The economics of the second visit
A return visit costs travel, labour and a slot that could have served another customer, plus the customer's time and goodwill. It is several times more expensive than getting it right first time.
Most organisations track first-time fix as a metric and treat it as a performance issue for engineers. Analysis usually shows the causes sit upstream, in what was known at dispatch.
What actually causes a failed first visit
- The required part was not on the van and not ordered
- The engineer did not have the right skills or certification for that asset
- Access was not available - site closed, key holder absent, equipment in use
- The reported fault was not the actual fault
- A specialist tool was needed and not carried
- Preparatory work by another trade had not been done
Only one of these is about the engineer's competence. The rest are information problems, which means they are addressable with better prediction at the point of dispatch.
Predicting the parts
The most valuable prediction is usually which parts will be needed. Inputs are the fault description, the asset type and age, its service history and what has been used on similar jobs.
The output should be a ranked list rather than a single part. Carrying the top three candidates is often practical where carrying every possibility is not, and that alone lifts first-time fix materially.
Van stock optimisation follows from the same model. Knowing what each engineer's likely job mix requires lets you stock vans by area and specialism rather than identically.
The description is the weak link
Fault descriptions come from customers who are not technical, through a call handler working quickly. 'It is making a noise' is common and carries little information.
| Improvement | Effect |
|---|---|
| Structured triage questions by asset type | Much better input, small time cost |
| Photos or video from the customer | Often decisive for identifying a model or fault |
| Asset history surfaced to the call handler | Context the customer does not have |
| Suggested follow-up questions from the model | Improves the description as it is taken |
The fourth row is a neat application. A model that suggests the next question based on what has been said so far improves the input rather than only working with it.
Measure the right thing
First-time fix rate alone can be gamed - by declining difficult jobs, or by recording a partial fix as complete. Pair it with repeat visits within a period and with customer-reported resolution.
Also separate what is controllable. A visit that failed because the site was locked is a scheduling and communication problem, not a parts problem, and lumping them together hides both.
Most failed first visits were decided at dispatch, not at the door.