Two meanings of the term
"AI pricing models" is used for two different things. SaaS companies use it for how they charge for an AI product: per seat, per token, per outcome. This page is about the other meaning: the statistical models inside pricing software that decide what a B2B quote line should cost. If you came for the first meaning, Chargebee and Maxio have good guides.
The signals a model reads
A quote line arrives with a customer, a part, a quantity and a date. The model turns those into signals it has learned weights for. Two come from the customer and the order.
Customer tolerance
How far below list this account and its segment go before the line is lost, from their own won and lost quotes.
Volume response
What a 400-piece order is worth compared with a 40-piece one, from order sizes and the prices they closed at.
Signals from the part
Two come from the part itself.
Seasonality
How each part's demand moves through the year, and whether to hold price in the peak months.
Stock pressure
Days of cover and lead times, and whether scarcity supports a higher price today.
Signals from the market and the supplier
The last two come from outside.
Market position
Where the line sits against matched competitor prices and stock right now.
Cost movement
How much of a cost change to pass through, from the part's own cost history.
From signals to one price
Each signal becomes a factor with a size, in percent off or on list. The factors are summed, the policy applies its floor, ceiling and rounding, and the result is the price. A confidence score says how much history stood behind it. Low confidence means the engine stays close to list and says so.
The learning loop retrains as outcomes sync. Learned offset drift shows where the model's weights have moved from the policy's assumptions, so a pricing manager can decide whether the market moved or the model did.
How to inspect a price
Open any line in the decision inspector. It shows the inputs the model saw, each factor and its size, the policy limits that applied, the confidence, who approved an exception if there was one, and when the price was written back. The same view is on the quote line for the rep, in a shorter form.