What AI actually does in field service today — reality vs the sales pitch

Call answering: the real deal, with a catch

AI receptionists that answer instantly, capture the problem, and book the visit genuinely work, and they are the biggest real shift in the industry right now. For a shop losing jobs to missed calls — and most small shops miss a third of calls during busy season — this is found money.

The catch is tone and trust. Customers learn quickly they are talking to a machine, and some hang up. The honest deployment answers fast, says what it is, and hands to a human the moment the call gets complicated. The dishonest one pretends to be Brenda.

Dispatch optimization: quiet, useful math

Reorder today’s stops to cut drive time, flag the job that is about to blow up the schedule, match the job to the tech who has the part. This is unglamorous optimization and it works well, because it is math, not judgment. It does not replace the dispatcher’s read on people — it gives them a better starting board.

Quote and follow-up drafting: keep a human on the send button

AI can draft a follow-up text or a quote summary faster than you can type it. That is useful. It also occasionally invents a price, a part, or a promise with complete confidence. The shops using this well treat drafts as drafts: a person reads and sends. The moment an unreviewed AI message reaches a customer, you have outsourced your reputation to autocomplete.

Photo diagnosis: a second pair of eyes, not a verdict

Reading a data tag — make, model, serial — is nearly solved and saves real typing. Diagnosing from a photo of a burnt board is genuinely impressive in demos and genuinely unreliable in attics: lighting, angles, and the dozen things a photo cannot capture. Use it as a hint that a human checks. Anyone selling it as diagnosis is selling liability.

The rule that sorts everything

Ask of every AI feature: does it make a person faster, or does it quietly decide instead of one? The first category — answering, routing, drafting, reading tags — is mature and safe to adopt carefully. The second needs gates: a named human approves before anything external happens. Software that cannot tell you who approved the AI’s action should not get your trust or your data.