Disclosure: Futuro sells the MasterMind system described here, so we have a commercial interest. The legal case is public record, industry figures come from the linked sources, and the call script is on our trades page for anyone to test. We never accept paid placement; see our Publishing Principles. Sources verified August 21, 2026.
Why does “how much will this cost?” tie plumbers in knots?
Because every answer carries a risk. Quote a firm number before you have seen the job and you own that number. Refuse to say anything and the caller hangs up and dials the next shop. Plumbing is a $126.4 billion industry of roughly 113,000 mostly small businesses (ConsumerAffairs), many of them one-truck shops (Census nonemployer data; SBA small-business profiles). The median plumber earns $62,970 a year, and the trade still adds about 44,000 openings annually (BLS), so the phone keeps ringing.
The shop’s real policy is nuanced: ranges for standard jobs, a flat diagnostic fee, “estimate before work” for everything else. Nuance is what generic answering services and generative chatbots both destroy, in opposite directions. One says nothing useful; the other says too much, confidently.
What happens when the bot invents a policy?
Ask Air Canada. In February 2024, British Columbia’s Civil Resolution Tribunal ruled in Moffatt v. Air Canada that the airline was liable for a refund policy its chatbot had invented, awarding $812.02 (CanLII; CBC News). The airline argued its chatbot was a separate legal entity. The tribunal called that remarkable and held the company responsible for everything on its website, static page or bot (American Bar Association; UBC Law Review).
Swap “bereavement fare” for “water heater install” and the exposure is yours. Meanwhile the caller who hears nothing useful moves on, the pattern behind missed-call economics. What a plumber needs is an AI that quotes like your best office manager on her most careful day. That is the design goal of MasterMind, and of this page.
What do three pricing calls sound like when the AI can only quote your rate card?
Three annotated exchanges. Every number traces to something the owner approved; every boundary to the owner’s policy.
Call one: the emergency, quoted from the rate card
This exchange runs on our trades page right now. A caller reports a flooding basement. The AI books Mike for 2:30 PM, asks where the water is coming from, then closes with: “The emergency service rate is $175 for the first hour, and Mike will provide a complete estimate before any work begins.”
Annotation: $175 for the first hour is a rate-card figure. “Complete estimate before any work begins” is the owner’s boundary statement. What the AI did not do is quote a repair total. That absence is the architecture.
Call two: the water heater, quoted as a range
Caller: “What does a water heater replacement run?” The AI answers from the approved range: “A standard replacement typically runs between $1,400 and $2,200 installed, depending on the unit and the hookups. Mike confirms the final price after inspection, before any work starts. Want me to book that estimate?” National cost guides put installed replacements in that band (HomeGuide), and ENERGY STAR’s guidance shows why the unit moves the number.
Annotation: the range is quotable because the owner marked it quotable. The caller gets a number to plan around; the shop commits to nothing it has not seen.
Call three: the refusal, delivered like a person
Caller: “Just give me a firm total. What’s the whole job going to cost?” The AI: “I don’t want to give you a number that won’t hold up. Mike confirms pricing on site, and the estimate comes before any work. I can have him out this afternoon.”
Annotation: this is where generative systems fail. A model trained to be helpful invents a plausible total. This system has none to give; the owner never approved one. The refusal is not caution; the AI has nothing else it is allowed to say.
Why is the refusal structural and not a setting?
What is the difference between retrieval and generation?
A generative system predicts the next plausible word. Asked for a price it has never seen, it produces one anyway, fluent and wrong. Researchers call this hallucination, a documented failure mode of generated text (arXiv survey), and NIST’s AI Risk Management Framework makes “valid and reliable” the first property of trustworthy AI for this reason (NIST).
MasterMind works the other way: the agent speaks from your knowledge graph and nowhere else, so no generative step exists to improvise a number (the retrieval architecture).
Why does that matter more than a careful prompt?
Because prompts instruct, and instructions get overridden by a smooth question at the wrong moment. A retrieval system cannot be sweet-talked into a figure that is not in the graph, which is why it survives Fridays, holidays, and pushy callers.
What do you control on the rate card?
Everything, and it takes one sitting inside the 24-to-48-hour build. Three decisions.
Which numbers are quotable?
Diagnostic or service-call fee, first-hour rates, emergency rates, standard-job ranges you stand behind. The AI says them verbatim, as often as callers ask.
Which numbers are forbidden?
Job totals, unlisted services, anything that depends on seeing the work. These do not exist for the AI. A caller asking for one gets the boundary, not a guess.
What exact words draw the line?
You write the boundary phrasing in your own voice, from “the estimate comes before any work” to however your shop says it. The ingestion story is covered in the knowledge-system guide, and the same grounding logic keeps policy answers honest in other verticals, as our salon coverage shows.
Which of these eight callers is on your line right now?
The same rate card handles all eight. What changes is the move after the answer.
The insurance-adjacent caller fishing for a number
They want a figure to anchor a claim, not a plumber. The AI gives quotable rates only and books the written estimate visit, never a total that could surface in a dispute.
The price-shopper comparing three shops
They have two other tabs open. An honest range plus fast estimate booking beats a suspiciously precise number. Straight answers read as confidence.
The property manager who needs written figures
Their file needs paper, not a phone quote. The FTC tells consumers to get written estimates with scope, materials, and price (FTC), and its Cooling-Off Rule covers in-home sales. The AI texts your rate card immediately and books the written-estimate visit (FTC contractor guide).
The landlord with a tenant emergency
Water is moving and authorization matters. The AI quotes the emergency first-hour rate, states the boundary, and confirms who approves work before dispatch.
The first-time homeowner who doesn’t know what’s normal
They are really asking whether they are about to be taken. The BBB’s hiring advice tells them to collect written estimates, so a calm rate-card answer with a stated boundary reads as the professional in the room.
The “just ballpark me” caller
The trap call. The AI gives the approved range for standard work and holds the boundary on everything else, as many times as pushed.
The commercial account asking about contract rates
Out of scope by design. The AI takes structured details and routes to you; contract pricing was never on the card and never should be.
The after-hours caller asking what emergency costs
Midnight, water heater leaking. Emergency work nationally runs $100 to $350 an hour plus a service-call fee (HomeGuide). The AI quotes your approved emergency rate, books the window, and you wake to a scheduled job instead of a voicemail (after-hours coverage).
Who keeps the rate card honest?
What happens when your prices change?
You tell us, and we say so plainly. The AI quotes only what is in the graph. Raise your diagnostic fee and tell no one, and the AI keeps quoting the old one, perfectly confidently, because that is what you approved. The discipline is a five-minute habit: prices change, one message to us, graph updated. That is the honest price of an AI that never freestyles.
What still belongs to your estimator?
Complex job pricing, always. The AI’s job is to make the estimate visit happen, not to replace the person walking the job.
How we verified this, and what we did not
Evidence level: mechanism demonstration on public artifacts. The emergency exchange is the live script on our trades page, quoted verbatim. The legal stakes come from the public record in Moffatt v. Air Canada (CBC News). Industry figures trace to the linked BLS and industry sources. Our voice-quality priors come from our published 1,000-participant blind study (read it), per our editorial standards.
What we did not do: we did not run a live deployment inside a plumbing shop for this piece, and the Moffatt ruling is a tribunal decision, persuasive rather than binding precedent in U.S. courts (ABA analysis). Futuro (our product) is the system described; the legal and industry sources are independent.
Last reviewed August 21, 2026. Errors get corrected under our corrections policy.