How Dry Cleaners and Laundries Use AI for Orders and Collections

Dry cleaners and laundries use AI in four places around an order: an AI phone or chat agent that books pickups and answers "is it ready?", automatic ready-for-collection texts, a reminder ladder for garments nobody collects, and AI-drafted replies to damage or lost-item claims. Start with whichever of those eats most of your counter time.

The part AI cannot do is the garment-level detail at intake: the stain on the left cuff, the missing button, the silk lining that needs hand finishing. If those notes are vague on the ticket, every automated step after them inherits the vagueness, and the claim you lose three weeks later started at the counter.

Follow me on Instagram@sagnikteaches

Where an order loses time between counter and collection

Before choosing any tool, walk one order through your shop and note where staff stop to look something up, call someone back or apologise. In most dry cleaners the same six points come up.

Connect on LinkedInSagnik Bhattacharya

The stages are in order for a reason. Each one depends on the data captured in the stage before, so fix them from the top.

Subscribe on YouTube@codingliquids

Stage 1: Make intake notes consistent (one to two days)

An AI agent can only tell a caller what the ticket says. Agree a short code list with your team and stick it by the till: ST for stain (with location), BT for button missing, TR for tear, BL for colour bleed risk, HF for hand finish, CL for customer's own label instructions. Most dry cleaning systems let you save preset notes, which is faster than typing.

Where staff prefer to talk, dictation on a phone plus a chat assistant turns rambling notes into the format. A prompt that works:

Turn these spoken intake notes into our ticket format.
Format per item: [item] | [colour] | [brand if said] | [codes] | [location of any mark]
Codes: ST stain, BT button missing, TR tear, BL bleed risk, HF hand finish.
Do not add anything that was not said. If a detail is unclear, write "CHECK".

Notes: "navy two-piece, wool I think, bit of red wine on the right sleeve,
jacket's missing the bottom cuff button, trousers fine"

What came back from that prompt in a test (illustrative):

Jacket   | navy | wool  | ST, BT | ST right sleeve; BT bottom cuff
Trousers | navy | CHECK | none   | none

Two things to fix. The assistant put "wool" in the brand column because the format had nowhere for fabric, so add a fabric field and mark it "customer says" rather than fact. And it logged the stain as ST without the words "red wine", which is the detail your cleaner needs to choose a treatment. Add "include the substance if said" to the prompt. The "CHECK" instruction itself is worth keeping: without it, a chat assistant will happily guess a brand or a stain type, and a guessed note is worse than no note when a claim arrives.

Stage 2: Let an AI agent take pickup bookings and status calls

For a shop with a van, this is usually where the hours are. The two laundry systems I checked both sell an agent that works inside their own software, which matters because it can read the ticket rather than just take a message.

• CleanCloud AI Voice. Its voice agent, called Grace, captures pickup details, confirms time slots, creates the order and can take payment, checks order status, signs up new customers and transfers to staff with an audio summary of the call so far. CleanCloud's plans start at $99 a month (Lite, 500 orders, no pickup and delivery) and $199 a month (Pro, 2,000 orders, with pickup and delivery, route optimisation and a driver app). The pricing page does not list AI Voice, so ask what it adds to your plan.
• Cents Assist. A $99-a-month add-on to Cents that takes orders by phone and text, answers FAQs, runs around the clock with unlimited calls and texts, and escalates to a person during business hours. Cents' own plans start at $89 a month.

In a self-service laundrette with a wash-and-fold counter the call mix is different: "are the big machines free?", "what time do you close?" and "can I drop off a duvet today?" come up more than order status. Cents' self-serve plans already include machine monitoring, so an agent that reads it can answer the first question honestly; if yours cannot see the machines, tell it to say so and give the busy and quiet hours instead of guessing.

If you are on another system, a general AI receptionist can still take pickup requests and messages; the limit is status calls. For the difference between answering-only and booking agents, see how an AI receptionist handles bookings.

Give the agent a tight brief. Things it should never do: quote for specialist items, promise same-day service after your plant cut-off, accept a pickup outside your delivery zones, or discuss a damage claim. Writing those limits down is most of the setup work, and call scripts and escalation rules for an AI receptionist covers the wording. Allow a few hours to fill in FAQs and zones, then a week of test calls from staff mobiles before customers hear it.

Those test calls earn their keep. In one illustrative run, a member of staff rang at 11.40am and asked for a suit back the same day. The agent said, "Yes, we offer same-day service, so it'll be ready by 5pm." The FAQ said "same-day service available" and nothing about the 10am plant cut-off, so the agent had no way of knowing. The fix was one line in the FAQ ("same-day only for items in by 10am, Monday to Friday; after that, next working day") and a rule to offer the next available time rather than a yes or no. Run the same test again after every FAQ change: agents are literal, and any promise your staff make with a caveat needs the caveat written down.

Specialist items need a clean hand-over rather than a refusal. When a caller asks for a wedding dress to be collected, the agent should take the details and pass them on, not quote. A useful hand-over note reads: "Caller wants a wedding dress collected for cleaning and boxing; ivory, beaded bodice, wedding was last month; would like a price before collection; best time to call back after 2pm." The person who calls back starts the conversation where the caller left it, instead of asking everything again.

Once pickups are booked automatically, the van round is the next bottleneck. Route optimisation comes with CleanCloud's Pro tier and with Cents' top Ultimate tier ($399 a month, which adds in-house fleet management); if you run deliveries outside either, route planning for small delivery fleets compares the options.

Stage 3: Ready texts people act on

This is automation more than AI, and it is cheap: Cents' entry plan includes order-ready notifications in 10 languages, and CleanCloud includes SMS on every plan, though its SMS credits cost extra. The wording is where shops lose collections. Compare these two.

Weak:   Your order is ready. Thank you.

Better: Hi [first name], your 3 items (ticket 4471) are ready to collect.
        We're open today until 6.30pm. Balance due: $24.50.
        Reply STOP to opt out.

The better version answers the three things people otherwise ring to ask: which order, when you are open, and how much. Use a chat assistant to write two or three variants and any translations once, read each aloud, then let the system send them. There is no benefit in generating a fresh AI message per order, and a template that goes wrong goes wrong for hundreds of customers at once, so a person signs off every template change.

Stage 4: A reminder ladder for garments nobody collects

Uncollected items tie up rail space and represent work you have done but not been paid for. A ladder of messages, sent by the system on fixed days, recovers a share of them without anyone on the phone.

1. Day 0: the ready text.
2. Day 10: a friendly nudge with opening hours and the balance.
3. Day 30: "Your items are moving to storage; they're safe, and here is how to collect."
4. Day 60: a final notice that quotes your ticket terms on uncollected goods.

The day-10 nudge works best when it sounds like the shop rather than the system. Filled in for an illustrative ticket: "Hi [first name], your two shirts and grey suit (ticket 4502) are still here and looking sharp. We're open until 7pm on Thursdays if that's easier after work. $31.00 to pay on collection." Naming the items jogs the memory far better than a ticket number alone, and mentioning the late opening answers the usual reason people haven't come.

Wording for the day-60 message, to adapt:

Hi [first name], we've been looking after your [items] (ticket [number]) since [date].
Under the terms on your ticket, items uncollected after [period] may be
[donated / disposed of]. We'd much rather hand them back. We're open
[hours], or reply to arrange a delivery.

Two cautions. First, rules on what you may do with uncollected goods differ between jurisdictions and often require specific notice, so have a solicitor check your ticket terms and final-notice wording before you act on either. Second, the ladder is only as good as your collection scans: if staff forget to mark items collected, customers who picked up last week get a final notice. For an existing backlog, export orders marked ready for more than 30 days and ask a chat assistant to group them by value and last contact date, so staff phone the few high-value ones and let the ladder handle the rest. Strip names and phone numbers from the export first, keeping only the ticket number, so the assistant sees the orders and not the customers. An illustrative summary of a 140-ticket export:

Over $60 owed, no contact for 30+ days:      9 tickets  -> phone
$20-$60, one reminder already sent:          38 tickets -> ladder, day 30
Under $20, ready 60+ days:                   71 tickets -> final notice
Tickets with no phone number or email:       22 tickets -> CHECK at counter

The 22 without contact details are the ones no ladder can reach. Flag them in the system so the next time a regular mentions a surname or shows a receipt, the counter staff can match them. And exclude account customers, such as a hotel or salon with standing weekly collections, from the ladder altogether: a "your items are moving to storage" text to a business whose van comes every Tuesday reads as a mistake, because it is one.

Stage 5: Claims replies drafted by AI, sent by a person

A shrunken jumper or a lost shirt produces the angriest message you will receive all month. AI is useful for a first draft because it stays polite when you might not, but it should never be the one to send.

You are drafting a reply for a dry cleaner. Do not admit fault, offer money
or promise an outcome. Acknowledge the problem, restate the facts from the
ticket, explain the next step and give a date for a decision.

Our claims policy: [paste the relevant lines]
Ticket notes at intake: [paste]
Customer's message: [paste, with phone number and card details removed]

Write under 150 words, in plain English, signed by [first name].

A customer writes: "My cream wool coat has come back two sizes smaller. It was fine when I dropped it off. I want it replaced." Here is the kind of draft that comes back (illustrative):

Dear [first name],
Thank you for letting us know, and I'm sorry the coat has come back
in this condition. I've checked ticket 5120: the coat was logged as
cream wool, care label "dry clean only", no marks noted.
We will inspect it with our cleaner on Thursday and compare it with
the care label and our process record. I'll write to you with our
decision by Friday. Please bring the coat in, or reply and
we'll collect it.
Kind regards, [first name]

Mostly usable. The fix is the first sentence: "sorry the coat has come back in this condition" can read as accepting that you caused it, so change it to "sorry to hear about the coat". Also check the date it promised is one your cleaner can actually meet.

The intake notes from Stage 1 do the heavy lifting here. "ST red wine right sleeve, customer advised may not fully remove" changes a claim conversation completely. The same goes for lost-item claims. A customer who says six shirts went in and five came back is usually right or wrong in a way the ticket can settle: if intake logged six, the search starts at the plant; if it logged five and the customer signed or received the receipt, the reply can say so gently and offer to check the rails anyway. Paste the item count into the claims prompt along with the notes, and the draft stops guessing. For the tone of public replies when the complaint lands on Google instead, replying to negative reviews with AI has worked examples.

Illustrative numbers: two counters and one van

Take a dry cleaner and laundry with two counters and one van, handling about 900 orders a month and 45 phone calls a day. A week of tallying shows 60% of calls are status checks or pickup bookings, at about two and a half minutes each: roughly 68 minutes of counter time a day. If an agent handles 70% of those calls fully, staff get back around 47 minutes a day, mostly during the morning rush when it matters most.

On the rails, a count finds 140 items ready for more than 30 days at an average of $18 owed. If the ladder brings back a third of them, that is about $840 of paid work collected, plus the space. Against that, an add-on agent at $99 a month and some SMS credits. These figures are an illustration; your own tally in week one is the only baseline worth using.

Signs the setup is misfiring

• Pickups booked outside your zones. The zone list in the agent's settings is out of date or draws the boundaries differently from the route your van driver actually runs. Check the first 50 AI-created pickups against the map.
• Final notices to people who already collected. Collection scans are being skipped at busy times. Make the scan part of taking payment.
• Callers asking for "a real person" in the first ten seconds. The greeting is too long or hides the transfer option. Say what it can do in one sentence.
• Claims drafts that sound like admissions. Phrases such as "we're sorry we damaged" slip in. Keep the "do not admit fault" line in every prompt and read before sending.
• Texts arriving at 7am. Scheduled messages are running on the server's clock, not your shop's. Check the time-zone setting.

To check it worked after a month, compare three numbers with your week-one tally: calls answered by staff per day, items on the uncollected rail over 30 days, and days from claim to first reply. If the first two fall and the third shortens, keep going; if not, the problem is usually in Stage 1.

Dry cleaning and AI: follow-up questions

Can AI identify stains or fabrics from a photo at the counter?

A general image model can describe a visible mark in a photo, but it cannot tell you the fibre content under a coating or whether a tannin stain will shift. Treat any AI description as a prompt for the person at the counter to look closer, and keep the care label and your cleaner's judgement as the deciding step. The value of photos is evidence for later claims, not diagnosis.

Do I need to switch point-of-sale system to get an AI phone agent?

Not always, but it matters. Agents built into dry cleaning systems, such as CleanCloud's AI Voice or Cents Assist, can read order status and create orders directly. A general AI receptionist can take a message or a pickup request, but unless it connects to your system it cannot answer whether ticket 4471 is ready, which is the most common call.

Will regular customers object to an AI answering the shop phone?

Some will, so give them an exit. Say in the first sentence that they are speaking to an automated assistant, offer a transfer to staff during opening hours, and keep a named person for your account customers such as hotels or salons. If you serve customers in the EU, the EU AI Act's transparency rules now require that disclosure for chatbots and voice agents anyway.

Further reads

• How to Set Up an AI Receptionist Without Losing Callers — The setup steps that stop callers hanging up on a new phone agent.
• Cutting Parcel-Tracking Calls With AI Delivery Updates — How couriers cut 'where is it?' calls, which maps closely to van rounds.
• Can AI Answer My Business Phone? What It Can and Can't Handle — What any AI phone agent can and can't handle, beyond laundry.
• How to Measure an AI Receptionist's Return in the First 90 Days — A simple way to prove the agent is paying for itself.
• How to Keep Customer Data Private When Your Team Uses AI — Keep addresses and payment details out of the wrong AI tools.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: CleanCloud pricing page and AI Voice product page; Cents plans and pricing page (including Cents Assist); EU AI Act Article 50 transparency obligations.