01
Does the answer say where it came from?
An answer without a source is an opinion. For torque specs, fluid capacities, and anything safety-adjacent, a tech needs to know whether they're reading the OEM manual, their fleet's standing practice, or a senior mechanic's override. If the tool gives you an answer and you can't tell which, it isn't built for the bay.
What good looks like
Every answer cites the source document and page number, or the name and date of the person whose note it is. Both, when both are relevant.
02
Can a tech use it without putting down the wrench?
If using the tool requires two hands and good lighting, it won't get used in the bay. Voice input and audio output aren't nice-to-haves for a shop floor. They're the difference between a tool that gets used and one that sits on the supervisor's desk.
What good looks like
Push-to-talk on any phone. Answer read back aloud. Works with background noise. No login screen between the tech and the answer.
03
Who controls what goes into the system?
An open system where anyone can add anything is a liability. A closed system where only admins can add content won't capture the knowledge you actually need. The right model is structured capture with a supervisor approval step — knowledge goes in from the people who have it, but nothing reaches a junior tech's answer until someone with authority signs off.
What good looks like
Knowledge capture is frictionless for the senior. Approval is required before anything goes live. Every note carries the author's name permanently.
04
Does it understand which vehicle is being worked on?
A generic answer about a Cummins L9 is less useful than an answer about the specific unit sitting in Bay 3 right now. Good knowledge software lets a tech scope a conversation to a specific vehicle, pulling that unit's make, model, year, and history into every answer it returns.
What good looks like
"Ask about Bus 247" returns answers specific to that unit's make, model, year, and any notes attached to its own record.
05
How long does it take to get running?
A system that takes six months to implement and a dedicated IT project to configure is not a realistic option for most fleet shops. The question to ask is how quickly a working workspace can be stood up with your own manuals and your own equipment roster. Days, not quarters.
What good looks like
Upload your manuals, tag your fleet, record a few voice notes with a senior tech. Working on day one. No implementation partner required.
06
What happens when two sources disagree?
The OEM manual says one thing. Your fleet's standing practice says another. A senior tech's note says a third. The right tool surfaces all three, labels each one, and flags the disagreement rather than silently picking a winner. The call belongs to the technician and the supervisor, not the software.
What good looks like
Conflicts are shown, not resolved. Each source is labeled by level. Disagreements are flagged in the answer and visible to the supervisor in the dashboard.
07
Where does the data live and who owns it?
Your OEM manuals, your fleet overrides, your senior tech's thirty years of fixes — that's your intellectual property. Before you sign anything, be clear on data residency, tenant isolation, and what happens to your data if you cancel the contract. For public sector buyers, ask specifically about compliance requirements.
What good looks like
Your data stays in your tenant. Isolation is enforced at the infrastructure level, not just by policy. You can export your knowledge base. SSO supported for enterprise.
08
Does the vendor understand a maintenance yard?
General-purpose AI tools can answer general questions. Fleet maintenance is not general. Mixed propulsion, equipment-specific service histories, OEM bulletins that contradict fleet practice, techs with greasy gloves and three minutes to find an answer — these are specific conditions. A vendor who can't speak to them in your first call hasn't built for your environment.
What good looks like
The demo uses fleet scenarios. The product handles OCR'd binder pages and voice input on a noisy shop floor. The founders take the first call.