Accuracy
Tested on your past. Every time.
A simulation is only worth what it gets right. Here's how Avir checks itself, on your data, before you rely on it.
Backtested before you see it
Avir hides your most recent months from itself, forecasts them from what came before, and compares. You see the error, how often the real month fell inside the range, and a trust grade, next to every forecast.
Calibrated on changes you really made
If you changed a price in the past, Avir replays it: the months before, the change, and what customers really did. How strongly the agents and the AI panel react is measured against that, and corrected.
Outside data earns its place
Seasons, holidays, weather, fuel, exchange rates and your own series are each tried in turn, and kept only if they cut the forecast's error by at least a tenth on your own history. You see which were kept and why.
A control in every panel
The AI panel is always also asked about no change at all. If it imagines an effect where there is none, you'd see it, and so would the correction.
On our demo business
How close it came on months it hadn't seen.
From our demo workspace. Your numbers will be your own, and shown with every forecast.
What we'll publish
Accuracy across real customers, method included.
As our first customers' forecasts meet their real months, we'll publish how close they came, across businesses, with the method and the misses. Until then, every customer sees their own accuracy in the product, next to every forecast.
Be one of the first