We sat down with Data Design’s Senior Advisor Pekka Autere to hear about his experiences from a project where machine learning and optimization were used to improve discount pricing.

Who are you and what do you do?
I’m Pekka Autere, Senior Advisor and Partner at Data Design. I work with many kinds of data and AI projects.

‍

‍You have long experience with machine learning and AI. Has any project been especially memorable?
There have been many, but one stands out: discount pricing. The project had been tried before but was left unfinished. This time we completed it successfully.

‍

‍What made the second attempt different?
The conditions were better. The organization was more willing to experiment and expectations were more realistic. Earlier, we analyzed results for too long without bringing them into practice. This time we implemented the solution directly, tested it, improved it, and achieved good results.

‍

‍What was the project about?
The idea was simple: could machine learning make discount sales smarter? The goal was to automate and standardize pricing and improve results while staying within the discount budget. We also looked at how campaigns affected inventory value and structure.

‍

What kind of solution was built?
We built a forecasting model and an optimization algorithm that suggested optimal discount prices at different stages. We also developed a user interface so users could adjust the recommendations if needed.

‍

‍How was the solution received?
Machine learning forecasts are never perfect. What matters is whether they work better than the old way. Users gave feedback on individual products, but overall the response was positive, especially because success metrics were defined clearly before testing.

‍

What challenges came up?
The main challenge was framing the problem so that the predictive model worked and remained stable with a large number of products. Data also varied between regions, which added difficulties.

‍

Any surprising findings?
Yes, one important lesson was about the scope of optimization. Pricing is usually done by department. But when we optimized the whole assortment together, the results were better than optimizing each part separately.

‍

‍Could the same solution work elsewhere?
Absolutely. The mix of forecasting and optimization fits many areas. In retail, for example, purchase planning, assortment design, and replenishment use similar logic. The same goes for hotel room and airline ticket pricing.

‍

What was the best part of the project?
The concrete benefits we could show, and the fact that the solution actually changed how people worked.