AI or business rules: 
When is AI pricing worth the investment?

Well-defined business rules are often sufficient for a narrow, stable, and loosely interdependent product catalog. AI becomes cost-effective when decisions number in the thousands, products influence one another, and price sensitivity varies.

But you still need a usable sales history. In most cases, the right answer is a hybrid approach: the rules set the framework, and the model optimizes within it.

AI has become a staple of discussions about pricing. However, it isn't always the best solution, and saying so is by no means an admission of weakness.

This article is intended for pricing, sales, and data teams that are considering whether to invest in an AI model. It outlines the signs that indicate business rules are no longer sufficient, as well as the often-overlooked factor: data.

A grid of clear lines that extends into a scatter plot connected by curves

The wrong question: “Do we need AI?”

Many pricing projects start with a question about technology. Do we need AI, machine learning, or a bot? That’s approaching the problem the wrong way.

The right question concerns the decision that needs to be made. How many prices should be set, how often, and with how many intersecting factors? The answer then determines which tool is appropriate.

For some companies, statistical methods and well-defined rules are more than enough. For others, they quickly reach their limits. The key question is which category you fall into.

What business rules do very well

A business rule is explicit: “Stay below a certain competitor on these products,” “Never go below a certain margin.” It is clear, predictable, and easy to explain.

For a streamlined catalog with few price changes and products that aren't closely related to one another, this is often the best solution. A more sophisticated model wouldn't offer much more.

Rules remain just as essential even when AI is involved. Legal requirements, differences between product lines, and adherence to a brand positioning aren't something you learn—they're decided upon and laid out in writing.

Their weakness becomes apparent over time and as the volume increases. Each specific case requires yet another rule, until we end up withan overwhelming number of rules: a system that no one can understand anymore.

Signs That We've Gone Beyond What the Rules Can Do

There is no universal threshold that says, “Once you reach a certain number of references, you need AI.” But certain indicators, when taken together, suggest that rules alone are no longer enough.

The number of decisions

Thousands of SKUs, multiple channels or price zones, and frequent price adjustments: the number of prices that need to be set exceeds what a single team can manage with precision.

The rules are then applied by broad categories, and the nuances are lost. This is often where managing the data in Excel reaches its limits.

Cross-effects

When lowering a price shifts sales of a related product, a rule- product by product al approach fails to account for the overall effect. It is necessary to measure both cannibalization and the halo effect.

This is where models really shine: they estimate these interactions based on past sales, whereas a rule-based approach would require us to guess them.

Varying price sensitivity

While customer reactions vary greatly depending on the product, season, or store, applying a one-size-fits-all rule leaves profit on the table. Measuringprice elasticity by product therefore becomes profitable.

LocationWhat is often enoughWhen a model provides
Streamlined catalog, stable pricesClear rules and regular reviewsRarely
Thousands of entries, frequent recalculationsRules by Major FamiliesTo refine product by product
Products That Cannibalize ThemselvesRules for the spacing between rowsTo measure cross-effects
Price sensitivity varies widelyCoefficients by sectorTo estimate the elasticity of each product
Short or incomplete sales historyRules: Time to Make Them More ReliableNot yet
Our Position
Pointing out where AI isn't useful is part of the job

A model applied in a situation where a rule works just as well is more expensive, harder to understand, and doesn't improve anything. Recommending it anyway is detrimental to the project.

We prefer to start with decisions where a model really has something to learn, and let the rules handle the rest.

The factor we tend to overlook: data

A complex catalog isn't enough. To learn, a model needs historical data: sales figures, price fluctuations, and documented promotions. Without that, it has nothing to learn.

63%

Some organizations do not have—or do not know if they have—data management practices suitable for AI.
(Gartner, press release dated February 26, 2025)

A product that sells a few times a year—or whose price has never changed—says nothing about customer response. We’ve outlined what’s required for a reliable sales history.

Quality matters just as much as quantity. Incorrect costs or incomplete competitor data lead to recommendations that are precise but inaccurate: this is the glass ceiling of AI pricing.

An AI project launched with data that isn't ready is costly and disappointing. It's better to know that beforehand.

60%

AI projects that do not rely on AI-ready data will be abandoned by the end of 2026, according to forecasts.
(Gartner, press release dated February 26, 2025)

Most often, the answer is a hybrid approach

It is misleading to pit AI against rules. In most cases, the two coexist: the rules set the framework, and the model optimizes within it.

The rules specify what is prohibited or required. The model searches for the best price among the options that are still permitted, taking into account factors that the rules cannot consider.

A single catalog can also combine both approaches. Best-selling products and high-volume long-tail items warrant a specific model, while slow-moving items often benefit from simple rules.

We described this balance in our article on how an AI pricing engine works and its safeguards.

How to Make the Right Decision for Your Business

Decisions are based on expected value, not on the appeal of the technology. A model that improves decision-making for a small portion of the catalog can already justify the investment, provided that portion carries significant weight.

Conversely, there’s no point in paying for AI when a rule works just as well. The right calculation compares the decision-making benefit to the project cost, as we explained when discussing the ROI of a pricing solution.

If you're unsure, a price diagnostic allows you to gauge where current rules leave room for improvement, and where a model could really benefit from learning.

The question to keep in mind
Which pricing decisions does your team struggle with the most because they can't measure the impact?

That's where—and nowhere else—a model is most likely to be profitable.

Are you wondering if your pricing strategy needs AI?

Our experts will work with you to review your volumes, data, and current policies to determine where a model would truly make a difference.

Let's discuss your pricing project

Frequently Asked Questions: AI or Business Rules in Pricing

No. For a streamlined catalog with few price changes and products that are not closely related to one another, well-defined business rules and regular reviews are often sufficient.

AI adds value when the number of decisions exceeds what a team can carefully weigh, when products influence one another, or when price sensitivity varies significantly.

The question we need to ask ourselves is about the decision, not the technology.

A business rule is written by the team: stay below a certain competitor, don't go below a certain margin. It is clear and predictable, but it only sees what has been defined for it.

An AI model learns from historical data, such as how sales respond to price changes. It can estimate effects that no rule captures, such as cannibalization between products, but it requires a sufficient and reliable history of data.

There is no universal threshold. The number of SKUs matters, but so do the frequency of price changes, the number of channels or regions, and the interactions between products.

A catalog of a few hundred closely related products may warrant the use of a model, whereas thousands of independent and stable SKUs can be effectively managed using rules.

It is the combination of these factors, along with the available data, that tips the scales.

At a minimum, a sales history by product that includes actual prices—ideally covering several seasons—and sufficient price variations to observe customer reactions. Promotions must be documented.

Purchasing costs and competitor prices must be reliable. A product whose price has never changed—or that is sold only a few times a year—does not allow a model to learn anything.

Yes, and that's the most common scenario. The rules set the framework: legal requirements, minimum margins, price differentials between product lines, and positioning. The model finds the best price among those that are still permitted.

A single catalog can also combine both approaches: a model for high-volume or highly price-sensitive products, and simple rules for slow-moving items.

The expected gain from pricing decisions must be compared to the project cost, including data costs. A model that improves decision-making for a small portion of the product catalog may be sufficient if that portion accounts for a significant portion of the margin.

A preliminary assessment helps determine where current rules leave room for improvement and whether the available historical data allows a model to learn. Without this step, there is a risk of paying for AI when a rule would have worked just as well.

‍

Articles
similar
Data flows from multiple sources pass through three stages before reaching human validation
November 13, 2026
Pricing Manager and AI: What the Machine Does, What the Expert Decides

AI handles the volume at every stage of the data journey: collection, product matching, verification, forecasting, and simulation. The expert selects competitors, resolves ambiguous cases, provides context, formulates the strategy, and makes decisions.

At Booper, Booper Mind makes suggestions, MPS monitors, and the user decides. Knowing how to interpret elasticity and question a forecast is more important than knowing how to code.

Read the blog post
Scales of Justice placed on a price dashboard, between three colored areas
November 11, 2026
Algorithmic Pricing: Where Is the Legal Line for a Retailer?

Dynamic pricing is permitted in France. Personalized pricing based on automated decisions is allowed, but customers must be notified of it. Algorithm-based price-fixing remains prohibited, even in the absence of contact between competitors.

The strongest safeguard for a retailer: being able to say, for each price, which rule generated it, what data it was based on, and who approved it.

Read the blog post
AI shopping assistant comparing price tags from multiple retailers on a screen
November 11, 2026
Agent-Based Retail: Why Price Perception Is No Longer Shaped on the Shelves

An AI agent does not simply look at a listed price: it compares the actual price—which includes applicable promotions, shipping, availability, and unit price—for the products the customer requests.

Competitiveness is now measured rather than perceived. The accuracy and consistency of pricing data across channels have become the brand's new calling card.

Read the blog post
Ready to
 booster
What about your profit margins?

The pricing intelligence solution for retail leaders. Accuracy, speed, and controlled margins.

Let's discuss your pricing challenges
‍
‍