Real İBB traffic · 250 orders/day · 22 days back-tested
Most route planners assume the city moves at one speed all day. Rotaist forecasts tomorrow's Istanbul traffic hour by hour and plans around the rush — then we drive every plan on the traffic that really happened.
Watch a real day ↓Back-test
Every dot is one planner on one real January day: driver hours spent vs. minutes delivered late. Rotaist (blue) sits next to the oracle; ignoring traffic (dark grey) falls off the chart.
Real arrival minus the ETA the customer was told. A static free-flow plan is systematically late.
At equal or better punctuality, Rotaist's savings grow as the fleet gets more room to plan.
Median speed across Kadıköy, Üsküdar and Ataşehir as % of free flow, every hour of every day. The model trains on the left, is tested on January.
Average street speed on a held-out day: reality vs. forecasts.
On-time rate on each held-out weekday (5 vans). One bad forecast day is visible — and real.
Under the hood
İBB average speed for every ~1 km cell, every hour — 398 cells × 117 days.
33,604 OpenStreetMap segments inherit their cell's congestion, hour by hour.
Gradient boosting predicts tomorrow's congestion from data up to tonight.
Our optimiser knows 3 km takes twice as long at 18:00 and sequences around it.
Every plan is driven on the traffic that actually happened that day.
Orders are synthetic (drawn from real OSM residential floor area); traffic and streets are real. Data: İBB Açık Veri Portalı (İBB Open Data License) · © OpenStreetMap contributors (ODbL) · tiles OpenFreeMap · code MIT.