Should window cleaners use AI to price jobs? · An honest 2026 answer

There's a lot of noise about AI in window-cleaning software right now. Some of it is real, most of it is packaging. This guide is a plain-English answer to what AI actually does in a quote flow, where it helps, where it fails, and how to spot the difference between "we use AI" as a genuine capability and "we use AI" as a marketing sticker.

I built the AI into Squeegify so bias upfront. I've tried to write this so it's useful even if you decide to use a competitor — the questions to ask are the same either way.

What AI actually does in a window-cleaning quote widget

The useful part of AI in a quote flow is one thing: counting windows off imagery.

A customer types their address. The tool pulls Google Street View + Satellite images of that property. A vision model (in Squeegify's case, Claude Haiku for the primary read and Claude Sonnet as a second-opinion pass on higher tiers) looks at the façade and returns "I can see 12 windows on the front and side, this appears to be a semi-detached 3-bed, medium property size."

That's it. Everything else in the quote flow — the pricing, the frequency rules, the add-ons, the follow-up — is deterministic code the operator controls. The AI does one job well and stays in its lane.

Where AI actually helps

- The customer doesn't have to count windows. Most people don't know how many windows their house has and don't want to walk around counting. AI removes that friction. - Consistency between quotes. Two operators quoting the same property will often give different window counts because they're eyeballing it differently. AI gives the same number every time (which the customer can then adjust — see below). - Speed. A vision read takes 8-15 seconds. Faster than walking around the house. Faster than a phone call. - Complexity flagging. A good vision pass will notice things like "this is a Grade II listed Victorian terrace with sash windows" and mark the quote for manual review, which stops the operator from under-pricing a complicated job.

Where AI fails

Honest list:

- Non-standard properties. Barn conversions, glass extensions, houseboats, anything with mixed roof heights and unusual glazing — AI window counts are unreliable. - Bad imagery. If Google's Street View was captured five years ago and the customer's added a conservatory since, the AI sees the old house. If the property has no clear façade image at all, the AI can't count. - Interior vs exterior. Vision models can only see what Street View can see — so any interior-cleaning quote is guessing. - Bay windows. Should a bay count as one window or three? The AI has to pick, and often picks wrong for your specific pricing model. - Difficult access. The AI can't tell whether there's a locked side gate, a steep drive, an overhanging tree, a nervous dog. All of those change the job.

Why "customer-confirmed" is more important than "AI accuracy"

The most important design decision in any AI-powered quote widget is: does the customer see and confirm the number before the price locks?

If yes, the AI is a helper. It gives the customer a sensible starting point, they nudge the count up or down based on what they know about their own house, and the price adjusts. Errors get caught before the operator sees them.

If no, the AI is a liability. Every mis-count becomes an angry email or a "you quoted £X but this'll be £Y actually" phone conversation.

Any tool that says "our AI is 97% accurate" without letting the customer confirm the count is selling you a problem. The right framing is "our AI is a helpful first guess and the customer always has the last word."

Squeegify does this via a big "12 windows · adjust if wrong" step between the AI count and the quote. The customer can tap +1 or -1 as many times as they like. Whatever they land on is the number the price uses. Boring, works.

What to ask before trusting an "AI window cleaner" tool

Six questions that separate real from marketing:

1. What model does the AI use? If the answer is "proprietary" or "custom," that usually means "we call GPT/Claude and add a system prompt." Fine, but pay accordingly. 2. Can the customer edit the window count? If no, walk away. 3. What happens if the AI can't count? A good tool falls through to a "please tell us your window count" step. A bad tool guesses and hopes. 4. Does the AI touch the price? It shouldn't. The AI counts. The rules price. Two separate systems, both auditable. 5. How much of the AI cost do I pay? Some tools bake AI cost into the flat monthly. Some meter it. Neither is wrong; you should just know. 6. Where does the AI run? European privacy-conscious operators may care about which data centre the images pass through. Most providers now offer EU-region inference; ask if it matters to you.

The economics

A well-designed AI window count costs roughly $0.005 to $0.015 per quote across current models. At £24.99/month for a Growth-tier subscription including 100 quotes, that's about £0.50-1.50 of AI cost against £24.99 revenue. Comfortable margin.

If a tool is charging £50/month and using the same models under the hood, you're paying for the interface and the CRM, not for the AI. Which is fine — just know what you're buying.

What AI won't do in 2026 (despite the marketing)

- It won't replace your judgement on difficult properties. For the top 10% of jobs, you'll still walk around. - It won't magically get you more customers. AI in the quote flow reduces friction; it doesn't create demand. Marketing still matters. - It won't write your customer emails better than you can. AI-generated follow-ups feel AI-generated. Customers can tell. - It won't price your round for you. The rates you charge come from your costs, your market and your competition. No AI can pull that out of thin air.

Where AI is quietly useful outside of quoting

The one place where AI has genuinely reduced my admin burden isn't quoting — it's job notes. Before a difficult clean, the AI writes a short "this property looks like it has these access considerations" note based on the Street View imagery. It's not always right but it's often useful, and it's the difference between arriving at a job with zero context and arriving with a 30-second read-in.

Also useful:

- Access-risk flagging — "this property has a first-floor conservatory, be careful placing ladders" - Job-time estimation — "medium property, ~35 minutes for a first clean" - Confidence signals — "we're 60% confident on window count, worth calling before committing"

None of these replace an operator's judgement. All of them save 30 seconds here and there, which adds up over a round.

The verdict

For UK window cleaners in 2026, AI in a quote widget is worth having if — and only if — the tool:

1. Uses AI for window counting and property assessment 2. Always lets the customer confirm the count 3. Doesn't touch your pricing rules 4. Falls through gracefully when imagery isn't available 5. Charges you honestly for what the AI costs

If those five conditions are met, it saves you 3-4 hours a week and converts more visitors into leads. If they're not met, you're paying for marketing.

If you want to see a widget that does all five in practice, the Squeegify homepage has a live demo. If you want the theoretical guide without the pitch, hopefully the above is useful regardless.

Questions? robin@squeegify.co.uk.