From a research perspective, retail pricing is undergoing a structural transition. What has traditionally been treated as a tactical or rules-driven function is becoming a dynamic decision system that must continuously respond to changes in consumer behavior, competitive actions and supply conditions.

Despite significant advances in analytics and machine learning, many pricing systems still fall short of delivering meaningful improvements beyond incremental gains. In my view, this is not primarily a limitation of data or algorithms, but rather a limitation in how pricing problems are framed.

Most current systems rely on either heuristic rules, such as competitor price matching or models trained on historical sales data. While both approaches provide value, they share a common limitation: they infer relationships from observed outcomes without fully identifying the underlying structure of consumer decision making.

In practice, this leads to pricing strategies that respond to surface level signals rather than true demand drivers. Retailers may match competitor prices even when those competitors are out of stock or adjust prices based on short term fluctuations that do not meaningfully reflect consumer sensitivity. More fundamentally, these systems often assume uniform price responsiveness across products, which is rarely the case.

A more robust framework treats pricing as a structured system built around three core components: a model of consumer choice, controlled experimentation to identify causal effects and constrained optimization aligned with business objectives.

Structural modeling of consumer choice

At the foundation of effective pricing is a clear representation of how consumers make decisions. Customers don’t evaluate prices in isolation. Their choices are influenced by a combination of factors including product availability, brand perception, reviews, convenience and fulfillment experience.

From a modeling perspective, this aligns with discrete choice frameworks where consumers select the option that maximizes their utility given a set of observable and unobservable attributes. Price is one component of this utility function but its effect is conditional on context.

This perspective is important because it shifts the focus from estimating isolated price effects to understanding substitution patterns across products and retailers. It also highlights that demand is inherently heterogeneous. Different consumer segments respond differently to price changes depending on their preferences and purchasing behavior.

The role of experimental identification

A central challenge in pricing analytics is identification. In most retail environments, historical price variation is insufficient to reliably estimate demand sensitivity. Prices are often adjusted in response to competitor behavior or internal policies, which introduces endogeneity. Additionally, external factors such as promotions, seasonality and marketing campaigns confound observed relationships between price and demand.

As a result, purely observational models tend to capture equilibrium outcomes rather than true causal effects.

Controlled experimentation provides a way to address this limitation. By introducing structured and bounded variation in prices, retailers can observe how demand responds under conditions that approximate exogenous change. These experiments must be carefully designed to maintain operational stability, typically through small magnitude adjustments and randomized or rotated exposure across products or time periods.

The objective is not to disrupt the market but to generate sufficient variation to identify key parameters such as price elasticity and cross-product substitution effects.

Heterogeneity in price sensitivity

One of the most consistent findings in empirical retail research is that price elasticity varies significantly across products and categories. This heterogeneity is not random but systematically related to structural factors.

High frequency purchase items tend to exhibit greater price sensitivity because consumers develop stronger reference points for what constitutes a fair price. In contrast, low frequency or high involvement categories often show lower elasticity, as purchase decisions are driven more heavily by brand perception, quality considerations or functional attributes.

Similarly, premium or differentiated products tend to exhibit lower sensitivity to price changes compared to commoditized items. These differences have direct implications for pricing strategy because they suggest that uniform pricing rules are inherently inefficient. A key implication is that pricing systems must explicitly account for heterogeneity rather than relying on aggregated or category level estimates.

Constrained optimization in practice

Once demand is structurally estimated, the pricing problem can be formulated as a constrained optimization problem. The objective may be defined in terms of revenue, profit or a combination of both, but in practice it must be evaluated within a set of operational constraints.

These constraints typically include margin requirements, price floors and ceilings, and strategic positioning considerations across categories or brands. In many cases, additional constraints arise from supplier agreements or channel consistency requirements.

Rather than attempting to encode all tradeoffs into a single objective function, a more effective approach is to optimize within clearly defined boundaries. This ensures that pricing decisions remain consistent with the broader business strategy while still allowing flexibility at the SKU level.

This formulation also improves interpretability, as it separates economic objectives from business constraints in a transparent way.

Validation through experimentation

Any pricing system intended for production use must be validated through controlled deployment. This typically involves assigning products to treatment and control groups, where the treatment group follows algorithmically generated prices and the control group follows existing pricing logic.

To ensure validity, it’s important to account for seasonality, product similarity and external shocks. Rotating treatment assignments across time or product groups helps isolate the effect of the pricing system from confounding factors.

This step is critical because it provides a credible estimate of incremental performance relative to a well-defined counterfact rather than relying on simple before and after comparisons.

Pricing as a connected decision system

An important extension of this framework is recognizing that pricing doesn’t operate in isolation. It’s tightly coupled with inventory allocation, assortment planning, promotion strategy and demand forecasting.

Decisions in one area directly affect the feasibility and optimality of decisions in others. For example, aggressive pricing strategies without corresponding inventory support can lead to stockouts, while misaligned assortment decisions can reduce the effectiveness of pricing optimization.

This interdependence suggests that pricing should be embedded within a broader decision architecture where multiple functions are coordinated rather than optimized independently. In platforms such as invent.ai, this type of integration allows pricing to respond dynamically to signals from inventory and demand systems rather than operating as a standalone function.

For a more detailed discussion of the methodology and how these ideas are applied in practice, read the full research in Harvard Business Review.

Santiago Gallino is a member of the invent.ai Scientific Advisory Board.