How we measure

Last updated: 11 October 2026

Forecast accuracy: 92.6% by day and zone, 14 days ahead

Foresight was trained on 392 delivery days from one live operation and tested on 56 days held out of training. At day × zone it was 92.6% accurate, against 90.9% for the naive rule "same weekday last week". 92% of its forecasts landed inside their band, against a target of 95%. At day × zone × slot, counting noise sets a floor: a single slot holds only a handful of orders, so no model can be much better than 79% there.

Planning speed: typically under 5 seconds

Measured on the optimisation history of a demo workspace: 210 optimisations, 97.6% completed, an average solve time of 3.7 seconds. Runs of 51–200 bookings with flexible delivery windows averaged 2.5 seconds. Every solve has a time limit, between 5 and 300 seconds depending on its settings, so very large or tightly constrained problems take longer.

AgentQ's strategy recommendation

AgentQ recommends a strategy from your own optimisation history. Its confidence grows with that history: the demo workspace reached 81% on 52 completed optimisations of medium workloads.

The other figures

Up to 30% lower operating cost, 100% auto-allocation and traceability, a 70.4% cut in workload imbalance, disruption recovery from 45 to 3 minutes and a 10–25% margin lift are targets from our pilot and test work. Results vary by operation, which is why every pilot ends with an ROI report measured on your own routes: on-time rate, cost per drop, workload balance and the share of decisions made by AgentQ.

Ask for the method

For the dataset, baseline and method behind any figure, write to sales@omnioiq.com.