How Zuma's AI Engine Works Across All 15 Services

One AI, Fifteen Use Cases

When most people hear "AI" in a consumer app, they think of a chatbot. Zuma's AI is something more structural — it's the connective tissue between all 15 services, learning from the entire Zuma ecosystem to make each individual service smarter.

1. Surge Pricing (Rides)

The ride-hailing surge engine evaluates supply-demand ratios every 90 seconds across every 500m × 500m grid cell in our operating cities. It adjusts per-km pricing in real time and sends push notifications to offline drivers when earnings opportunities exceed a configurable threshold. The model was trained on 18 months of Nairobi traffic data before launch.

2. Route Optimisation (Courier & Food)

When a food delivery driver picks up from a restaurant, the AI calculates the optimal route to the customer's address, accounting for current traffic (via TomTom Traffic API), building access rules (e.g. "no motorcycle delivery in Westgate precinct"), and the probability of the customer being home based on their historical order-time patterns.

3. Credit Scoring

Our credit scoring model runs a gradient boosting classifier on 40+ features derived from in-app behaviour. It achieves 84% accuracy at predicting 30-day loan repayment — meaningfully better than the 71% accuracy we benchmarked against traditional income-based scoring in our first-year test cohort.

4. Fraud Detection

Every transaction on Zuma is scored by our fraud model within 200 milliseconds of initiation. The model looks for: device fingerprint anomalies, velocity checks (too many transactions in a short window), geolocation inconsistencies, and network graph patterns (accounts that transact primarily with each other — a signal of circular fraud). Flagged transactions are held for manual review before processing.

5. Personalised Recommendations

The recommendations engine uses collaborative filtering to suggest restaurants, products, and gig workers based on what similar users in your area order. It's also time-aware — if you order coffee on weekday mornings but pizza on Friday evenings, those patterns are captured and used to surface relevant content at the right moment.