Which sales signals actually predict a closed deal.

The four everybody names.

Reply speed. A buyer answering inside two days is in a different state from one answering in nine, and the gap widens as a deal ages.

Multi-threading. A second and third person joining the thread is the most repeated finding in every write-up of this question, and it holds up more often than not.

A scheduled next step. Not whether somebody typed one into the CRM, but whether a meeting actually exists on a calendar.

Procurement, legal or a security review opening. Somebody spending their own political capital on your deal is doing it because they expect to buy.

All four are reasonable. All four are guesses until they are checked.

Why a list is not an answer.

A security review is routine in one business and a six week stall in another. A discount request is a buying signal in one and a stalling tactic in the next. The same list cannot be worth the same in both, and a tool that ships one set of weights for every customer is asserting that it is.

The question is not which signals matter. It is which signals matter here, and by how much, and that is a measurement rather than an opinion.

How to check them against your own deals.

Take every deal you closed in the last year or two, won and lost. For each one, look only at what was known before it closed, so the outcome cannot leak backwards into the inputs.

Count how often each signal appeared on a win and on a loss. A signal that shows up equally on both predicts nothing, however sensible it sounds.

Then hold a slice of those deals back. Fit the weights without them and ask the model to rank them. If it does well on the deals it learned from and poorly on the ones it never saw, it has memorized your history rather than learned anything about your next quarter.

That last step is the one almost nobody does, and it is the only number that is a claim about tomorrow.

What the buyer did, not what the rep typed.

Most of what sits in a CRM is a rep describing a deal: a stage, a probability field, a next step written to satisfy a pipeline review. It is the account of somebody with a quota, and it is the most confident data in the system.

What the buyer did is different. They replied or they did not. They brought a colleague or they did not. They opened a security review, named a competitor, asked for pricing in writing, went quiet for twenty eight days. None of it is typed by anybody and none of it has a reason to flatter.

What Kaypo does with this.

Kaypo reads the buyer side of your CRM, email and calls, drops anything your reps said before matching, and looks for 929 phrases across 55 concepts and eight categories, weighted by how recent each match is and discounted for repetition.

It fits those weights on your own closed deals, then tests itself on a slice it was never shown and reports that number next to the flattering one. Every input is visible, every weight is editable, and if the fitted model loses to the industry benchmark, Kaypo keeps the benchmark and tells you.

It needs about thirty closed deals to fit anything. Below that it uses the industry starting point and says so.

Questions

Related reading