Question
Run the model on day 1 of past quarters with only what was known that day, compare with what closed, and check it beats the simple baseline.
The short answer
An AI forecast is as accurate as its back test. Run the model on the first day of each past quarter with only what was known that day, compare it with what closed, and compare that miss with a simple baseline: booked revenue plus open pipeline times your win rate. A forecast that loses to it is not worth using.
How Kaypo answers this on your pipeline
30 minutes with the founder
Thirty minutes on your pipeline or the example one. Pick a time and it is booked.
The miss is the gap between the forecast and what closed, as a share of what closed. It depends on the day the forecast was made: on day 1 most of the quarter has not happened, and by the last week most of it has.
So a single accuracy number means little. Ask for the miss by day of the quarter, across several quarters, and ask what the forecast was compared against.
Every input has to be what the CRM and the connected tools showed on that day. The stage as it stood then, the close date as it stood then, the signals up to that day and no further.
A back test that reads today's stage for a deal that closed last quarter is reading the answer. It will look excellent and mean nothing. This is the most common way forecast accuracy gets overstated.
Booked revenue plus open pipeline times your historical win rate is free and takes one query. Any model worth paying for should beat it, on your own quarters, from day 1.
If it does not, the extra machinery is not earning anything, and the honest thing for a vendor to do is say so.
A forecast that gives one number hides how unsure it is. A range built from how much the outcome has varied, including deals that were not open yet on day 1, tells you how much to lean on it.
Kaypo shows the expected number, the likely range, what is already booked, the reps' commit beside what Kaypo backs of it, and one line with its own back test, on the Overview.
Each deal type gets the same treatment, because new business, expansion and renewal close for different reasons and miss by different amounts. On the example pipeline the test missed by 9% for new business, 9% for expansion and 14% for renewal, and each type carries its own range.
Recorded in the sandbox, the same example pipeline you can open yourself. Pick any part.
Deals. Every open deal, scored on what the buyer did.
There is no universal number. Compare the forecast's miss on your own past quarters, by day of the quarter, with booked revenue plus pipeline times your win rate. A model that does not beat that baseline is not adding anything.
It needs closed deals to learn from. With fewer quarters the range is wider, and a careful model leaves out the inputs whose effect it cannot measure yet rather than guessing at them.
Kaypo keeps the reps' commit. Its forecast sits beside the commit and checks it deal by deal, so the forecast call starts from the deals where the two disagree.
The product page for this: How Kaypo works.
Reviewed November 8, 2026 against the product as it behaves today.
30 minutes with the founder
Bring this question and your own deals. Pick a time and it is booked.
Next step
The real product on invented B2B software companies. Nothing to connect and no signup.
Open the Overview in the sandboxThe whole product on an example pipeline: scores, next steps, plays and agents. No signup and no call.
$1,500 a month, or $15,000 a year, for one admin and up to 15 users. Free for 14 days, no card.
Or get the free report on your own closed dealsOr talk it through with a person, 30 minutes