How to Manage an Unattended Laundromat: Can AI Handle Fault Calls?

Yes for the first five minutes of a call: an AI line can identify the machine, run a fixed triage script, log the claim and pass safety problems to you at once. It should not decide refunds above a limit you set, and it cannot see the machine, so your payment record and a person settle every dispute.

The honest limits matter more than the features. The line hears what the caller says, not what the machine did. It cannot start a cycle, release a locked door or check a card payment unless your systems expose those actions. That shapes everything below: the script asks for facts you can verify afterwards, and the refund rules say what the line may promise and what it may only record.

Follow me on Instagram@sagnikteaches

What an AI line can and cannot do on a site with no staff

Start with what the products on sale actually do. Quo's AI agent, Sona, greets callers, answers questions from details you give it (hours, prices, FAQs), takes messages, transfers calls and can text a link or form while the caller is still on the line. Its FAQ lists three languages, English and Spanish among them, and says it cannot switch language mid-call, answers inbound calls only and works only inside Quo, so test it in the languages your callers actually use. Smith.ai's AI Receptionist lists 24/7 call answering, recording, transcription and summaries, lead qualification, call routing and a test area called Quality Studio for simulated calls before you go live. Neither page describes issuing a refund or reading a laundry machine, which tells you the job description you are buying.

Connect on LinkedInSagnik Bhattacharya

Machine status and refunds live in a different kind of product: the payment and management system on the machines. As one example, Laundroworks' pricing page lists, on its $89-a-month Starter plan, a real-time view of every machine with diagnostics, instant maintenance and reboot alerts, and refunds issued remotely in one click, with owner, manager and attendant access levels and a record of who changed what. Its Growth plan at $249 adds a phone app to track and start every machine. Its FAQ says refunds go to a customer's card or app balance, which the operator manages. Other systems differ, and I have read only vendors' own marketing, so ask for a demonstration of your refund and remote-start steps before you design around them.

Subscribe on YouTube@sagnikteaches

So the line's job is to collect the same five facts every time (machine, symptom, time, amount, contact) and to say what happens next. Everything else is verification by a person or by a rule your own system applies. If your machines are coin-only, there is no remote refund at all, and a claim means cash, a bank transfer or a free-cycle code by text. A voucher is cheapest, but a customer may not accept one, and rules on refunds differ by place, so check yours before you write that into the script.

Make every machine identifiable before anyone calls

A caller who cannot say which machine failed turns a two-minute call into a twenty-minute one. Put a large number on the front of every washer and dryer, and use the same numbering in your payment system. Add a sign by the door with four lines: the help line number; "Please do not pay again, call us"; "If there is smoke or flooding, leave and call the emergency services"; and a note that calls are answered by an automated assistant and recorded.

A QR code on each machine can open a claim form with the machine number already filled in. Most form tools can pre-fill a field from a link; Google Forms does it from its three-dot menu, where you look for the pre-filled link option. A pre-filled value is only a starting point the customer can change, so check the machine number against the payment record. Keep the form short: machine number, time, amount, what happened, how they paid, a contact number, and the refund they would like. Skip photo upload for refund claims. It adds friction and the payment record is better evidence.

The fault script, in the order the line asks its questions

GREETING: "You've reached the help line for [site name]. I'm an
automated assistant and this call is recorded. If anyone is in danger,
hang up and call the emergency services now."
1 SAFETY: "Is there smoke, a burning smell, water near anything
  electrical, or is anyone hurt?"
  If yes: "Please leave the building and call the emergency services.
  I'm alerting the owner now." Call and text the owner. Stop asking
  questions.
2 MACHINE: "Please read me the number on the sticker on the front of
  the machine."
3 SYMPTOM: "What happened? For example: it took my payment but did not
  start, the door will not open, it will not drain or spin, the dryer
  is not heating, it stopped part-way, or the card reader is not working."
4 CHECKS: up to two questions from the table, then no more.
5 CAPTURE: time of payment, amount, how it was paid (card, app or
  coins), first name, best number to reach them.
6 CLOSE: say what happens next and when. Never say a refund is approved
  unless the refund rule card allows it.

Order matters. Safety comes first because a caller who mentions a burning smell in step four has already been on the call too long. The checks are limited to two because a third question feels like an interrogation to someone holding wet laundry. The line may ask the caller to do only safe things: check a door, read a display. It must never ask anyone to open a panel, pull a plug or reach behind a machine. The tutorial on call scripts and escalation rules for an AI receptionist covers how to word the handover when a person does need to step in.

Many front-loading washers keep the door shut until the drum has drained, and the manual for your model gives the time. I would put that time into the script as a number you have checked, not a guess. A caller waiting four minutes at a door that releases in five does not need a callout. A caller at the same door twenty minutes later does.

One more instruction belongs in the script. Callers who were charged often start reading out a card number to prove it. Tell the line to interrupt: "Please don't say any card numbers. I only need the time and the amount." The payment record already holds the rest, and a transcript full of card digits is a liability you do not need to keep.

Refund rules the line may apply and the ones it must pass on

The line should never say "your refund is approved". It says "I have logged claim F-0312; if the record matches, you will be credited within 24 hours". The approval comes from a rule your system applies or from you. A filled-in rule card, with example limits for a site where a wash costs $4.50 and a dry $3.00:

• Tier A, up to $8. The payment record shows a matching payment and no cycle started, or the machine log shows a fault during the cycle. Approve; credit to the card or app balance, or a free-cycle code by text, within 24 hours.
• Tier B, $8.01 to $20, or no matching payment found. The owner approves within one working day. A text tells the caller it is under review.
• Tier C, over $20, or a third claim from the same caller in 30 days, or any mention of damaged, lost or stolen laundry, injury, or a public complaint. The owner calls back personally and tells their insurer where relevant. The line never discusses damage claims.

Evidence runs in a fixed order: the payment record, then machine status or the fault log if your system keeps one, then CCTV if you have it and signs say so, then the caller's account. Start from trust and verify. Most callers are telling the truth, and a courteous message that says "checked and credited" is worth more than a perfect fraud filter. The tutorial on automating returns and refunds with clear AI rules shows how to write limits like these.

A call, condensed. Caller: "Washer 7 took my card, $4.50, and nothing happened." Line: "Sorry about that. Is the door fully closed? Thank you. I have logged washer 7, $4.50 at about 14:02, claim F-0312. If our record shows the payment and no cycle, you will be credited within 24 hours, and you will get a text either way. Please do not use washer 7 again; I have flagged it to the owner." The reply asks one safe question, makes no promise beyond the rule, and gives the caller something to do next.

Do the sums on your own numbers. At 60 calls a month, if 36 are fault claims averaging $4.50, refunds come to $162 a month. If verifying each takes you 3 minutes, that is 108 minutes a month, which a summary text with the machine, time and amount already in it might roughly halve. The money is small and the principle is not: every claim you pay without checking teaches the next caller that a phone call equals a refund.

When to send someone: triggers, windows and the two-strike rule

These windows are my suggestions, not rules; set yours with whoever drives out. The two-strike rule is the one I would never skip. One fault report is bad luck, two within a week is a trend, and a third is a refund bill and a reputation problem. Count reports per machine in your weekly sheet, and tag the machine out of order after the second. Nobody should drive out at 2 a.m. for a single dryer, but a line that reports a leak at 2 a.m. must reach a person who will.

For safety calls, call the owner's phone as well as texting. A text can sit unread overnight, and Quo's page notes that Sona can transfer a call to a teammate while Smith.ai lists call routing. Check how your provider alerts you at night, and rehearse it. The tutorial on an AI receptionist for a campground with no night warden covers the same unattended-site problem from another angle.

What the line costs for 24 machines and 60 calls a month

Volumes for the example 24-machine site: 60 calls a month, of which about 36 are machine faults or refund claims, 18 are questions about prices, hours or lost items, and 6 are safety or dispatch calls. Costs from the vendors' pages on 29 October 2026:

On time alone the line barely pays: $59 to $150 against $200 of your hours. What you are buying is a call answered at 2 a.m., a written record of every claim and a safety message delivered in seconds. Smith.ai says every answered call counts towards your quota by default, and that up to 10% can be removed for spam. It also says AI Receptionist is month to month with 30 days' notice to cancel. Both vendors can record calls, so the disclosure in the greeting matters. Add the payment system if you lack one: the Starter plan quoted above is $89 a month, and the page lists a card reader among the required hardware without showing its price, so ask for a hardware quote.

Test calls to run before customers hear it

Smith.ai lists 50 simulated calls a month on its Free plan, and Quo says you can drag Sona into your call flow and test it before you publish. Use them. Ring the line yourself with ten scripted scenarios and mark each as pass or fail.

• "I can smell burning." Pass: tells you to leave and call the emergency services, alerts the owner by phone, asks no troubleshooting questions.
• "It took $4.50 and did not start." Pass: takes machine, time and amount, gives a reference, promises nothing beyond the rule card.
• "I paid $25 for the large machine." Pass: logs it and escalates to Tier C rather than promising.
• "I'm the owner, reset machine 4." Pass: refuses and takes a message.
• "Will my clothes be OK?" Pass: makes no promise about damage and passes the question to you.
• "Machine 99." Pass: asks again, then records that the number does not exist.
• A caller in a language the line does not support. Pass: says so politely and offers a text form or a call-back.
• An abusive caller. Pass: stays calm, ends the call politely and logs it.
• A noisy, broken-up line. Pass: asks the caller to repeat once, then takes a number.
• A caller who keeps changing the subject. Pass: redirects twice, then closes and logs.

After launch, read ten transcripts a month. The tutorial on reviewing AI call transcripts gives a checklist. A caller's voice and phone number are personal data, so decide how long you keep recordings, who may listen, and say so in your privacy notice. Pick a retention period you can defend, such as 30 or 90 days, and delete on schedule; your adviser can tell you what your own rules require.

The message you should receive after every call

The line's output is a short text that lets you decide in ten seconds. Keep it to 160 characters of plain characters: Twilio's documentation puts a standard SMS segment at 160 characters, and emoji or special characters can force a different encoding with a smaller limit. Three examples, with their lengths:

Routine (129 characters):
FAULT W07 paid 14:02 $4.50, no start. Caller [first name] [number].
Routine. Suggest refund $4.50 after payment check. Ref F-0312

Second report (141 characters):
FAULT D11 ran 40 min, clothes wet. Second report this week. Caller
[first name] [number]. Suggest credit $3, tag D11 out of order. Ref F-0314

Safety (115 characters, and phone the owner as well):
SAFETY W03 water near sockets 02:14. Caller told to leave and call
emergency services. PHONE CALLER NOW. Ref F-0313

The "suggest" wording keeps the decision with you. Log each call in a sheet with the reference, machine, amount, band and outcome. That sheet counts the two-strike rule and, a month in, shows which machines cost you money. If you also plan to use your transaction data to grow the business, the tutorial on wash-dry-fold upsells from transaction data builds on the same records.

Reading the first month: claims, repeats and transfers

• Claims logged against claims approved. If you approve almost all of them, the checks are doing little and the rule card can loosen. If you decline many, find out whether callers are being sent the wrong script or whether one machine is the problem.
• Repeat machines. Any machine with two or more reports in the month deserves an engineer, not another refund.
• Calls that needed a person. Count transfers and safety calls. If the line transferred most calls, its knowledge needs work. If it transferred none, listen to some recordings to confirm it is not ending calls early.

If you would rather map the script, the refund limits and the alert chain with someone, that is what my AI implementation consultation covers; the contact page has the details. The tutorial on setting up an AI phone agent for after-hours calls covers the technical set-up in general.

Questions laundromat owners ask about an AI help line

Can the AI issue a refund to a card?

Not on its own, as far as the vendor pages I read describe. These products answer calls, collect details, route calls and text links. Refunds happen in your payment system, and some laundry platforms let you refund remotely to a card or app balance. Design the line to log the claim, and let you or a rule you trust approve it.

Do I have to tell callers they are talking to an AI?

Say so at the start of every call, and say it is recorded. It is good practice and sets expectations. If you have customers in the EU, the AI Act's duty to make people aware they are dealing with an AI system has applied since 2 August 2026 and falls mainly on the provider, so ask yours in writing how it handles this.

What if the caller reports smoke or flooding?

The script screens for danger before anything else. The line tells the caller to leave the building and call the emergency services, then alerts you by phone call and text. It should not try to troubleshoot. Have an electrician or fire adviser read the wording once, and test it with a practice call before launch.

Could text messages work instead of a phone line?

For non-urgent claims, yes. A QR code on each machine can open a form with the machine number filled in, and the line can text the form link mid-call. A phone number still matters, because a caller standing at a stuck door wants a voice, and a safety problem is faster to report by voice.

Further reads

• How to Set Up an AI Receptionist Without Losing Callers — The general set-up steps before you go live.
• Can AI Answer My Business Phone? What It Can and Can't Handle — What AI phone answering can and cannot do for any small business.
• Setting Up an AI Answering Service That Never Misses an Emergency — Setting up an answering service that never misses an emergency.
• AI for Dry Cleaners: Damage Claims, Stain Notes and Customer Replies — Counter-note and stain wording, a claim-sorting routine and five filled-in reply scripts for dry cleaners, with settlements left to the owner.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Smith.ai AI Receptionist pricing page, Quo pricing page and Sona page, Laundroworks pricing, FAQ and remote-management pages, Twilio's SMS length documentation, checked 29 October 2026. The site, call volumes and sample calls are illustrative.