AI for Window Cleaning: What Helps a One-Van Business and What Doesn't

A one-van window cleaner gets value from AI in four places: quote replies built from a price list, invoice chasing, review requests and rebooking messages, for about $20 a month. Receptionist add-ons, route optimisers and photo-counting apps tend to fail a simple sum for one van: they cost more each month than the hours they win back.

The sum is the monthly price divided by what an hour of your time is worth. At $40 an hour, a $20 plan has to save 30 minutes a month and a $139 plan three and a half hours. Saved time only counts if it turns into something, such as an extra clean, a chased quote or an evening that ends at six, so be honest about where each hour goes.

Follow me on Instagram@sagnikteaches

The break-even rule for a business that earns on the round

Start with what your time is worth, because every tool below is judged against it. Illustrative figures: 320 regular customers on a four-weekly cycle, 16 houses a day and an average of $20 a visit. That is $320 a day over about eight hours of cleaning and driving, or $40 an hour. Yours will differ, so use your own takings and hours, and take off the van's running costs if you want to be strict.

Connect on LinkedInSagnik Bhattacharya

Outside the round, say there are six hours of admin a week, about 26 a month: two hours on enquiries and quotes, an hour and a half on invoices and chasing, an hour and a half on rebooking and rearranging days, and an hour on reviews and posts. Those 26 hours are the whole prize. If AI trims a fifth of them, that is about five hours a month, worth $200 at $40 an hour. Every price below gets read against that figure.

Subscribe on YouTube@sagnikteaches

The prices are Jobber's listed USD figures for one user as I read its pricing page on 1 November 2026, and ChatGPT's and Claude's pricing pages for the $20 plans (Claude Pro is $17 a month if billed annually). Check the live pages before you buy, because offers come and go.

Turn the table round and it gives you a ceiling for every job: the most a tool is worth is the hours it saves times $40. Ten quote replies a month at seven minutes saved each is 70 minutes, so a ceiling of about $47. Read every pricing page against its ceiling, not against its feature list.

Quote replies: price bands in, a checked reply out

New enquiries arrive while you are on the round, and the reply that goes out first has a head start. The fix is a price list the AI can read, kept in a Project or a pinned note, so a reply takes three minutes instead of ten. Write the list yourself, in bands, with the awkward cases spelled out:

Price list (four-weekly regulars; windows, frames and sills, ground and first floor):
2-bed terrace $15 | 3-bed semi $20 | 4-bed detached $28 | 5-bed or larger: visit first
Conservatory: sides $12 extra, roof $15 extra. Gutters, fascias, soffits: quoted separately.
First clean for a new regular, and one-off cleans: regular price x 1.5. Minimum $15.
Rules: quote "from" prices only. If the customer mentions a bay, an extension, roof
windows, a third storey or more than 10 windows, ask for a photo before confirming.
Never promise a day: offer the two days that have gaps in my diary (pasted below).

An enquiry arrives (illustrative): "Hi how much for windows in a 3 bed semi with a conservatory? We've got a big bay at the front too. Thanks". The prompt is the price list, the message, your free days and one line: "Draft a reply in my voice, under 90 words, following the rules." A typical result:

Hi [first name], thanks for getting in touch. For a three-bed semi on a four-weekly clean it's from $20, plus $12 for the conservatory sides and $15 if you'd like the roof done too. A big bay can add a few windows, so could you send a quick photo of the front? Then I can confirm the price before I come. The first clean is $30 because it's usually heavier, then it drops to the regular price. I have Tuesday or Thursday free. Which suits? [your name]

What you'd check takes under a minute: the arithmetic ($30 is the $20 band times 1.5), that the days offered are really free, and that the reply asks for the photo instead of guessing. This draft also exposes a gap in the price list itself: it applied the 1.5 to the house only, so a first clean with the conservatory sides comes to $42, while the same rule applied to the whole visit gives $48. Decide which you mean, write it into the list, and the next draft will follow it. The usual fix is tone, such as "we" when you work alone; one example reply in your own words, saved in the Project, cures that within a few messages.

The costly mistake is not a typo. It is a flat $20 quoted to a "3 bed" that has 14 windows. On site, a normal three-bed semi might take 25 minutes, which is $48 an hour for the clean itself; the 14-window house takes 55 minutes, which is about $22 an hour. One careless quote becomes a 12-month price you are stuck with, which is why the rules above force a photo and a from-price when a message mentions anything unusual. Pricing window cleaning per window and quoting from photos goes through that side in detail.

The time sum, illustrative: ten new enquiries a month, ten minutes a reply by hand and three with the draft, so 70 minutes saved. That alone clears the $20 plan 2.3 times over (70 minutes is 1.2 hours, worth about $47).

Round planning: what AI can slot in and what a map already does

On a stable round the order of streets hardly changes from one cycle to the next, so there is little to plan. The work comes from changes: a new customer to place, a cancellation leaving a gap, a rain day pushing 16 houses into the next. These are the planning jobs a general AI assistant can help with, as long as you feed it streets (never names or house numbers) and then check its geography.

Here are my days and the streets on each. I have a new regular on Beech Avenue.
Monday: Alder Road, Beech Close, Cedar Avenue (13 houses)
Tuesday: Dunmore Street, Elm Grove, Ford Lane (15 houses)
Wednesday: Gorse Way, Hill Rise (10 houses)
Which day should she go on? Give your reasons, list every pair of streets you are
assuming are close to each other, and say what you are unsure about.

The result, illustrative: "Monday, because Beech Avenue is probably near Beech Close." This is the classic slip: street names are not geography, and matching on a shared word is a guess. Because the prompt asks the AI to list the pairs it assumes are close, you can check that one pair on a map in two minutes. Suppose it turns out to be a 12-minute drive from Beech Close but two streets from Gorse Way: the customer goes on Wednesday, which also has the most gaps. The AI saved you the shuffling; the map made the decision.

Rain days are the other regular job. The AI drafts the text to the affected customers ("Hi [first name], the rain stopped me getting to you today, so I'll be round on Thursday instead. Nothing you need to do.") and a one-line order for the catch-up day: missed streets first, then the new day's list. Sixteen messages typed by hand might take twenty minutes; with a draft and a copy-and-paste each, perhaps eight (illustrative timings).

What about the paid optimiser? Jobber's help page says route optimisation is on its Connect, Grow and Plus plans, works on "anytime" visits only (visits without a fixed start or end time), covers up to seven days at a time and can be run only by admin users. For one van, moving from Core to Connect costs $70 to $90 more a month, so the optimiser must save 1.75 to 2.25 hours. Take a stable round with about 90 minutes of driving a day and assume the optimiser trims a generous 5 per cent: that is 4.5 minutes a day, 90 minutes over 20 working days, or 1.5 hours worth $60. It loses. The related tutorial on window cleaning round software covers the cases where it wins, such as inheriting a patchwork round of one-off jobs.

Invoicing and late payers: let the app send, let AI sort the mess

Sending the invoice or payment link is the field-service app's job, or your bank's, and it needs no AI. What goes wrong in a one-van business is the tail: nine customers a couple of weeks late, no time to decide who gets a text, who gets a call and who is simply a day behind. That sorting and wording is the part to give an AI assistant.

Export the overdue list, delete surnames and addresses, and add a notes column from your own memory. A cut-down illustration with five of the nine (the nine owe $280 between them):

You would overrule one line at once. The AI saw "28 days late" and suggested pausing visits to customer C, but your note says why the payment slipped, and the right move is a phone call to the daughter, not a pause. That is the pattern to expect: the AI sorts by the numbers you gave it, and you supply what it cannot know. The more the notes column holds, the better the sort.

Then ask it to write a three-step ladder in your voice. Day 3: "Hi [first name], a quick note that the $20 for the clean on the 4th hasn't come through yet. Pay by card or transfer using the link below, and thanks!" Day 14 names the visits and asks for a date. Day 28 says visits pause until the balance is cleared, and offers to talk if something has gone wrong. Send them yourself; don't let anything chase automatically until you have read a month of drafts. For wording by situation, the tutorial on cleaning invoices and chasing payment with AI has more.

The sum, illustrative: sorting nine customers and writing nine messages takes you about 30 minutes with a draft, against the hour or more it takes when it drifts to Sunday night. If the tidy-up recovers even two balances that would otherwise have drifted into write-off, that is $40 to $100, enough to cover the plan.

Review requests that go out because they cost one tap

A review request saves no time. It is on the list because it is cheap, and a new customer may well glance at your reviews before texting you. Save your Google review link as a text shortcut on your phone and send it after each first clean, then once a year to regulars. The AI's job is to write four or five versions to rotate, so your regulars don't all get the identical line, and to draft replies to what comes back.

One version (illustrative): "Hi [first name], thanks for having me today. If you're happy with the windows, a quick Google review helps a one-van business a lot: [link]. If something wasn't right, tell me and I'll sort it." The last sentence matters, and not for tactics. A message that invites both outcomes is honest, and it keeps you inside Google's rules.

Those rules are plain. Google's review policy says businesses may not offer payment, discounts or free goods or services in exchange for a review, may not selectively solicit positive reviews, and may not discourage negative ones. So there is no "get 10 per cent off your next clean for a review", and no asking only the customers who said the job looked lovely. Ask everyone the same way. Getting more reviews after every job with AI follow-ups sets out a longer routine.

For replies, give the AI the review and one real detail from the job and ask for three sentences. The realistic fault is sameness: ten replies that all begin "Thank you so much for your kind words". Edit each to include the detail (the conservatory, the dog that supervised, the first clean) and drop the "we" if you work alone. Never let it invent details; it will if you don't supply them.

The sum, illustrative: ask 30 customers a month, and if one in ten leaves a review that is three reviews. The time is about five minutes a week for the messages and replies, and it costs nothing beyond the assistant plan the quote replies already justify.

The skip list, with the sum behind each no

None of these tools is bad. Each carries a fixed monthly cost that a single van struggles to absorb, and each is easy to buy because the demo shows a result and the pricing page shows a number you can shrug at. Where I verified a price on the vendor's own page, it is in the table; where I didn't, the cell says to ask for the monthly figure before the feature list.

The receptionist deserves a second look, because it is the one with a case. Say twelve new people a month contact you, and you miss four because your hands are wet or you are up a ladder. If a quarter of those four would have become regulars at $20 a visit across 13 visits a year, that is one new regular a month, each worth $260 a year in visits, against a $29 monthly fee. But if you ring each missed caller back that evening and they book anyway, the add-on bought you nothing. Only the log can tell you which of those you are, and check that your plan can take the add-on at all: Jobber's help article says it can be added to select plans and doesn't list which. The tutorial on AI receptionists for trades, built in or standalone compares the options if the log says yes.

A ten-van firm with a dispatcher is a different sum, and so is a cleaner with commercial contracts where one missed call is worth thousands. The table is for one person who is the whole workforce. If you do move into that second group, rerun the break-even with the new figures.

Fitting it around a day on the round

A solo operator's problem is not finding AI uses; it is having 20 minutes at the end of a working day. Plan for that. Spend two hours once writing the price list, the rules and two or three example replies in your own words. After that, the routine is ten to fifteen minutes in the van or at the kitchen table: paste the day's new messages, read and edit the drafts, send. I would expect anything needing more than 30 minutes of weekly upkeep to be dropped within a month, so don't build it.

Keep customers' details out of consumer chats: first names and street names are enough for everything above. On the plans I looked at, ChatGPT's pricing page shows training on your content as an opt-out for its individual tiers, and Claude's shows none by default on Team; read the controls in whichever account you use. Keep the price list somewhere other than the AI too, in a note on your phone, so the business doesn't stop on the day you cancel or the service has a bad hour.

A 30-day test, then keep or cancel by the numbers

Run one tool at a time for 30 days and write down three things: what you paid, minutes saved by job, and anything it got wrong. Here is a filled-in log for the assistant plan (illustrative):

• Cost: $20.
• Quote replies: 14 at about 7 minutes saved each, so 98 minutes.
• Invoice chasing: three sessions saving about 40 minutes in total.
• Total: 138 minutes, or 2.3 hours, worth about $92 at $40 an hour.
• Errors caught: two quotes that needed a photo before confirming; one reply that said "we".
• Decision: keep. It cleared its $20 and the ceiling comfortably.

For anything dearer, set the stop rule before you start: if it hasn't covered its cost by day 30, cancel on day 30. Pay monthly for the test even though annual billing is cheaper. Jobber Core, for instance, is $29 a month billed annually against $49 monthly, which is $240 a year, but only worth locking in after the test has passed.

The receptionist test is the missed-call log above, run for two weeks before you spend anything. Everything else on the skip list passes or fails on the same arithmetic: write the monthly cost, divide by $40, and ask whether you can name the hour or the job that pays for it. If you can't, it is not for this van yet.

Questions a solo window cleaner asks about AI

Can ChatGPT or Claude quote a window clean from photos?

They can count obvious windows and spot a conservatory, but they often miss frosted panes, roof lights and anything hidden by a fence or a tree. Use the photo to ask better questions, quote a from-price, and confirm on the doorstep before the first clean. A miscounted first visit costs far more than the minutes the quote saved.

Is a $20 AI plan enough for a one-van business?

For quote replies, invoice-chasing wording and review requests, yes, and the free tiers of ChatGPT and Claude may cover a solo round on their own. Both free plans let you keep your price list and rules in a Project; a paid plan matters once you keep hitting message or upload limits. Buy nothing else until a month of use shows what is missing.

How do I check whether missed calls are costing me jobs?

For two weeks, note every call or text you could not answer, whether you replied the same day, and whether they booked anyway. If nearly all booked, a receptionist add-on adds little. If several went elsewhere, work out what a regular customer is worth over a year and compare that with the add-on's monthly price.

Can I offer a discount to customers who leave a review?

No. Google's review policy bans offering payment, discounts or free goods or services in exchange for a review, and it bans asking only the customers you expect to be happy. Ask everyone the same way, thank them either way, and reply to each review you receive.

Further reads

• Window Cleaning Price Increase Letter: Text and Leaflet Versions With AI — Text and leaflet wording for raising regular prices.
• How to Get Window Cleaning Customers: Leaflets, Local Search and AI — Leaflets and local search for new regulars.
• How to Price Commercial Window Cleaning: Per-Pane Quotes From Your Labour — Pricing shopfronts and offices from your labour.
• AI Receptionist for Tradespeople: What It Costs and When It Pays — What a call-answering add-on costs and when it pays.
• Free vs Paid AI Tools for a Small Trades Business — Free tiers against paid plans for a small trades business.
• The Intake Questions That Make AI Cleaning Quotes Accurate — The questions that make a cleaning quote accurate.
• Replying to Trade Enquiries in 60 Seconds With AI — Reply to new enquiries before they ring the next cleaner.
• Commercial Window Cleaning Contract: Proposal, Access and Safety With AI — Survey notes become a scope table, access and safety clauses, frequency pricing and an insurance section, checked five ways by AI before a lawyer sees them.
• Upholstery Cleaning Business: Jobs AI Can Take Off a One-Person Round — Booking-form fabric questions, photo-based quotes, aftercare notes and rebook texts a one-person upholstery cleaner can hand to AI, and what stays with you.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Jobber pricing page and help articles (Receptionist, route optimisation), claude.com/pricing, chatgpt.com/pricing, and Google's review policy page, all read on 1 November 2026.