How to evaluate AI for field service
The short answer
Ask eight questions, then run a pilot on your own jobs. The questions tell you whether the tool can work for you. Only the pilot tells you whether your technicians will use it.
Eight questions for the vendor
- What does it answer from: one system's records, our documents, or the linked history of each asset?
- Does it show the source of each answer?
- How does a technician reach it: an app, a messenger, a phone call?
- Does it write back to our system of record, or only recommend?
- Which features are generally available, and which are in preview or beta?
- Do the AI features work without a connection?
- Which plan, add-on or licence does it need?
- What happens to our data, and which certifications are finished?
How to run the pilot
One team, your own jobs
Connect the tool to your real work orders. Five to ten technicians for two weeks is enough.
One number
Pick first-time fix, callbacks or time spent on reports, and record it before you start.
Count use, not opinions
How many technicians used it in the second week matters more than what they said in the first.
Signs it will not work
The vendor cannot say what the answers are based on. The demo only runs on the vendor's data. The technician has to type a full question into an app. The answer never reaches the work order.
On the last question in the checklist, hold us to the same standard: Kenner's ISO 27001 certification is in progress and its SOC 2 audit is in preparation.
Questions and answers
How long should a pilot run?
Long enough for the novelty to wear off. Two weeks shows whether technicians still use it once nobody is watching.
Should we choose the AI in our suite or a specialist?
Start with where the missing information is. If it is in the suite's work order history, the suite's AI will find it. If it is in manuals, old jobs or a colleague's memory, that is the case for a specialist.