Predictive Service Time

Predictive Service Time (PST) uses machine learning to estimate how long a driver will need at each stop, so your routes and ETAs reflect real-world conditions instead of fixed assumptions.

Why service time matters

Service time is hard to get right. How long a stop takes depends on what is being delivered, the structure of the building, the neighbourhood, the time of day, and more.

Small errors add up across a route:

• Underestimated service times put drivers behind schedule, delaying every stop that follows and frustrating both recipients and drivers.
• Overestimated service times waste capacity and create unnecessary waiting.

PST replaces static values and simple rules with a prediction for each individual task.

How it works

PST learns from your own delivery history. The model is trained only on your platform's data and is never shared with or influenced by other customers.

It considers factors such as:

• Item size and weight
• Floor number and elevator availability
• City and zip code of delivery
• Time of day and day of week

The model retrains every weekday (Monday to Friday), so it keeps improving as new deliveries come in.

A prediction is calculated whenever service time is needed: when a task is created or announced, when a task is updated, and when you request available time slots.

Each prediction comes with a confidence level: LOW, MEDIUM or HIGH. Only HIGH-confidence predictions are ever applied to your tasks. If a prediction has lower confidence, or cannot be calculated, your usual service time is used instead.

Getting started

PST is rolled out in two stages, so you can review predictions before they affect any tasks.

Stage 1: Activate PST on your platform

Contact your Urbantz representative to activate PST on your platform. Once activated:

• The model trains on your historical data and is ready on the next working day (activation on a Friday means predictions start on Monday).
• Predictions are calculated and shown alongside your tasks, but not applied. Your existing service time logic continues as usual.
• Your platform needs at least 50 tasks in the last 3 months for the model to train. More data means more confident predictions.

Stage 2: Apply predictions per hub

When you are happy with the predictions, you can start applying them hub by hub:

1. Go to Administration > Platform > Hubs and select the desired hub.
2. Open the Settings tab and find the Predictive Service Time card.
3. Enable Apply Predictive Service Time and save.

The change takes effect immediately for new tasks and time slot requests in that hub. Existing tasks keep their current service time. You can enable as many hubs as you like, one at a time.

Viewing and controlling predictions

Where to see predictions

Once PST is active on your platform, two new fields appear: Predictive Service Time and Predictive Service Time Confidence. You will find them in:

• The General tab of task details
• The task list, as columns
• Spreadsheet exports

When a prediction has been applied, the task's Service Time matches its Predictive Service Time exactly.

Recalculating a prediction

If you change details that affect how long a stop takes, you can ask PST to predict again. A recalculation is triggered when you update any of these fields: task type, dimensions, hub, location, contact, time window or service time.

To apply the new prediction:

• In the UI: keep Calculate service time automatically checked when editing the task.
• Via API: include recalculateServiceTime: true in the POST /task/{id} request.

Other fields, such as metadata, do not trigger a recalculation.

Using your own service time instead

You can skip the prediction for a single request and use your usual service time, even on a hub where PST is applied:

The prediction is still calculated and visible on the task, and the hub setting stays in place for all other requests. Changes appear in the task history like any other update.

Turning PST off

To stop applying predictions in a hub, go to Administration > Platform > Hubs, select the hub, open Settings, disable Apply Predictive Service Time and save. Predictions will still be calculated for review.

To stop PST entirely, contact your Urbantz representative to deactivate it on your platform. This stops all predictions, regardless of hub settings.

Frequently asked questions

I don't see the Apply Predictive Service Time toggle or the new task fields. PST must first be activated on your platform. Contact your Urbantz representative.

Can I enable a hub with little delivery history? Yes, but predictions there are likely to have lower confidence and won't be applied until the model has learned enough.

Can I apply LOW or MEDIUM confidence predictions? No. Only HIGH-confidence predictions are applied, and this threshold can't be changed.

Why does a prediction have LOW or MEDIUM confidence? The model may not yet have enough similar historical deliveries, or the task may be missing details that help it predict.

Why is the Predictive Service Time field empty on a task? No prediction was made for that task. PST may have been activated very recently and still be training, it may be turned off, or the prediction may have failed. In every case, your usual service time is used.

What happens to my original service time when a prediction is applied? It is still calculated and kept by the system, and is used automatically whenever a prediction can't be applied.

Are predictions applied to existing tasks? No. Predictions only apply to tasks and time slot requests created after a hub is enabled.

Why has confidence dropped compared to before? If your operations changed significantly (new areas, new product types), the model needs time to adapt. If nothing has changed, please contact Urbantz support.

Does PST slow down requests? PST adds a very small amount of processing time that shouldn't be noticeable. If you see an impact, please contact Urbantz support.