriptions use credits and limits: compare the usage with what the extra review actually clarifies.

Judge your first attempt by an observable outcome. Did you find an unsupported assumption, revise a plan or identify a question the client must answer? More pages of prose do not establish value. Start with one decision for which you have documents and a way to verify the conclusion.