The blind spot of experience: Why we let students challenge our retail logic

The risk of the expert loop

Experience builds speed, but it also builds defaults. After years of pricing optimisation, certain decisions become automatic. Under time pressure, continuity is rewarded. You reuse last season’s playbook, you follow the established discount cadence, and you adjust depths at the edges, and move forward.

Over time, these patterns become invisible. The risk is not that experts are wrong; it is that we stop questioning which parts of our logic are based on current evidence and which are based on inherited habit.

We often call it experience when it is actually institutional memory. Students don’t inherit that memory. They construct the logic from scratch, surfacing assumptions that have been part of the process for so long they are no longer noticed.

A deliberately open brief

This year, we shared simulated retail datasets, based on real-world patterns, with Econometrics students at Erasmus University. The assignment was intentionally broad: “Recommend a markdown strategy.”

In an academic setting, that sounds like a clean optimisation problem. In retail, the reality is more volatile. A markdown recommendation is a complex schedule of depth, timing and cadence. It must work across categories, channels and countries while respecting margin floors, inventory pressure, and promotional rules. Above all, it must be executable in-store.

We weren’t only interested in the models students would build; we wanted to see how they would define the problem itself.

Where “optimal” starts to break down

In many academic cases, the objective is predefined: maximise X, minimise Y. Retail pricing is rarely that tidy. Markdown decisions involve objectives that often collide:

• Revenue vs margin: Accelerating sell-through usually requires deeper discounting, while protecting margin slows volume. Both paths are defensible depending on the seasonal goal.

• Inventory distribution: Stock in a Distribution Center behaves differently from stock in stores. End-of-line items require a different cadence than core carryovers.

• Legal and promotional constraints: Discount architecture and claims are not infinitely flexible. A strategy that requires weekly re-ticketing in 200 stores is simply not practical, regardless of its mathematical elegance.

The challenge is deciding what “optimal” actually means when different stakeholders define success differently.

Framing before modelling

We deliberately provided only a broad guideline —optimising revenue— rather than a fixed objective function. We wilde students to choose their own constraints. Would they prioritise margin protection, stock clearance, revenue acceleration, or simplicity of execution?

“We were surprised by the variety of solutions the students presented. By giving them the freedom to deviate from standard guidelines, each team spotted a different pricing challenge and tackled it with completely fresh eyes.” - Igor Custodio João, Senior Consultant – Pricing, ADC

In retail pricing, framing determines the outcome as much as the algorithm. The chosen constraints shape the solution space long before any code runs. A strategy that looks efficient in Excel can collapse once operational realities enter the picture. Conversely, a commercially viable strategy might look less “pure” mathematically.

By letting students define the frame themselves, we forced ourselves to look at the assumptions embedded in our own work.

Making assumptions visible

This collaboration was a structural exercise. Letting external thinkers work with retail data forces us to articulate choices we normally take for granted. It highlights what we are truly optimising for this season, and which trade-offs are intentional versus historical leftovers.

In pricing, the framing step matters as much as the model itself. Staying sharp in a competitive retail environment requires the logic behind the algorithms to remain explicit and open to challenge. Even the most experienced team has blind spots. Sometimes, the fastest way to see them is to let someone without your history look at the same data.