Blog: Pricing & AI—Retail Analytics and Strategies
The newspaper of the pricing
An AI-powered pricing engine prepares the data, compares your products to those of competitors, models demand, and then suggests prices filtered through safeguards. The output is a quantified recommendation, validated by the team for sensitive cases.
Its weak point is rarely the algorithm. It’s the distinction between regular prices, promotions, and stockouts in the sales history, and the quality of safeguards—including regulatory ones. A recommendation with no expected impact remains merely an opinion.
We structure decisions, not an organizational chart. The hybrid model—the most common one—is defined by what headquarters prohibits, and the role that is most lacking is not that of the analyst but that of the executor, who ensures that the price decided upon is indeed the price displayed.
A single approver per type of decision and a delegation threshold known to everyone are better than an exhaustive matrix. In B2B sales, the key issue is discounts granted beyond the standard pricing schedule.
An effective pricing strategy distinguishes between image products (KVI)—which customers compare—and margin drivers—which they do not compare. Applying a single “ Markup ” to the entire store or matching competitors’ prices across the board erodes margins without gaining customers.
The most costly pitfalls we see among our clients relate to measurement: a volume-un Price Index , price changes made without a control group, and prices tracked in an Excel file by brand. Clear governance and a single reference system correct these issues.
Price image is the perception of a retailer’s price level. It is determined by a handful of benchmark products, which are chosen based on what customers remember and compare—not just on sales volume. An index calculated as a simple average may indicate lower prices when customers actually experience the opposite.
Effective management tracks two metrics side by side: a volume-weighted index, calculated weekly, and customer perception, measured quarterly. When these metrics diverge, the cause is almost always related to a few key products or to communication.
Most pricing dashboards measure what is easy to collect—Price Index —rather than what indicates whether the strategy is working. An index measures a position, not a result.
Three metrics are more important: the margin by product category, the difference between the set price and the actual price charged, and the rate of deviations from the rules. Price perception, on the other hand, is measured among customers, not in a spreadsheet.
AI-based sales forecasting estimates demand for each product, store by store, by separating baseline demand from promotional demand. AI takes in more data points than a spreadsheet, but its value lies solely in the decisions it helps inform.
In retail, this decision is primarily about pricing or a promotion. A useful pricing forecast—one that estimates how much will be sold at each considered price—is evaluated based on its bias rather than its MAPE, and is validated against control stores.
Price elasticity measures the change in volume relative to the change in price: at -1.5, a 1% price increase results in a roughly 1.5% decline in volume. It is measured excluding promotions and stockouts, preferably by store cluster.
The threshold of 1 distinguishes price increases that reduce revenue from those that do not—not those that erode profit margins. With a 25% price- Gross margin, a 10% price increase remains profitable up to an elasticity of approximately -2.9.
Strategic pricing defines, for a period of six to twelve months, the role of each category, the target price level, and the parameters within which daily prices may fluctuate. It does not set prices; rather, it defines the range within which each price is allowed to fluctuate.
A pricing strategy rarely falls apart all at once; rather, it erodes under the weight of accumulated tactical decisions and an ever-growing list of image products. Drawing the line between strategic, tactical, and operational considerations is more important than the sophistication of the calculations.
Promotions will account for 21.9% of consumer goods revenue in 2025—a record high. At this level, the question is no longer whether to run more or fewer promotions, but rather what the purpose of each one is: a promotion without a stated objective cannot be evaluated, and it is repeated out of habit.
There are seven strategies in use, ranging from price image protection to control group testing. Each has its pitfalls: selling off the most frequently compared products at rock-bottom prices, offering discounts to existing customers, and matching every price cut by a competitor. Profit margins are improved by prioritizing these strategies and discontinuing the least effective ones.
A calculation yields a price; a simulation shows how that price will affect volume, related SKUs, inventory, and the category’s margin. Automating the calculation isn’t enough: it’s the projection of the effect—scenario by scenario—that influences the decision.
A simulation analyzes price and volume effects—using conservative, balanced, and aggressive scenarios—and then compares the results to actual performance. Its value lies primarily in the fact that it rules out adverse scenarios before they occur in the real world.
A pricing alert is a notification that is triggered when a price falls outside a predefined range and prompts a designated person to take action. The workflow follows this sequence: who processes it, within what timeframe, and who approves it. In retail, six alerts cover the essentials, ranging from “ below-cost resale threshold ” to local prices falling outside their ranges.
Thresholds should be set based on the team’s processing capacity, not just on risk; otherwise, alerts will be ignored. Workflows must remain short: only issues that fall outside the thresholds are escalated. A good metric is the percentage of alerts followed by a price change.
Endogenous data describe what the company has done (sales, net prices, costs, inventory), while exogenous data describe the context (competition, timing, weather, inflation). Sound pricing decisions always take both into account, and an external signal is only valuable if someone can act on it.
The most costly pitfall lies in the historical data itself: past prices were set with knowledge of demand, and stockouts distort sales figures. Without correcting for these biases, the calculated price elasticity overestimates the revenue generated by a price reduction.
A pricing tool project is based on a pricing decision that takes effect today, not on a list of features. The scope defines this decision—an initial, representative but narrow scope—and determines the price once the tool is in place.
The generalization criterion is defined before the test and is evaluated against a control set. There is no requirement to go through an intermediate step between a spreadsheet and a more advanced tool: this often doubles the scope of the project without reducing the risk.

The Barbotteau Group previously used a single " Markup " in Excel. It now calculates three optimal prices per product—based on volume, revenue, or margin—selected according to each category’s role, and achieved a 1-point increase in the " Markup " rating in one quarter for the products tested, with a significant improvement in its price image.
The group made the transition directly from Excel to a hybrid AI system, skipping the intermediate step of using rules. Support from management and a pilot test—which was first limited, then scaled up—helped secure buy-in, while the teams retain control over every decision.
AI does not make elasticity estimates more accurate for products that are already well known. Above all, it allows us to estimate a credible value for the long tail by sharing information between a reference product and its product family, and to separate the effect of price from that of promotions, out-of-stock situations, and seasonality.
Extreme flexibility is often the first sign of a data issue. And a model is only useful if the team understands its recommendations: projects fail more often because of organizational issues than because of the algorithm.