How a deal is scored

Back to the guide

Worth understanding, because the score is only useful if you know what it means and where it can be wrong.

Three steps

1. Reading. The document is read and the figures are pulled out: revenue, earnings, addbacks, customer concentration, employees, asking price, industry, location. Each figure is recorded with the page it came from, so every number can be traced back to the book.

2. Checking the hard filters. Your buy box contains rules that are not negotiable: an earnings range, a revenue range, a customer concentration ceiling, industries you will not look at, states you will not look at, and any deal-breaker terms you have written down.

A deal that fails any one of these scores zero and is banded pass, and the reasons are listed on the deal. No amount of being good elsewhere rescues it, because that is what a hard filter means.

3. Scoring what is left. Deals that clear the filters are scored out of 100 against your weighted factors. The weights are yours; the arithmetic is fixed.

The bands

Band	What it means
Pursue	Fits the buy box well. Worth a person's time now.
Review	Plausible but imperfect. Worth a look when you have capacity.
Pass	Outside the buy box, or failed a hard filter.

The band is a ranking, not a decision. Nothing is deleted or hidden because it scored badly, and you can open any deal and disagree.

Where the model is used, and where it is not

This matters more than anything else on the page.

The model reads. It does not score. Extracting figures from a hundred page document is a reading task, and that is what the model does. Turning those figures into a number out of 100 is arithmetic against your rules, and that is ordinary code.

So the model cannot decide it likes a company. It cannot award marks. If a score looks wrong, the cause is either a figure read incorrectly, or your buy box saying something you did not intend. Both are visible, and both are fixable.

Readiness: Look, Wait or Pass

Separate from the score, each deal gets a readiness verdict about the document rather than the business.

Look — there is enough here to form a view.
Wait — too much is missing to judge it properly. Usually a teaser that needs the full book first.
Pass — not worth going further.

A strong score with Wait means the figures look good but rest on too little. That combination is the single most useful thing on the screen, and it is why teasers and CIMs are judged by different standards.

Versions

Every score records which version of the buy box produced it.

Change the buy box and old scores stay as they were. They are not quietly rewritten, so a deal you passed on in March still shows the rules you were applying in March.