How to know if a deal will actually close
You cannot know whether a single deal will close, and any method that claims otherwise is selling something. What you can do is become calibrated: put a number on each deal, record it, and check afterwards whether deals you called at 70% closed about seven times in ten. A calibrated 70% is far more useful than a confident yes, because you can plan with it.
Below: why the usual signals do not work, the three questions that do, and how to build the record that turns judgement into something measurable.
What does not predict a close
Stage. A deal in Negotiation has reached a point in your process, which tells you about your process rather than about this buyer. Stage moves when your rep updates a field. It is an output of activity, not evidence of intent.
Rep confidence, asked directly. "Will it close?" gets a yes, because the question has a socially correct answer and asking it in a group makes that worse. Confidence also rises with time invested, which is the opposite of what you want: the longer a rep has worked a deal, the more certain they become and the less that certainty means.
Recency of contact. A reply last week is not progress. Buyers are polite. Activity data in particular flatters deals that are being worked hard and going nowhere.
The common failure of all three: none of them can be wrong. A signal that cannot be falsified cannot be evidence, and a forecast built on unfalsifiable signals will feel confident right up to the end of the quarter.
Three questions that do predict
1. Has the buyer done something that costs them something? Scheduled a technical review, pulled in legal, named a budget line, given you a security questionnaire. Effort on their side is the only signal that is expensive to fake. Enthusiasm is free; a procurement process is not.
2. Do you know what happens after yes? Who signs, in what order, by what date, and what has to be true for that date to hold. A deal whose owner cannot describe the path from verbal agreement to signature does not have one, however warm the last call was.
3. What would tell you it is not going to close? Ask the owner to name the specific thing that, if it happened, would change their mind. Deals that are genuinely alive have a clear answer. Deals that are hope have "nothing really" — and that is the tell.
Scoring yourself afterwards
None of the above gives you certainty. What they give you is a better number. The value arrives when you check that number against reality, repeatedly, which almost nobody does.
Take every deal that reached a given probability and ask how many of them actually closed. A calibrated pipeline shows a straight line: 70% of the deals you called at 70% close. Most real pipelines do not. They look like this:
Twelve months of a four-rep team. The top of the ladder roughly holds. The middle sags, which is where most forecasts lose their money.
Two things to read off a ladder like this. Deals called 90% closed 76% of the time, so the top of the pipeline is roughly honest and slightly optimistic. And deals called 50% closed 49% of the time — which sounds healthy until you notice that a 50% call is, by definition, no information at all. Half the pipeline sitting at a coin flip is the single most common shape in sales forecasting, and it means nobody has made a judgement yet.
The useful question is not "is this deal going to close" but "is my 70% a real 70%". The first is unknowable. The second is measurable within two quarters, and fixing it improves every forecast you make afterwards.
What to do on Monday
- Replace stage probability with the owner's own number. It will be worse for a quarter and much better after two.
- Record it. One row per deal per month: what was called, what happened. Without the record nothing above can be checked.
- Ban 50% as a lazy default. If a rep cannot choose between 40 and 60, the deal has not been qualified, and that is the finding.
- In the review, ask question three — what would tell us this is dead — rather than asking whether it will close.
I ran this on my own pipeline for a year. I did not get better at reading individual deals in any way I could describe. My forecast error fell from roughly 50% to about 14% anyway, because I finally knew which of my own numbers to trust.