When the Algorithm Makes Clearance Worse: How a Fashion Retailer Used Market Research and Consulting to Fix a Markdown Analytics Model That Was Eroding Margins

Markdown optimisation is one of the more commercially consequential applications of retail analytics, because the price at which a slow-moving line is cleared affects not only the margin recovered on that specific inventory but the signal sent to the broader market about the brand’s residual value positioning. A markdown that is too shallow leaves inventory unsold and pushes clearance cost into a later period.

The client’s trading team had been using the third-party platform’s recommendations as a starting point but overriding them frequently, because the recommended markdowns consistently felt too aggressive for the actual pace of sale the team was observing on the shop floor. The overrides were not being tracked systematically, which meant the platform’s vendor had no visibility into how often its recommendations were being rejected, and the client had no structured record of whether the overrides were producing better outcomes than the model would have.

The result was an expensive platform generating recommendations the trading team did not trust, a vendor attributing the performance gap to demand conditions rather than model calibration, and a client with no independent analytical basis for either accepting the vendor’s explanation or refuting it.

Key Challenges

• No independent market benchmarking of markdown optimisation platform outcomes at comparable mid-market fashion retailers to contextualise the client’s own performance deterioration.
• No consulting audit of the platform model’s training data, input assumptions, and recommendation logic against the client’s specific trading environment.
• A trading team overriding recommendations frequently without a systematic record of whether overrides produced better outcomes than the model.
• A vendor attributing the performance gap to external demand conditions rather than addressing the model calibration question directly.
• Gross margin on clearance lines declining against the pre-platform baseline despite the investment in analytics capability.
• Board pressure to either demonstrate a return on the platform investment or make a decision about its continuation before the next trading year.

Retail markdown analytics models are only as reliable as the data they were trained on and the demand signals they are reading. A model trained on an incomplete picture of a retailer’s own channel mix will systematically misread the remaining demand for a slow-moving line, and the resulting recommendations will consistently push markdowns deeper than the actual pace of sale warrants.