A window is a promise. Make one you can keep.
Flat 30-minute windows quietly break trust every time the operation misses them. The window model runs before the promise is made - a live simulation of what the assigned rider will actually be able to do, on the roads they are actually driving, with the queue they are actually joining.
A window per order, not per store.
Every order carries its own predicted window, generated at assignment time and refined as conditions change. Two orders leaving the same pickup point five minutes apart will not necessarily share a window - the second one is joining a different fleet state.
- Point estimates with a confidence band, not a single hopeful number
- Automatic tightening or widening as the rider progresses
- Escalation to dispatcher when confidence drops below your threshold
- Configurable per client or per SLA tier
Window life for one order
Assigned
Order goes to R11. Path takes them through two known slow segments.
Window set
Predicted delivery 12:34 - 12:41. Point estimate 12:37. Confidence 92%.
Refined
Traffic on segment 2 clearer than modelled. Window tightens to 12:33 - 12:38.
Watched
Pickup queue depth grew at neighbouring hub. Not on this order's path - window unchanged.
Delivered
Recorded at 12:35. Inside window, inside confidence. Fed back into corridor history.
What the window is actually built from.
Live segment speeds
Per-road-segment speeds refreshed continuously, weighted by how recently they were observed.
Pickup queue depth
How long the assigned rider will wait at origin - a genuinely large factor for single-store or dark-store operations.
Fleet load
How busy the wider fleet is - a heavily loaded fleet is more likely to reassign, which affects the confidence band.
Corridor history
What that specific route actually took at that time of day, learned from your own operation over recent weeks.
SLA tier
Whether the promise is a hard window, a target window, or an ASAP - drives how narrow the band should be.
Parcel type
Signature-required, fragile, or refrigerated parcels carry handover-time overheads the window has to include.
How accurate is accurate.
Confidence, not certainty
Windows come with a confidence figure. A 92% confidence means 92 in 100 orders with a similar shape landed inside the predicted band across your historical operation.
Cold-start behaviour
For a brand-new service area, the model leans on the routing layer and widens confidence bands until enough corridor history exists to tighten them. This is usually two to three weeks of live volume.
What happens on a miss
A missed window is a first-class event - the reason is recorded (traffic, pickup delay, rider stall, reassignment cost) and shown in fleet analytics so the model, the routing, or the SLA can be adjusted.
Next in the loop
A window at risk should not sit at risk. See how orders get reassigned before it becomes a miss.
Show us your fleet size. Watch the swarm move at your scale.
Type a rider count and a rough daily order volume. The view on the right rebuilds to that scale, with a sample reassignment playing every few seconds - the same kind of decision the delivery engine makes for a live fleet.
An illustrative view based on typical patterns - not a live read of your actual fleet. Numbers move with your inputs so you can see the shape of the problem the engine solves.