AI search and recommendations tuned for how Nigerians shop
Semantic search and personalised recommendations across millions of products, tuned for Nigerian spelling, brands and budgets, served in under 100 milliseconds on existing infrastructure.
The client
Jumia — e-commerce, Lagos, Nigeria.
The challenge
Keyword search missed the way customers actually type: local spellings, brand nicknames and budget phrases like "cheap android phone". Recommendations were generic, and any change had to run within the existing infrastructure budget.
What we did
A Business AI integration package rebuilt search and recommendations on the existing catalogue.
- Semantic search with typo tolerance and a Nigerian product vocabulary built from search logs
- Personalised recommendations from browsing and purchase history with cold-start rules for new visitors
- Evaluation set of 5,000 real queries with expected results, reviewed with the merchandising team
- Vector index and caching tuned to serve under 100 milliseconds on current infrastructure
- A/B testing framework and dashboards for conversion, basket size and latency
Results
Conversion from search rose 23% and average basket size 14% within a month of roll-out, with search responses under 100 milliseconds and no increase in infrastructure spend.
More ai integration work
An AI support assistant that resolves most routine requests
A support assistant inside the app and on WhatsApp that answers from Kuda's own help content, resolves card, transfer and statement requests and hands complex cases to agents with full context.
Read the case study →
Document AI that cut claims processing time by 70%
Document AI that reads claim forms, receipts and police reports, checks them against policy rules and routes clean claims for approval, leaving assessors to handle only the exceptions.
Read the case study →