Somewhere right now a vendor is telling a buyer that their forecasting is 98% accurate. The buyer will not ask on what error metric, over what horizon, against what baseline, measured by whom. The claim will hang there, technically unfalsifiable, emotionally reassuring, and the deal will move forward on the strength of a number that means nothing.
This is the equilibrium of the enterprise AI market in 2026: every claim confident, every benchmark self-administered, every case study a survivor. And it has produced exactly the buyer behaviour it deserves, with leaders re-checking everything these systems produce, because nothing in the sales process gave them grounds not to.
In a market where every statement is confident, the only differentiated statement is an honest one.
What honest looks like as product behaviour
Honesty in an intelligence system is not a tone. It is a set of concrete behaviours, each of which costs something to build and pays back in credibility:
- Say what is not predictable. Some questions, on some data, have no learnable signal. The honest system says so at the start: this target, with your current history, cannot be forecast better than a naive baseline. That sentence loses a line item and wins the account, because it makes every other claim believable.
- Publish the misses with the hits. Every model in production carries its backtest, including the quarters it got wrong. One miss, on the record with the rest, is worth more than ten case studies, because it proves the record is real.
- Test against luck. Run the backtest 20 times, once on the real signal and 19 on noise. If the real result does not stand clearly apart from the noise results, it is not a result. A finding that cannot be told apart from luck should never reach a customer slide.
- Keep the probabilities honest. When the system says 70%, about 70% of those calls should land, and the reliability curve proving it should be published each cycle, not asserted at purchase.
The cheapest thing to manufacture in AI is confidence. That is precisely why it is worthless.
The commercial case, since it has to be made
The objection is obvious: honesty narrows the pitch. It concedes territory a competitor will happily claim. This is true for one meeting and false for the relationship. The confident vendor owns the demo; the honest one owns the renewal, because their claims were the ones that survived contact with production.
The harder commercial edge cuts deeper. An honest-limits posture converts the sales conversation into an evaluation the buyer can run: agree a metric and a baseline in week one, run on the buyer's data with the buyer's users, and measure. A vendor who publishes misses can offer that without flinching. A vendor who cannot, cannot, and buyers increasingly notice which is which.
The deepest reason, though, is not commercial. Enterprises are about to delegate real decisions to these systems. The ones that get trusted should be the ones that earned it, and the earning mechanism is a published record with the failures left in. Any market that punishes honesty about limits will eventually be corrected by its own incidents.
Trust the system that tells you what it cannot do. It is the only one whose other claims mean anything.
Honest limits are built into how Prophesee works. Unpredictable targets are named as such, every published backtest includes the predictions it got wrong, and the calibration record is published every cycle. Ask it what your data cannot support. Start here.