How to Diagnose Through a Customer Who Is Your Only Eyes
Why this matters
Sometimes there is no visit coming. The property is out of your service area, the customer is overseas, the site is seasonal with only a caretaker on it, or the equipment is on a boat, a farm, or a rental you will never stand in front of. You still have to produce something useful: a decision about whether to shut it down, a part to ship, a trade to refer, or an honest "this needs hands on it." Working through untrained eyes is a real skill with real rules, and the biggest one is that you stop asking questions and start assigning observations.
Step 1: Set the safety boundary before you ask for anything
Say the limits out loud, in the first minute, before curiosity takes over. A customer who wants their system working will do more than you asked, and they will do it while holding a phone.
- Fuel or gas smell, anywhere in the conversation: stop. Everyone leaves, from outside call the gas supplier or emergency services, nothing gets switched on or off, no lights, no phone use inside. Diagnosis resumes when the property is clear and someone qualified has made it safe.
- Water and electricity together: nobody touches the equipment or anything metal near it. Power gets isolated at the panel by someone competent to do it, or not at all.
- Anything that stores energy: capacitors, pressure vessels, springs, raised or suspended loads, hydraulic accumulators. Covers stay on. No exceptions for "I just want to look."
- Height, confined space, or hot surfaces: the observation is not worth the fall or the burn. If it needs a ladder or a crawl, it waits for a trade.
Then give the affirmative version, because a bare list of prohibitions makes people improvise: "Everything I ask for can be done standing on the floor, covers on, without touching anything hot or live. If I ask for something that does not fit that, tell me and we will find another way."
Step 2: Decide what you are actually being asked to produce
You are not producing a diagnosis. You are producing a decision, and there are only about five of them: run it, shut it down and stop using it, ship a part and have someone fit it, send a local trade with a defined scope, or gather more evidence and talk again.
Name the target before the questions start, because it sets how much certainty you need. You can and should shut something down on a suspicion. You should not ship a part on one.
Step 3: Build a picture of the thing you cannot see
Five minutes here saves twenty later. Before any fault questions, get the physical layout.
Ask for photographs, not descriptions: the whole unit from a step back, the data plate, the immediate surroundings, and anything connected to it. Ask in that order, because a data plate photo without context leaves you guessing which of two similar units you are looking at. Then have them list what is connected, in their own words. Without that mental model, "the pipe was hot" is meaningless, because you do not know which pipe.
Step 4: Convert every question into an observation they can perform
This is the whole method. An untrained customer is unreliable at interpretation and surprisingly reliable at four narrow things: binary states, counts, timings, and side-by-side comparisons. Ask for those and nothing else.
| Instead of asking | Assign this observation |
|---|---|
| "Is it running normally?" | "Start a stopwatch when it kicks on and tell me the time when the noise changes." |
| "Is there a light on?" | "How many lights are on the panel, and which numbered one from the left is lit?" |
| "Is it hot?" | "Hold your hand near it, not on it. Can you hold it there comfortably, yes or no?" |
| "Is it leaking?" | "Put a dry paper towel under it, leave it 30 minutes, then tell me if it is wet and where." |
| "Is the noise bad?" | "Count how many times the sound repeats in one minute." |
Every one of those returns a number or a yes/no, and every one of them can be repeated later to see if something changed. Interpretation stays yours.
Step 5: Sequence so each observation splits the field
Do not run a checklist. Pick the observation whose two possible answers eliminate the most, then pick the next one based on what came back.
The first split is nearly always the same: does the system attempt to operate and then fail, or does it not attempt at all? Nothing at all points at supply, control input, or a lockout. Attempts and fails points at the working parts, and it means the system is proving to you that most of the chain is intact.
The second split is usually timing: does it fail immediately, after a consistent period, or at random? Consistent timing is a gift, and step 7 explains why.
The third split is usually recovery: does it come back on its own, after a wait, or only when someone does something. That separates thermal from latched from momentary without you seeing anything.
Step 6: Refuse conclusions, accept only observations
Customers hand you conclusions constantly. "The pump is bad." "It is the thermostat." "The motor is burnt out." Treat each one as a lead, never as evidence, and immediately convert it: "What did you see or hear that told you that?"
Half the time the underlying observation is genuinely useful and the conclusion attached to it is wrong. "The motor is burnt out" often decodes to "there was a smell," a real observation pointing at several possible components. Take the smell, discard the conclusion.
Be equally strict on yourself. "Would you say it sounds like grinding?" gets a yes from most people regardless of the sound. Ask what it sounds like, wait, and let them reach for their own comparison.
Step 7: Get the fault to repeat and time it
A single event gives you a story. Three timed repetitions give you data. If the fault can be safely reproduced, and often it can because starting a system is not dangerous, have them run it three times with a stopwatch and report each duration.
Consistent durations mean a physical quantity is being consumed or accumulated: a volume, a charge, a temperature rise. Wildly varying durations mean something intermittent, and that you are unlikely to close this remotely.
Step 8: Keep a written remote fault log
Remote diagnosis happens across days and often across different people at the property. Memory will not carry it. Keep one running record on the job: every observation, who made it, when, in their exact words alongside your translation, and which possibilities it eliminated.
Without it you re-ask questions you already asked, which burns the customer's patience faster than anything else, and you lose the thread that a different person answered differently, which is often the actual clue.
Step 9: State your confidence and what would change it
Close every remote contact the same way: what you believe, how sure you are, what you are recommending, and what single new piece of information would change your mind. Put it in writing.
"I am confident this is on the fuel side. I am not confident which part of the fuel side. I am recommending you shut it down and stop relying on it, and that a local trade attends with these three things in hand. If it turns out to run fine for an hour on the next attempt, that changes my answer completely and I want to hear about it before anyone buys anything."
That paragraph separates a professional remote opinion from a guess, and it protects you: a remote conclusion that later proves wrong is defensible when the confidence and the limits were stated up front.
The worked example
A seasonal property several hours away, a caretaker on site, and a standby generator that starts and then dies after a couple of minutes. No visit is possible for at least a week.
Safety first, and it is not optional here. Fuel plus electricity plus exhaust. The caretaker was told: do not open the fuel system, do not work on it while it is running, stay clear of the exhaust because the gas it makes can kill in still air, and do not touch the transfer switch. Everything from that point was done standing back, covers on.
Baseline first. Photographs of the unit, the data plate, and the surroundings, plus the caretaker's description of the fuel supply.
Then the splits. It cranks and starts, so supply, starting circuit and ignition are proving themselves. It runs and then dies, so this is a running fault, not a starting fault.
Timing next, three repetitions with a phone stopwatch. The runs came back as 2 minutes 10 seconds, 1 minute 52 seconds, and 2 minutes 5 seconds. That is 130, 112 and 125 seconds, totaling 367 seconds, so an average near 122 seconds and a spread of 18 seconds. Eighteen seconds on a 122 second mean is about 15 percent, which is tight clustering for a stopwatch in an untrained hand.
Character of the failure: "does it stop dead or does it sag first?" The answer was that it sags and then quits, over a second or two. A dead stop points at an electrical cut or a protective device. A sag points at the engine being starved of something.
Recovery: it restarts immediately, with no wait and no reset, and then runs the same two minutes again. That eliminates the thermal family outright. A heat-driven fault needs cooling time before it will run again, and this one needs none.
Load: it does the same thing with no load connected, which eliminates the electrical side of the machine, because the fault does not care whether it is making power.
History and change: last season it ran its weekly test for an hour at a time without complaint, and over the winter the caretaker refueled from cans stored since the previous year.
Put those together. A consistent, repeatable run duration with a sag at the end, immediate restart, no thermal recovery needed, indifferent to load, and a recent change of fuel from long-stored containers. That is a fuel delivery limit, and the most likely members of that family are a loading filter or fuel that has taken on water during storage.
What was not concluded remotely: which one. Both produce the identical symptom set from the caretaker's side. So the recommendation stopped at the family: shut it down and do not rely on it for backup, do not let the caretaker touch the fuel system, and dispatch a local trade with a fuel filter, water-detection paste, and the expectation of draining and disposing of the stored fuel properly. The stated flip condition: if it ever runs materially longer than two minutes, or dies dead instead of sagging, the whole conclusion is back open.
What doing this wrong looks like
The common failure is shipping a part on a family-level conclusion. The reasoning felt strong, the customer wanted action, and a part in a box feels like progress. Then the part does not fix it, the customer has paid for freight and a fitter, and your credibility for the rest of the job is gone. Ship parts on member-level certainty, not family-level.
The second is building three more questions on top of an interpretation. If observation two assumed observation one's conclusion, and observation one was wrong, everything after it is fiction. That is why the log keeps the raw words next to your translation.
The third is asking for something unsafe by accident, usually through a question that implies access. "Can you see whether the connection looks discolored?" sounds harmless and is an invitation to open a panel. Ask what is visible with everything closed, and say so.
How to verify you got it right
Predict something checkable and then check it. If your reasoning is sound you can usually name a second symptom the customer has not mentioned yet: "you will probably find it is worse when it is asked to work hard." A correct prediction from a remote read is strong confirmation. A miss means go back to the log.
After a local trade attends, record what was actually found against what you concluded. Remote work only gets better if the outcome goes back into the file, and shops that skip this stay exactly as accurate at year five as they were at year one.
References
- OSHA guidance on hazardous energy control and on carbon monoxide exposure from fuel-burning equipment
- Manufacturer documentation for equipment shutdown, lockout, and fuel-system service
- Trade-standard practice for remote support and evidence documentation
- See related: The Limits of Diagnosing Anything Over a Phone Call
- See related: What You Can and Cannot Conclude Without Being On Site