The quote gets cross-examined

The homeowner is still at the kitchen table when the estimate arrives.

It is a PDF. Six pages. Polite logo. A photo of a bathroom that is probably not theirs. The total is £18,600, which is both more than they hoped and less than they feared. There is a line for “labour”, a line for “materials”, a fixture allowance, a note about “standard installation”, and a sentence saying unforeseen works will be charged separately.

In the old version of this moment, the homeowner sighs, forwards it to their partner, maybe texts a friend who once renovated a kitchen, then replies with one awkward question about timing.

In the new version, they upload the PDF before replying.

The answer comes back as a nasty little checklist:

• Why is the permit responsibility not specified?
• What brand and model are included in the fixture allowance?
• Is disposal included or assumed to be arranged by the owner?
• Why is labour a single lump sum?
• What happens if the subfloor is rotten?
• Who decides when a change order has been triggered?

The contractor thought they had sent a quote. What they actually sent was evidence.

The second sales meeting

A quote used to be sales collateral optimised for a tired human under time pressure. It had to look credible, reduce anxiety, and get the buyer to say yes before the next contractor turned up.

That object now has a second audience: software that extracts, compares, flags, rewrites and asks questions without social embarrassment.

The model is not afraid to ask what “miscellaneous” means. It does not feel rude asking for the same clarification in a different way. It will happily spend ten minutes interrogating a £600 ambiguity because its time is free and the buyer’s money is not.

QuoteChecker is already built around this behaviour. A homeowner uploads contractor estimates and gets a line-by-line audit for vague scope, hidden fees, missing permits, inflated pricing, allowance traps and missing assumptions. The interesting part is not that it “uses AI”. The interesting part is the posture. It treats the estimate as an object to be inspected, not accepted.

The same habit is showing up in uglier categories. Granted Health lets people upload medical bills so AI can review them for errors and help reduce what they owe. Car buyers have been moving this way for years with pricing data, invoice explanations, dealer fee scripts and add-on comparisons. A dealer no longer negotiates only with the person in the chair. They negotiate with the person, their spreadsheet, their forum screenshots, and now a calm little procurement analyst in the browser.

Insurers have had this machinery for longer. In auto claims, CCC’s estimate review tooling lets carriers review repair estimates at line-item level, apply rules, flag anomalies and route reviewer actions. That used to be institutional audit infrastructure. Now a lighter version is leaking into everyday buying.

The pattern is simple: audit interfaces are escaping the back office.

The quote is now an interface

Calling this “AI negotiation” makes the category sound too theatrical. Most buyers do not want a robot haggling on their behalf. They want to avoid being the idiot who missed the obvious trap.

The better frame is hostile linting for commercial documents.

Software developers know linting as the annoying thing that catches errors before code runs. Missing semicolon. Bad variable. Weird dependency. Nobody mistakes the linter for the engineer. It just creates friction at the exact point where friction is cheaper than failure.

A quote auditor does the same thing to buying decisions. It scans for missing assumptions, ambiguous terms, asymmetries, suspicious units and clauses that become expensive later.

The ordinary workflow looks like this:

1. A quote, renewal, invoice, lease addendum, medical bill, event venue estimate or software proposal lands in an inbox.
2. The buyer uploads a PDF, screenshots, prior emails, policy terms, product pages, old invoices or competing quotes.
3. The assistant extracts line items, normalises units, spots missing assumptions and benchmarks charges.
4. It produces red, amber and green notes, then drafts the next action.
5. The buyer still clicks send.

That last step matters. The product is rarely a full agent closing a deal. The handoff is an email draft, portal message, support ticket, sales call brief or procurement intake note. The human remains the legal and social front end.

But the seller now faces machine-shaped suspicion delivered through a human mouth.

“Can you clarify whether disposal is included?” sounds like a normal customer question.

“Please separate labour into demolition, installation, plumbing, electrical and project management, and confirm whether permit fees are included or excluded” sounds like a customer who brought procurement to a bathroom fight.

Trust becomes parseability

The strange incentive is that more detail invites negotiation, but less detail starts to look like a red flag.

For years, vague quotes protected margin. A bundle let a seller hide risk, judgement, contingency, overhead, hassle and profit inside one rounded number. That was not always dishonest. Sometimes it was the only sane way to price messy work.

A bathroom renovation is not a bag of screws. A contractor prices sequencing, availability, subcontractor risk, callback probability, liability, experience and the pain of dealing with your building control officer. A medical bill has coding rules. A SaaS contract has seat minimums, support commitments, usage tiers and renewal mechanics. Some margin pays for knowing what will go wrong before it goes wrong.

Still, the commercial physics changes once the buyer has cheap inspection.

A vague quote used to protect margin. Now it creates suspicion debt.

The best seller will not always be the cheapest. The best seller may be the easiest for the buyer’s AI to verify without opening a dispute. Trust becomes parseability.

That does not mean publishing every supplier invoice or pretending judgement can be itemised to the penny. It means making the important parts inspectable:

• Scope in plain English.
• Exclusions in a dedicated section, not buried in prose.
• Allowance ranges with named examples.
• Permit responsibility.
• Change-order triggers.
• Optional tiers.
• Evidence links.
• Expiry logic.
• Risk separated from margin where possible.
• A sanctioned lane for questions.

The quote still has to persuade a human. It also has to survive extraction by a machine. Same object, two surfaces.

This is why the PDF is such a funny survivor. It is probably the worst possible format for interrogation, yet still the default because anyone can email it, print it, forward it and sign it. The next product move is not a prettier PDF. It is a quote portal with an exportable evidence pack: each line item opens into basis, options, exclusions, dependencies and why it is priced this way.

Not transparency theatre. Dispute prevention UI.

The office manager gets teeth

The consumer examples are easy to picture because homeowners and patients feel visibly outgunned. The same pattern gets more interesting inside small businesses.

An office manager receives a SaaS renewal. Last year’s contract is in one folder. Usage logs sit in the admin console. Support tickets are in Zendesk. Public pricing is on the vendor’s website. Three competitors have cleaner packages. The renewal email says the price is going up 14% due to “expanded platform capabilities”, which is a phrase designed to die under interrogation.

An assistant can compare:

• Last year’s seat count against current active users.
• Paid modules against actual usage.
• Support response times against the promised tier.
• Auto-renewal notice periods.
• New AI surcharges.
• Implementation fees that look suspiciously like sales make-up.

Then it drafts the email nobody had time to write:

“Please remove 18 inactive seats or justify their continued inclusion. Please explain the £4,800 platform uplift against our usage of the new features. Please confirm the cancellation deadline in writing.”

The office manager did not become a procurement expert. They acquired a workflow weapon.

That will irritate suppliers because junior buyers will suddenly ask better questions than their category knowledge deserves. Senior salespeople will spend time rebutting generated objections from people who barely understand the product. Some of those objections will be valid. Some will be nonsense pasted from a model that confused a London bathroom remodel with a cheap suburban forum post.

Both things can be true. AI will make weak buyers less weak, and it will make some weak buyers confidently annoying.

Sellers need a different object

The obvious business response is to “add AI to quoting”. That is too shallow.

A seller-side quoting assistant that writes warmer proposal language, inserts case studies and optimises follow-up timing does nothing for the new buyer behaviour. The buyer’s assistant is not reading for warmth. It is reading for unresolved liability.

The response is a cross-examinable quote.

That means designing the commercial document as if it will be audited before anyone signs. Not legalistic sludge. Not a tax return with a logo. A persuasive human layer backed by a structured, inspectable layer.

For a contractor, that might mean:

• “Fixture allowance: £1,200, based on Brand A basin, Brand B tap, Brand C shower mixer. Customer may select alternatives; difference charged at cost plus agreed handling.”
• “Disposal: included for standard debris up to X bags. Hazardous materials excluded.”
• “Subfloor: visual inspection only. Rotten subfloor triggers change order before replacement work begins.”
• “Permit: contractor to advise; owner responsible for application unless selected as paid option.”

For a SaaS renewal, it might mean:

• Seat table with active, inactive and provisioned users.
• Feature usage for paid modules.
• Support tier evidence.
• Renewal window.
• Price movement by component, not a single uplift line.
• A customer-facing explanation of new fees.

For an event venue, it might mean the difference between “staffing included” and “staffing included for six hours, overtime billed in 30-minute increments, security required after 11pm”.

The point is not radical openness. The point is to remove avoidable ambiguity before the buyer’s assistant turns it into a 14-question email.

Support inflation is coming

There is an ugly operational side to this.

Once every inbox has an “inspect with assistant” button, vendors will receive floods of semi-reasonable clarification emails. The same 12 questions, slightly different wording. Some buyers will forward the generated packet without reading it. Some will ask for alternative tiers they cannot afford. Some will cite policy clauses that do not apply. Some will accuse sellers of inflation because a model benchmarked against the wrong geography, quality level or year.

Sales teams will respond by creating quote FAQ agents to answer buyer agents. The first wave will feel absurd: two systems arguing over “reasonable market rate” while the humans wait for tone to become spicy enough to intervene.

There is a game theory problem here. If buyers can generate objections at near-zero cost, sellers must either absorb the support load, automate the rebuttal, or redesign the quote so common objections have nowhere to attach. The cheapest long-term answer is usually the third.

That will change pricing.

Opaque bundles may lose conversion in high-anxiety categories: medical bills, home renovation, car buying, insurance repairs, legal services, enterprise software renewals. Clean packages may earn a premium because they reduce the buyer’s unresolved questions. “Defensible quote” becomes a conversion feature.

Free estimates may also get more expensive. If a detailed scope becomes unpaid procurement work for a buyer and their assistant, sellers will start charging for it. I would not be surprised to see more consultation deposits: pay £250 for the detailed inspection and quote, refunded if the project proceeds. Buyers will complain. Good sellers will frame it as the price of a serious, defensible scope.

The limits are the product

There is a real failure mode here. Models can invent rights, misread contracts, overfit to irrelevant benchmarks and pressure buyers to squeeze quality out of work they do not understand.

Line-item obsession can be dumb. Craft work does not always decompose cleanly. A surgeon, builder, barrister or specialist consultant is not selling units of output in the same way a stationery supplier sells pens. Some price is judgement. Some price is availability. Some price is taking responsibility when the tidy plan meets reality.

That limit is not a reason to ignore the shift. It is the design brief.

Businesses need to decide which parts of their offer are inspectable, which parts are judgement-based, and which parts cannot be itemised without lying. Then they need a response ready for the moment every customer can cheaply ask: “Explain this line.”

The uncomfortable near future is mundane. A quote goes out on Monday. By Tuesday the vendor receives nine numbered questions, three requests for alternative tiers, two cited clauses and one false accusation. Today that feels like an edge case. Once the assistant lives in the inbox, it becomes the default buyer routine.

Most quotes are still built like brochures. Pricing pages still hide assumptions. Proposals are bloated with persuasion and thin on evidence.

AI will read them like a hostile accountant. That is the product surface now.

---