Retail & Media
Tier-one merchandising platform - planogram optimization
Category arrangement optimizer ran in 6–10 minutes per scenario, blocking interactive planning sessions
The challenge
A provider of image recognition tech and market data services for tier-one manufacturers needed to dramatically reduce optimization run time from 6–10 minutes to near real-time so merchandisers could test scenarios during live planning sessions. They also needed to improve sales uplift from prior baselines through better alignment between store characteristics and target outcomes.
What we built
The team upgraded a genetic algorithm based optimizer that recommends category arrangements by size and adjacency. By strengthening selection, mutation, and randomization operators, and by reframing the fitness function to better match store features with sales targets, run time dropped from minutes to near real-time and uplift improved materially. The optimizer now recommends commercially useful layouts rather than purely academic optima.
What changed
Optimization speed
from 6–10 minutes to 30–60 seconds
Sales uplift
improved from ~3% baseline
Interactive what-if scenario testing in planning sessions
Built with
Genetic Algorithms · Constrained Optimization · Image Recognition · Real-time Processing · Production ML · Fitness Function Design