Definition
The degree to which the AI pricing model's demand sensitivity assumptions match current market behavior, measuring whether the model is pricing based on today's reality or on a market that no longer exists.
The Measurement Gap
The Gap
Pricing models are trained on historical data. They learn demand curves, competitive dynamics, seasonal patterns, and price sensitivity from past behavior. Then they price today's products in today's market based on those learned patterns. The problem is that markets shift. Customer demographics change. Competitors enter and exit. Economic conditions fluctuate. And the model's understanding of demand becomes stale -- sometimes gradually, sometimes abruptly.
Model accuracy metrics measure overall prediction quality but typically aggregate across the entire price range, which masks localized drift. Revenue variance (actual vs. predicted) captures deviations but does not diagnose whether the cause is model staleness or random demand variation. Pricing engine dashboards show the recommended price and the actual price but do not measure whether the model's demand assumptions behind that recommendation still reflect reality. Mean absolute error on demand forecasts provides a technical accuracy score but does not translate into the operational question: is this model's understanding of the market still current enough to trust?
The structural problem is that pricing models drift silently. A model that was 90% accurate when deployed six months ago may be 70% accurate now, but it is still producing price recommendations with the same confidence. No alarm fires. The recommendations just get gradually worse, and the margin impact accumulates before anyone notices.
The Decision
We propose PMCS to answer: Is our pricing model's understanding of demand still accurate, or is it pricing based on a market that no longer exists? This metric would track how closely the model's demand predictions match actual demand at each price point, producing a single currency score that flags when recalibration is needed.
Formula
Plain Language
Every time the pricing AI sets or changes a price, it implicitly predicts how demand will respond. PMCS measures how well those predictions match reality. For each price change event in the measurement window, compare the model's predicted demand to actual observed demand. Calculate the average absolute percentage error across all events. Subtract from 1.0 to produce a score where 1.0 means perfect prediction and lower scores indicate increasing staleness. A model scoring above 0.8 is reasonably current. Below 0.6 is significantly stale.
Notation
PMCS = 1 - (1/N) x Sum|Demand_predicted_i - Demand_actual_i| / Demand_actual_i
Variables
| Variable | Definition | Units | Source System |
|---|---|---|---|
| Demand_predicted_i | The pricing model's predicted demand at price point i when the price was set | Units or revenue | Pricing Engine / ML Model Logs |
| Demand_actual_i | The actual observed demand at price point i after the price was set | Units or revenue | E-commerce / POS |
| N | Number of price change events in the measurement window | Count | Pricing Engine |
Proposed formula, pending validation pilots.