The Diligence Red-Flag Checklist

A deal model is a story told in numbers. The seller wrote it. Your job is to find the line where the story stops being true.

The new AI models read a data room in an afternoon and they are very good. They are also, sometimes, confidently wrong. So this checklist works on both: the 14 places a deal model bends the truth, and the one place a brilliant model hands you a clean answer that is dead wrong. Use it as a gate. Every target. Before the investment committee, not after.

## Earnings quality (is the EBITDA real?)

1. **Kitchen-sink addbacks.** "Adjusted" EBITDA carrying costs that come back every year. Demand a bridge from reported to adjusted and challenge every line over 2% of EBITDA.
2. **Run-rate revenue off one good quarter.** Annualizing the best three months. Ask for the trailing twelve and the quarterly trend, not the annualized snapshot.
3. **Revenue pulled forward into the sale.** Bookings that spike in the two quarters before a process. Compare deferred revenue and billings to recognized revenue across the run-up.
4. **Synergy addbacks that have not happened.** Strip every unrealized synergy out of the entry number. Pay for what exists, not what is promised.

## Cash and working capital (does the EBITDA convert?)

5. **EBITDA that never becomes cash.** A widening gap between EBITDA and free cash flow is the single most reliable warning in the book. Track the conversion rate over three years.
6. **Working capital normalized to a flattering point.** Stretched payables and pulled receivables before sale inflate the cash that looks free. Use a normalized, seasonally fair level.
7. **Capex understated, maintenance disguised as growth.** Ask what spend is required just to hold revenue flat. That is the real floor.
8. **Deferred revenue masking a declining book.** Look at new logos and gross bookings, not just recognized revenue.

## Leverage and covenants (does it survive a bad year?)

9. **Covenant headroom modeled off the adjusted number.** Re-run every covenant on reported EBITDA and on the downside case.
10. **Refinancing assumed at today's rates for the whole hold.** Stress the rate path. Ask what returns look like if the cost of debt is 200 to 300 bps higher at refi.
11. **PIK and springing terms that look benign until they do not.** Model them in the bad case, where they actually bite.

## Projections and exit (is there a margin of safety?)

12. **The hockey stick with no history behind it.** Demand the operational reason for the year-three inflection. If it is "the market," it is a wish.
13. **Exit multiple at or above entry.** Underwrite the deal on multiple compression and see if it still clears.
14. **A downside case that is not a downside.** If the "bad case" still returns capital, it is a second base case. Build a real one: revenue down, margin down, multiple down, rates up, at the same time.

## The 15th: where the AI lies

A capable model reads all fourteen faster than any associate. On a hard deal it will also: trust the seller's framing because the narrative is coherent (coherent is not true); extrapolate a trend off thin data and state it with the confidence of a fact; miss a contradiction between two documents in the room; and fabricate a tie-out, stating two figures reconcile when they do not.

The fix is not a smarter model. Everyone has the same model now. The fix is a layer on top whose only job is to check the model against the evidence, flag every number that does not tie to a source, and refuse a verdict until the contradictions are resolved. That layer is the difference between a model that is usually right and a verdict you can put capital behind.

## Want this run on a deal you already know?

We build the full engine into your fund, in your own environment, your deal data never leaves the building. Then we prove it on a deal you pick. The first one is free. Reply DILIGENCE or send a message, and we will run it.