Blog: Pricing & AI—Retail Analytics and Strategies
The newspaper of the pricing
Price alignment involves setting prices relative to those of identified competitors, within a selected scope and with a target margin. It is an internal policy, distinct from the lowest-price guarantee promised to the customer, and its natural scope is the core set of products the customer compares.
When applied across the board and automated, it creates a ratchet effect: price cuts follow competitors’ promotions immediately, while price increases are delayed, and margins erode without any deliberate decision. A well-designed alignment policy defines the scope, verifies product matching, and accommodates discrepancies elsewhere.
There is no single "best" pricing policy in absolute terms: you must decide which policy to apply to which subset of the catalog, and why. Three factors shape the decision-making framework: the product category and its price elasticity, the product's life cycle, and the retailer's competitive position.
A single " Markup " applied to the entire store is not a policy—it is a lack of choice. A retail group in the Caribbean that assigned a specific role to each product category gained one point in the " Markup " for the products tested over the course of a quarter.
A comprehensive competitor pricing monitoring system is built on five inseparable components: data collection, product matching, alerts, reporting, and governance. If even one of these components is missing, the system becomes ineffective.
Data collection is no longer the weak link: according to Diamart Group (2026), 94% of French retailers collect competitive data, but about half still perform the matching manually.
Coverage, timeliness, and compliance are prerequisites for a competitor price monitoring tool. Its value depends on the accuracy of product matching and on linking each price to a traceable decision, yet approximately half of French retailers still perform this matching manually (Diamart Group, 2026).
The listed cost is only part of the total cost. Manually reconciling discrepancies, maintaining in-house development, and making decisions based on unreliable data often end up costing more than the subscription.
A PIM can store, calculate, and publish a price. It generally cannot determine whether that price is the right one: price elasticity, competition, cannibalization, and impact simulation fall outside its scope.
The threshold beyond which it is no longer sufficient to count the number of distinct prices to maintain—SKUs multiplied by regions and channels—rather than the number of SKUs. The two tools complement each other: the PIM ensures the reliability of product data, while the pricing solution turns it into a simulated and governed decision.
AI is only as good as the data it’s fed, and a more powerful model trained on poor-quality data produces more convincing errors. Obvious errors are easy to spot; it’s the plausible errors—such as a purchase cost that’s off by a few percent—that erode the margin.
Waiting for perfect data is another way to fail. A broken data model is fixed before configuration; the rest is made reliable through use, starting with the data points that drive sales, with a manager assigned to each type of data.
A general-purpose language model predicts the most plausible price, not the actual price: it has neither an up-to-date catalog nor a signal of uncertainty, and its training leads it to provide an answer rather than admit that it doesn't know.
For a retailer, the risk is twofold: inaccurate figures in internal decision-making, and incorrect prices displayed to customers who shop using a virtual assistant. The solution lies in connected data, not in the power of the model.
An AI that executes carries out a human decision within a written framework. An AI that decides makes its own choices. A permanent alignment rule, executed thousands of times, is in reality a delegated decision that warrants a review date.
The slider is set based on the type of decision by considering the stakes, volume, and lead time. In France, the 30-day reference price makes certain automatic price reductions costly for subsequent promotions.
In pricing, “AI” encompasses three components: statistical methods that apply a formula, machine learning that learns from data, and generative AI that creates models. They address neither the same needs nor the same risks.
Margin thresholds and legal rules must remain explicit—never implied. To spot a rebadge, ask what the system does when your data changes, and who approves it before a price changes.
A successful pilot test proves almost nothing: on a small scale, human attention masks the flaws in matching, alerts, and frequency that become immediately apparent on a large scale.
Systems that are in place advance in stages, but no stage is mandatory: we move on to the next one when maintaining the previous one costs more than moving beyond it. And what scales up is not the number of employees, but the traceability of decisions.
Excel works well as long as the volume remains low. Once you have more than 3 or 4 competitors or a few hundred SKUs, compiling a complete report takes days of data entry, and data freshness, product matching, and traceability gradually deteriorate without anyone noticing.
The most costly limitation is the least obvious one: a cell does not indicate when the price was updated, nor does it indicate whether an empty cell means the product is unavailable, out of stock, or simply not recorded.
A marketplace is not a competitor, but rather a constant competition among sellers, governed by an algorithm that considers more than just price. Monitoring it requires more frequent data collection, a seller-by-seller analysis, and a total price that includes shipping.
When setting prices, the right benchmark is not the lowest price but the net margin after commission. A minimum price calculated based on this margin, a tolerance threshold for each category, and an accepted price difference compared to the direct channel provide better protection for the price image than automatic price alignment.
Monitoring the entire catalog is not a sign of thoroughness. Sales and price perception are concentrated on just a few items, and exhaustive coverage dilutes the focus on the discrepancies that really matter.
Four categories deserve close monitoring: products that drive price perception, those that drive profit margins, those your competitors are highlighting, and your private-label brands, which should be tracked in comparison to the national brand. The rest can be monitored from a distance—by choice, not by oversight.
In online supermarkets in the eurozone, nearly one in ten food prices changes every week, even during periods of low inflation, and about four out of ten of these changes are temporary. Data covering just a few days can therefore be inaccurate in either direction—upward or downward.
The main risk isn't a missed price; it's when a promotion that has already ended is mistaken for a permanent price. The optimal data collection frequency should align with the pace of your pricing decisions, and each competitor's price must include its date and status.
Price image is a perception based on a few key reference points (KVI), not an average of the catalog. Aligning the entire catalog with the lowest price squeezes margins without improving that perception.
Two effects are often underestimated: prices that change too frequently erode trust, and an automatic price drop lasting just a few days becomes the basis for the next promotion for 30 days. Monitoring without undermining means making a decision before adjusting prices.