Problem
Megamarket's marketing team planned promo-code mailings by expert judgment and the results of earlier campaigns. A promo code is a discount such as 1000 rubles off an order of 4000. The business sets a GMV or new-customer target and a budget, and wants to hit the target without overspending.
An expert plan cannot show whether the same result was possible more cheaply. It is hard to revise when conditions change, and it ignores the current state of segments and their contact history.
Example from the article: with a budget of 5 million rubles and two promo types, 500 off 1000 and 1500 off 2000, the team must decide how many codes of each type to send.
Solution in brief
Treat the plan as an integer linear program: choose how many promo codes of each type to send, maximizing expected GMV within the budget (or minimizing budget for a target GMV). Expected conversion, GMV and cost per user come from historical segments. A what-if calculator lets marketers check and adjust any plan.
Questions this case answers
- How to allocate a fixed budget across discount types with integer unit counts to maximize expected value, using conversion history by segment
- How to find the cheapest allocation that reaches a target outcome and estimate its result before launch
- How to plan promo-code mailings that maximize GMV under a fixed budget
- How to find the cheapest promo plan that reaches a target GMV
Tools and methods
- Integer linear programming (knapsack view) — budget is the knapsack capacity, promo codes are items with expected value and cost
- Python-MIP — the optimizer implementation (Pyomo and SciPy named as alternatives)
- Historical segment heuristic — what-if calculator fills the best-converting segment first
- Conversion, GMV and cost per segment — inputs taken from past campaigns
Google or a one-time email code. No password.
