MaShop/Journal/Tools/The Estimate Your AI Wrote Is Now a Promise
● ToolsSeptember 21, 2026
Read · 5 min
ai written quote · quoting

The Estimate Your AI Wrote Is Now a Promise

A model drafting a quote fills every field, including the ones it cannot know. The invented lead time is still a lead time you offered to a customer.

Key takeaways
  • A model drafting a quote fills every field the template has, including ones it cannot know, and a lead time it invented is still a lead time you offered.
  • Under the Consumer Rights Act, a service contract that fixes no price is treated as requiring a reasonable price and no more, and what is reasonable is a question of fact.
  • That means silence does not protect you. It hands the number to somebody else to decide after the work is done.
  • The CMA published price transparency guidance on 18 November 2025 covering mandatory fees, drip pricing and partitioned pricing, and it applies from early advertising through to purchase.
  • The FTC fee rule is narrower than its reputation. It covers live event tickets and short term lodging, so most service businesses are outside it and still inside general deception law.
  • The fix is structural rather than legal. Name the scope boundary, the exclusions and the validity period, because those are the three fields a model guesses.

A plumber asks a model to turn six lines of notes into a quote. Thirty seconds later there is a tidy document with a scope of works, a price, a two week start date and a payment schedule. It reads better than anything written by hand at nine in the evening. Two of those four things were invented.

This is the most useful and most dangerous administrative task a service business has handed to AI. Useful because quoting is the bottleneck that decides how much work you win. Dangerous because a quote is not a description of a job, it is an offer, and an offer that is accepted becomes the contract you have to perform.

What is the model actually doing wrong?

Filling fields. A quote template has a lead time box, so it produces a lead time. It has a payment terms line, so it produces payment terms. Nothing in the process distinguishes between a number you supplied and a number that simply belongs in documents of that shape.

The three fields most often invented are the start date, the duration, and the exclusions. The first two come out as plausible industry averages. The third comes out empty, because your notes did not mention what is excluded and a generative draft rarely volunteers a limitation you did not ask for. An empty exclusions section is the expensive one.

We described the same failure on the wholesale side, where a drafted quote quietly commits to a lead time that ignores the kiln schedule, in what a maker should watch when a shop asks for wholesale terms. Service work makes it worse because the scope is discovered on site rather than counted in a warehouse.

Diagram breaking a defensible quote into a scope boundary, a stated total, named exclusions, a validity period and a variation clause

What happens if you leave the price open?

Somebody else decides it, later, with the benefit of hindsight. In the UK this is not a grey area but a statutory default.

Section 51 of the Consumer Rights Act 2015 applies where a consumer has not paid consideration, the contract does not fix a price or say how it will be fixed, and nothing incorporated into the contract sets one either. In that case the contract is treated as including a term that the consumer must pay a reasonable price for the service, and no more. The section then adds the line that decides most arguments: what is a reasonable price is a question of fact.

A question of fact means it is established by evidence about the circumstances rather than by your rate card. So a job that ran long because the joists were rotten does not automatically justify the final invoice. What justifies it is a record made before the work of what was included, what was not, and what would happen if the unexpected turned up.

Note

This is why an estimate with a variation clause beats a vague quote. An estimate that says how a change gets priced is more defensible than a firm sounding number with no mechanism for the thing that actually happened.

Does a quote bind you differently from an estimate?

Yes, and the words matter less than the structure. A document that reads as a fixed total for defined work behaves like a fixed price, whatever it is titled. A document that presents a projection and explains how the final figure is reached behaves like an estimate.

What decides which one you produced is whether the scope has a boundary. A quote for replacing a bathroom suite is a fixed price for a defined set of items. A quote for refurbishing a bathroom is a fixed price for whatever a customer later says refurbishing meant. The second is the one that ends in a dispute, and a model asked to smarten up your notes will happily produce the second while sounding like the first.

Three lines fix most of it. What is included, listed as items rather than as a description. What is excluded, named explicitly, including the discoveries you consider likely. And how long the price holds, because materials move and a quote with no expiry is a standing offer.

FieldWhat a model tends to produceWhat it should sayCost of getting it wrong
Start dateA plausible interval such as two weeksThe date you can actually start, or noneA promise made before you checked the diary
DurationAn industry average for the job typeYour own working estimate with a basisOverrun that reads as your failure
ExclusionsNothing, because you did not mention anyThe likely discoveries, namedUnpaid work on the least predictable part
Total priceA single confident figureThe total including all mandatory chargesAdd ons later that look like a tactic
ValidityOmittedA stated number of daysAn old price accepted after costs rose

What do the pricing rules now require?

That the number a customer first sees is the number they will pay, where you are able to calculate it. This tightened recently in the UK and it applies well before anybody signs anything.

The Competition and Markets Authority published its price transparency guidance on 18 November 2025. It covers what to include in pricing information, including mandatory fees, taxes and charges. It covers drip pricing, which it describes as prices being added as consumers proceed with a transaction. And it covers partitioned pricing, where the component parts of a price are given but the overall price the customer would pay is not.

Partitioned pricing is the one that catches trades and studios who thought they were being transparent. A quote listing labour, materials and a call out charge as three figures, without the total, is exactly that pattern. It feels more honest than a single number and it is the practice the guidance names.

Is the American position the same?

No, and the difference is worth knowing before you copy advice written for the wrong market. The rule everybody cites is much narrower than its nickname suggests.

The FTC's questions and answers on its rule on unfair or deceptive fees states that it applies to any business that offers, displays or advertises live event tickets or short term lodging, including platforms, resellers and travel agents. Live event tickets cover concerts, sporting events, theatre and similar, while pre recorded performances and film screenings are excluded. Short term lodging covers hotels, motels, vacation rentals and home shares, while long term rental housing in a landlord and tenant relationship is not covered.

So a plumber, a photographer or a consultant in the United States is outside that rule. What they are not outside is the general prohibition on deceptive practices, which has no industry list. The rule tells you what specific disclosure looks like when a regulator writes it down, and it is a reasonable standard to adopt voluntarily even where it does not bind you.

One detail from the rule is worth borrowing regardless of jurisdiction. The total price has to include all charges or fees the business knows about and can calculate upfront, including charges for mandatory goods or services bought as part of the same transaction. Knows about and can calculate is the right test to apply to your own quotes. A charge you always apply is not a variable, it is part of your price.

What may legitimately sit outside a quoted total?

A short list, and only if it is disclosed properly. The FTC rule is instructive here even where it does not apply to you, because it draws the line a regulator considered defensible.

Three categories may be left out of the upfront total price under that rule: government charges such as taxes, shipping charges, and optional ancillary goods or services. The conditions attached matter as much as the list. Excluded charges have to be disclosed before payment is requested, and the final amount has to be displayed as prominently as the total price.

Apply that to a service quote and the test becomes practical. A congestion charge or a permit fee set by somebody else is a genuine pass through. A materials delivery you arrange is arguably shipping. Your own minimum call out charge is none of those things, because it is not optional, not governmental and not shipping. If it applies to every job, it belongs inside the number you quote.

The word optional deserves a moment of honesty too. A service is optional if the customer can decline it and still get what they came for. Waste removal on a job that generates waste is not optional in any meaningful sense, whatever the line item says.

Does this change when you quote another business?

It changes the legal backdrop and not the commercial risk. Section 51 sits in consumer law, so it does not supply a reasonable price default to a business to business contract in the same way. What fills the gap there is ordinary contract interpretation, which is less forgiving rather than more.

That cuts against a common assumption. Trade customers are often thought of as the safer audience because they understand the work, and in practice a business customer is more likely to hold you to the literal document, more likely to have a purchase order process that overrides your terms, and more likely to pay late. A public sector buyer takes that further still, since what you promise in a tender answer becomes a contract term you are measured against. The quote is where you lose or keep control of all three.

So for a business quote, make the document the whole agreement rather than a friendly summary. State the total, the exclusions, the validity period and the payment terms in the same place, and say what happens on late payment rather than leaving it to be discovered. The credit side of that decision, including what to check before offering terms at all, is in what to establish before giving a business buyer thirty days.

Where does AI genuinely earn its place in quoting?

In the parts that are reading and restructuring rather than deciding. This is where the time actually goes for most service businesses, and it is the work models do well.

Turning a voice note from a site visit into a structured list of tasks, so nothing gets forgotten between the visit and the document. Comparing a new enquiry against three similar jobs you already priced, so you can see whether your instinct matches your own history. Checking a draft against a checklist of required fields and telling you which are empty. Rewriting a technical scope into language a domestic customer will actually read, which reduces the misunderstandings that become disputes.

Notice what those have in common. Each takes information you already hold and changes its shape. None of them produces a number out of nothing. The moment a model is supplying the figure rather than formatting it, you have handed over the only part of the quote that carries risk.

Card listing four things a model cannot know when drafting a quote, covering capacity, site conditions, supplier lead times and your walk away price

How do you build a template that resists this?

By making the risky fields impossible to fill silently. The trick is to design the document so that an invented answer looks obviously wrong.

Put required inputs at the top as explicit values: your earliest start date, your day rate, your minimum charge, the validity period in days. Then instruct that any field without a supplied value must be left as a marked gap rather than estimated. A document with three visible gaps is a document you can finish in two minutes. A document with three confident inventions is one you will sign without noticing.

Keep your exclusions as a standing list rather than writing them each time. Every trade has a set of recurring discoveries, and after twenty jobs you know yours. That list is the most valuable page of business writing a service firm owns, and it is the page nobody gets around to making.

Then read the total out loud before sending. It sounds trivial and it catches the specific failure that matters, which is a figure that is plausible for the job type and wrong for this job.

When should a quote be refused rather than written?

When you cannot see the work. A model will produce a confident document from a vague enquiry, and that capability quietly removes the friction that used to stop you quoting blind.

Two situations are worth declining outright. An enquiry that describes an outcome rather than a task, where the customer wants a room to feel bigger or a system to be faster, needs a conversation before a number. And any job where the decisive variable is hidden behind something, under a floor, inside a wall, in somebody else's codebase, is a job to quote in two stages: a priced investigation, then a quote for the work the investigation reveals.

The two stage approach is easier to sell than most people expect, because it is honest about the thing the customer already suspects. It also protects the relationship, since the alternative is a fixed price that has to be renegotiated at the worst moment.

What if the customer asks the quote a question?

Then you are in a different kind of exposure, because an answer given in your name can extend what you promised. A quote is a controlled document you reviewed. A reply generated in a chat window is not.

This is the same problem as an assistant on a storefront making a commitment about delivery or a refund, which we went through in what happens when your bot promises something you cannot deliver. The practical rule for a service business is narrow and easy to hold: anything that changes price, scope or timing goes through a person, even when everything else is automated.

The quiet economics of this

Quoting faster wins more work, which is exactly why the risk is worth managing rather than avoiding. A service business that returns a clear quote within a day of the site visit beats one that takes a week, and for most one person firms the week is caused by the document rather than by the thinking.

So use the tooling on the document. Keep the judgement, the capacity check and the number. And write down your exclusions once, properly, because the single cheapest protection available to a service business is a list of the things it does not include, and no model can write that list for you. It comes from your own jobs, which is also the case for the pricing and margin work we set out in measuring the return on automating one task. If you are building the quoting and ordering side yourself, keeping the templates and the data under your own control is part of the point of running the commercial side on code you own.

Discussion 0

0 / 4000Your email address is not displayed with your comment.
No comments are published yet.

Explore — related articles.

Build something. Move your work forward.

Start with a software project or an agent task. Describe the result you need, review the work and keep control of your connected accounts.

Open the workspace →