When Competitor Intelligence Makes Your AI Pricing Model Worse

FutureUP
22 hours ago
3 min read

Updated: 11 hours ago

The challenge

A customer once asked us to identify optimal prices across multiple geographies. The challenge looked familiar:

Demand varied significantly by location, purchasing power differed across markets, and a single pricing approach was clearly leaving money on the table.

One of the customer's first hypotheses was straightforward:

Competitor prices should be a strong predictor of the optimal price.

After all, competitors operate in the same market. They target similar customers under common economic conditions and demand patterns, so their pricing should contain valuable information.

In our first attempt, we didn't use any competitor intelligence, and the AI pricing model performed very well.

We naturally thought that adding competitor prices would make the model even better and more realistic. However, the result surprised us:

After adding competitor prices, the AI model's performance deteriorated significantly.

It did not improve slightly less than expected — it became substantially worse.

More data does not always create a better AI pricing model

Our first reaction was to investigate the data and methodology.

Were the competitor prices collected incorrectly?

Were we missing important differences between offerings?

Did competitors have different positioning or target audiences?

But after looking more closely, the explanation became clear: the customer was one of the relatively few companies pricing consistently and rationally according to customer willingness to pay and market conditions.

Many competitors were not. Some were systematically underpricing. Others were charging prices that were difficult to justify based on local demand and economic conditions.

Bottom line:

Competitor prices were not providing a clean market signal. They were adding noise.

The model became worse because it was trying to learn from pricing behavior that was inconsistent or economically irrational.

Your own pricing data already reflects the market

This was a valuable reminder.

A company’s historical data does not exist in isolation.

It already — at least partly — reflects:

changes in customer demand
competitor actions
market growth or contraction
economic conditions
promotions and discounting
customer and product differences
previous pricing decisions

When competitors reduce prices and customers react, that reaction may appear in your volumes, conversion rates, discount levels, or customer behavior. When competitors increase prices, the resulting shift in demand may also become visible in your own transactions.

This means that, in many cases, companies can learn a great deal about the market from the data they already own.

This may include their own pricing and transaction data, supplemented by available market intelligence, economic indicators, or industry reports.

Competitor information may still be valuable. But it should be considered as one possible explanatory factor, not treated as an irreplaceable input for pricing decisions.

A simple test

How can you determine whether competitor pricing data is genuinely needed?

Start by building the strongest possible model using the information already available (what we did). Then evaluate how accurately it predicts customer behavior and sales performance.

If your AI pricing model is already accurate, then adding competitor intelligence probably won't add much, or may even hurt accuracy, as it did in our case.

If the initial model is not sufficiently accurate or if competitor intelligence is readily available without much effort, then it is worth adding it and comparing the results.

The point is not to assume that competitor information is necessary — but to test its incremental value before deciding what is important and what is not.

A final thought

The strongest pricing decisions begin with an understanding of your own customers, products, segments, and price-response patterns.

Competitor intelligence, if available, should then be used to provide context, challenge conclusions, and improve the analysis where it adds measurable value.

Not the other way around.

The lesson from this case was simple:

More data does not automatically mean more intelligence.