Think Build Implement Repeat
London, UK +44 7367 067226
WhatsApp FOLLOW f in X
  1. Home
  2. Blog
  3. Predicting Service Time, Not Just Volume
AI & Machine Learning

Predicting Service Time, Not Just Volume

Capacity depends on duration as much as count, and most planning uses an average that fits nobody. Where duration prediction pays and what drives it.

Updated 2 min readBy SpiderHunts Technologies

Free estimateNo obligation

Get a free estimate

Tell us what you need. A senior engineer reads every enquiry.

Takes under a minute. We never share your details.

  • Free consultation
  • No commitment
  • NDA on request

Prefer to talk? Book a free 30-minute call →

Quick answer — TL;DR

Planning on average job duration guarantees overruns and idle time in roughly equal measure. Duration is usually predictable from job attributes, the person doing it and the location - and better estimates improve schedules more than better volume forecasts.

The forgotten half of capacity

Capacity planning normally forecasts how many jobs, appointments or deliveries are coming. Actual capacity is jobs multiplied by duration, and duration is treated as a constant taken from an average.

The variation is usually large. Two jobs of the same type can differ by a factor of three depending on access, complexity, and who is doing the work. An average fits neither and the schedule fails in both directions.

What actually drives duration

  • Job type - the obvious one, and usually already used
  • The individual doing it - consistent differences between people, which planning rarely reflects
  • Location - access, parking, floor, whether the site is known
  • Customer or site history - some sites reliably take longer, for real reasons
  • Time of day and position in the run - the last job of a shift differs from the first
  • Whether preparation happened - parts present, site ready, previous work completed

The individual variation is worth handling carefully. Used to improve scheduling it is valuable; used as a performance metric it will be gamed, and the data quality will degrade within weeks as people learn what is being measured.

Predict the spread, not just the middle

A schedule built on median durations runs late whenever jobs exceed it, which is half the time. Some slack has to exist somewhere, and the useful question is where to put it.

Schedule elementBasis
Individual job slotMedian, or slightly above
Buffer between jobsSized from variability of that job type
End-of-day protectionAccumulated uncertainty across the run
Emergency capacityHeld back separately, not absorbed into slots

Putting the slack at the day level rather than padding every slot generally produces more throughput, because unused padding in one slot cannot help a later overrun but shared buffer can.

Where it pays most

The benefit is largest where jobs are scheduled back to back and an overrun cascades - field service, clinics, mobile repair, installation. One bad estimate at nine in the morning affects everyone that day.

It matters less where jobs are independent and queued, since variation averages out across the day. That is a useful screening test before scoping the work.

The data is usually already there

Job start and end times exist in most field service, clinical and dispatch systems. What is often missing is what the job actually involved, because the work description is free text.

Extracting structure from those descriptions is typically the enabling step - the same pattern as in most operational machine learning. The modelling is straightforward once the history says what was done.

Everyone forecasts how many jobs. Almost nobody forecasts how long they will take, and that is what fills the day.

FAQ

Frequently asked questions

The questions readers ask us after this guide.

Still have a question?

Ask us directly — a senior engineer will get back to you.

Ask about your project

Is it fair to include who does the job?

For scheduling accuracy, yes. Using it as a performance measure is a different decision with predictable consequences for data quality.

How much history is needed?

Enough examples per job type, ideally across different people and sites. A few hundred per common type is a reasonable start.

What about genuinely unpredictable jobs?

Predict the range instead and schedule them where an overrun does least damage - typically at the end of a run.

Does this help with quoting as well?

Yes. Duration prediction feeds labour cost, and quoting on averages systematically underprices the difficult jobs.

Keep reading

More on AI & Machine Learning

Start here

Want machine learning project details from us?

Tell us what you are trying to predict and roughly what data you hold. We will come back with an honest view on whether machine learning is the right tool, what the work would involve and a realistic cost range. If a spreadsheet would do the job, we will say so.

  1. You tell us what you needTwo minutes on the form, or a message on WhatsApp.
  2. A senior engineer reviews itAnd comes back with questions, a realistic range and an honest view on fit.
  3. Free 30-minute scoping callWe talk through scope, options and a realistic estimate — with no obligation.
Free estimateNo obligation

Talk to someone who builds this

Send a short brief and we will come back with an honest view and a realistic range.

Takes under a minute. We never share your details.

  • Free consultation
  • No commitment
  • NDA on request

Prefer to talk? Book a free 30-minute call →