Loading…
The capstone of the Data Scientist track. Kasuwa, an online shop, loses about ₦40m a year to pay-on-delivery orders that fail at the door, and its head of operations wants "a model that stops bad orders". You'll frame it as a decision (whom to phone before dispatch), audit 67,000 orders for leaks, build customer history as it was known at checkout, and validate models over time, testing only on orders the trial didn't touch. Then you'll check calibration, use a randomised trial of confirmation calls to measure what a call actually prevents at each level of risk, and set the threshold where calls pay for themselves. Finally, you'll check the model city by city (it badly misses Kaduna, a market it has never seen), decide whether it should use location at all, write the model card, and plan monitoring that would catch the problems a single drift number hides. Every number comes from running the code.