Reviewing a deal model before it goes to the partner: a line-by-line checklist

A deal model gets reviewed twice: once by the partner before it goes out, and once, far more aggressively, by the buyer’s diligence team. The errors that matter are the ones that pass the first review and fail the second. They are rarely broken formulas — those get caught. They are silent errors: a number that’s plausible, internally consistent, and wrong.

This checklist is built to catch that class of error before the model leaves the building. It is organized by layer, from traceability through to presentation. Use it as a gate, not a suggestion: a model that can’t pass it is a model that will fail diligence.

1. Traceability

The first question a serious reviewer asks of any number is “where does it come from?” If the model can’t answer instantly, nothing downstream can be trusted.

Every input-layer figure carries its source: document, page, and line.
The authoritative version of each source document is identified (which management accounts, which QoE draft) and stale versions are not feeding the model.
Adjusted EBITDA can be walked back to reported EBITDA, with every add-back sourced and rationalized.
A reviewer can pick any three figures at random and verify each in under a minute without asking the analyst.
No figure exists in the model that no one can attribute to a source.

2. Inputs versus logic versus outputs

Most structural errors come from these layers being mixed. The review should confirm they are clean.

No hardcoded numbers are buried inside formulas in the logic layer.
No output cell has been manually overtyped to “make it tie.”
Each assumption appears exactly once and is referenced everywhere it is used — no growth rate living in four different cells.
Inputs that came from outside the model are visually distinguishable from calculated cells.

3. The historical build

The history is the foundation; an error here propagates into every projected number.

Line-item labels are mapped consistently across all periods — the same economic cost is on the same line in FY22, FY23, and FY24.
The P&L ties to the balance sheet ties to the cash flow for every historical period.
Management accounts reconcile to audited figures where both exist, and any difference is explained.
Every historical period foots, including the partial latest-twelve-months period.
Any restatement in the source documents is reflected, not averaged over.

4. Normalization and add-backs

The single most scrutinized section in diligence, and the most common place for an indefensible number.

Every add-back has a documented rationale, not just a label.
One-off adjustments are genuinely non-recurring — no “one-off” that appears in two consecutive years.
Owner’s remuneration normalization is to a defensible market rate, with the basis stated.
Run-rate adjustments are supported by evidence of the run-rate, not an assertion.
The aggregate normalization is sense-checked: if adjustments inflate EBITDA by 30%, the model anticipates the scrutiny that will draw.

5. Assumptions and projection logic

Where judgment is most exposed. The review is testing defensibility, not just arithmetic.

Every projected line traces to an explicit assumption and a historical base.
Growth assumptions are consistent with the historical trajectory, or the divergence is explicitly justified.
Margin assumptions are anchored to historical performance and, where claimed, to a specific operational lever.
Working capital ratios (DSO, DPO, DIO) are consistent with the historical pattern and the business’s seasonality.
Capex assumptions distinguish maintenance from growth and tie to the operating case.
The debt schedule is internally consistent: drawdowns, repayments, interest, and covenants all reconcile.

6. Outputs, sensitivities, and presentation

The layer the partner and the buyer actually read.

Every output figure traces back through the logic layer to an input — nothing is presentation-only.
Sensitivity tables flex the two or three assumptions that genuinely move the answer, not arbitrary ones.
The figures that flow into the IM match the model exactly — no manual divergence between the document and its source model.
Returns metrics (IRR, MOIC, or the relevant strategic measures) are computed on a basis a buyer will recognize.
Charts and summary tables tell the same story the underlying numbers tell.

7. Currency with the data room

A model is only as current as its inputs, and live deals move.

The model reflects the latest version of every source document, not the version that existed when the build started.
Any document uploaded since the last model update has been checked for impact on the existing figures.
Figures dependent on a recently changed document are confirmed updated, not silently stale.

How to use this in practice

Run the checklist top to bottom, because the layers depend on each other: a traceability failure makes the historical-build checks meaningless, and a broken historical build makes the projection checks moot. Stop and fix at the first layer that fails rather than pushing through.

The recurring theme across all seven sections is traceability. Almost every silent error — the un-sourced number, the overtyped output, the stale figure, the indefensible add-back — is, at root, a number that can’t be walked back to where it came from. A model where every figure carries its source doesn’t just pass review faster; it makes most of this checklist self-evident, because the questions it asks are already answered in the model itself.

That is the direction the workflow is moving: from review as a manual hunt for un-sourced numbers, to models that are source-traced by construction — extracted from the data room with every figure tied to its document and page, so the reviewer is checking judgment rather than re-deriving facts. The checklist gets shorter as the model gets more traceable. The judgment it protects does not.