Essential details of the third step.
1 - The retailer operated with MRP (minimum retail price) and RRP (recommended retail price) of its suppliers. Imprice automatically uploaded current sets of MRP and RRP, utilizing these values as permitted price range limits in pricing rules; MRP was a price limit from below, and RRP was a price limit from above.
During the pilot, pricing experts tested a hypothesis about the willingness of consumers to buy some items with relatively inelastic demand at a price higher than the RRP. For such items, pricing experts set higher upper limits of price range permitted for price optimization.
2 - In this case, AI-driven price optimization algorithms were learning and sensing the demand under higher complexity conditions.
The reason was the ongoing "noise", caused by varying promotions, like "discount for skincare brand XXX". The store's consumers were accustomed to these promotions, and actively responded to such ad activities. Sometimes the price of a SKU had raised, and then its sales increased sharply. Such an increase in sales was caused by promotions of other items, not by the SKU's price changing. In reality, consumers bought a lot of items with promo markdowns, and also they bought the SKU with raised price as a complementary good.
Thus, during demand sensing, price optimization algorithms had to correctly handle such "price paradoxes".
3 - Since the company rarely changed regular prices, there was no history of price changes for many items.
For such SKUs, the algorithms performed an expanded demand sensing, increasing the number of price experiments.