Model council for due diligence

The job

You are about to commit — to an acquisition, an investment, a key vendor, a partnership. Someone has handed you a story: the market is growing, the customers are loyal, the risks are managed. Your job is to stress-test that story before the money moves.

Diligence is not reading. It is deciding which claims to verify first, which risks you are underweighting, and which questions to ask while you still have leverage.

• Rank the claims worth verifying before the next call
• Surface the risks the pitch is built to hide
• Turn a gut feeling into a written question list

Why one model is not enough

Ask one model to assess a deal and you get one narrative — fluent, structured, confident. It reads the same whether the model caught the real risk or missed it. A plausible risk memo is not a checked risk memo.

Diligence is the worst place for an unchecked answer, because the miss is the whole cost. Different models weight risks differently. Only comparison shows you which risks every model sees and which one model alone is worried about.

How the council runs your diligence

Three stages, and in diligence the third one earns its keep: the dissent is often the deliverable. To date the platform has judged 237,177 answers in peer review.

• Stage 1 — Independent assessment. Each model reads your material and writes its own risk analysis. No model sees another's take, so no one anchors on the first opinion.
• Stage 2 — Anonymous peer review. The models critique and rank each other's assessments without knowing whose is whose. Thin reasoning gets flagged regardless of which model produced it.
• Stage 3 — Synthesis. One assessment: the risks the council agrees on, ranked, with minority views kept visible. The minority flag is often the risk you came looking for.

A worked example

"I am evaluating a small SaaS business for acquisition. Here is the seller's summary and my notes so far. Which claims should I verify first, which risks am I underweighting, and what should I ask the founder before the next call?"

What comes back is a position, not a summary. A council informs your judgment; it does not replace your advisor — or the data room.

• The verdict: the council agrees the revenue story depends on how churn is defined and on a small set of large customers, and puts those two claims at the top of the verification list.
• The strongest dissent: one model ranks founder dependency above customer concentration — the business may not survive the person selling it. The majority disagrees. You get to decide who is right.
• The document: export the question list as a Word or Excel file, one row per claim, ready to work through in the data room.

From free, on your phone

LLM Council works in any mobile browser. Diligence rarely happens at a desk — run the council between meetings and read the dissent on the train.

The free tier includes one real council every day, all three stages. Pro is $25 a month for more. Fox is $100 a month: the strongest model pool, the deepest effort setting, and a document agent. Exports cover Word, PDF, PowerPoint, and Excel.

When not to use a council

A council is slower than a chatbot because peer review is real work. Some questions do not need it.

• Simple lookups. A founding date, a ticker, a headquarters address — one model, one answer.
• Time-critical single facts. If the call starts in two minutes, a council is the wrong tool.
• Questions that need primary sources. The council reasons over what you give it. It does not open the data room for you.