Most of the automations I write about here recover a contact that was missed — the call that rang out, the form abandoned halfway, the quote that went quiet. This one is different, and it is the one owners argue with hardest. The contact was not missed. Somebody answered. The caller opened with four words — "how much do you charge?" — and ninety seconds later the call was over, nothing was booked, no name was captured, and nobody in the business considers that a loss. It gets filed under "just a price shopper" and forgotten.
It is, by volume, the most common inbound contact a local service business gets, and it is the only one with no process attached to it. Every other contact type has a path: the booking call has a calendar, the complaint has an owner, the quote request has a form. The price question has a shrug and a number, or a shrug and a refusal to give a number. As Gideon Wafula, AI Automation Engineer, I have now built the same triage layer for enough trades, clinics and studios that the pattern is clear: the category is not unprofitable, it is unsorted.
Start with the thing that makes it expensive rather than merely annoying. The business already paid for this call. If it came from a search ad, a Local Services Ads click or a map pack listing, the acquisition cost is spent whether the caller books or hangs up. Home-services marketing benchmark roundups published this year put the share of consumers who call from search and go on to buy at around forty percent when the call is handled properly — meaning the majority of that spend is decided inside the first two minutes of a conversation nobody has written down a process for.
Second, the same benchmark work keeps landing on a less comfortable finding: close rate, not cost per lead, is where the money separates. Two companies paying identical costs per lead can book twice the work if one converts better, and the reported gap between average and top performers runs to something like fifteen to twenty percentage points across the trades. That gap does not live in the ad account. It lives in calls exactly like this one.
Third — and this is the part that decides the design — the price question is usually not a price objection. It is the only question a caller knows how to ask. They do not have vocabulary for their problem. They cannot describe the fault, the model number, the square footage or the treatment. "How much" is the one question that works in every industry, so it is the opener regardless of what the caller actually wants. Treating it as a signal of price sensitivity is a category error that costs real revenue, because a meaningful share of these callers have a defined job, a real address and a near-term deadline, and simply opened with the wrong sentence.
Businesses answer this call in one of two ways, and both are wrong in opposite directions.
The number. Someone gives a figure to make the call end. It is either a lowball designed to get in the door, which sets an anchor the technician has to fight on site, or a high defensive number that ends the conversation. Either way the business has now competed on a single dimension against companies whose numbers cover different scopes. Nothing was captured — no name, no address, no callback number — so there is no second chance and no record that the contact ever happened.
The refusal. "It depends, I'd have to send someone out." This is technically true and conversationally fatal. The caller hears evasion, and they are not wrong to: a business that cannot describe its own price structure sounds like one that makes it up. Pricing guidance written for contractors this year is consistent on the alternative — acknowledge the concern, then reframe on scope and consequence rather than dodging — and the reframe only works if the person answering has been given something specific to say.
The honest third answer is neither. It is a structure: here is the starting point, here is what moves it, here is what the visit costs and what that fee covers, and here is the next available slot. That answer takes fifteen seconds, requires no improvisation, and is identical every time — which is precisely what makes it automatable.
Each of these needs a different fix, and lumping them together is why most attempts at this fail.
The largest leak and the easiest to close. A price call that ends without a name and a number is a contact the business cannot retry, cannot measure, and cannot attribute to the ad that produced it. Most booking-focused phone processes only capture details once the caller agrees to book, which means every non-booking call is invisible. The fix is not a longer script — it is capturing identity before the answer rather than after, framed as a reason rather than a demand: a number to text the breakdown to, which most callers accept because they are being given something.
The caller who says "how much for a boiler service" and, if asked two more questions, turns out to have a specific make, a specific fault, a house inside the service area and a tenant arriving Friday. This is a booking, mislabelled. It is lost because the person answering hears "how much", gives a number, and never asks the two questions that would have revealed it. Structured qualification — what, where, when, how they found you — is a fifteen-year-old sales practice and the published pre-qualification frameworks all follow the same shape: opener, motivation, timeline, then bridge to the appointment. The automation's job is to make those four fields non-optional without making the call feel like a form.
Some callers really are collecting three numbers. They are not worthless; they are badly timed. What they need is a written summary they can compare against the other two quotes, and then silence, and then one well-spaced follow-up. What they usually get is either nothing at all or an aggressive callback sequence built for hot leads, which is how a business trains someone to block its number. These two populations need different cadences, and the split has to happen at the moment of the call, not later.
The same question increasingly does not arrive by phone. It comes through a Google Business Profile text button, a website chat widget, or Instagram. Google retired its in-dashboard Business Profile chat in mid-2024, pushing the interaction toward SMS and WhatsApp buttons on the listing itself, and coverage of profile messaging this year keeps reporting that only around a third of eligible profiles respond through those channels at all. The gap matters more than it sounds, because a typed "how much for a deep clean?" is a lead that has already told you the service, in writing, timestamped — and is sitting unanswered in an app nobody has open.
An afternoon of export work settles whether this is worth an engineer. Pull ninety days of inbound calls from the phone system, filtered to calls under three minutes that produced no booking — that population is your price-shopper volume and it is nearly always two to four times what the owner guesses. Cross it against ad spend for the same window to get the acquisition cost already sunk into calls that produced no record. Then count how many of those numbers appear again later, which tells you what share came back on their own and is the honest baseline any automation has to beat.
Do the same for messages: count unanswered or slowly-answered inbound texts and profile messages, and measure time-to-first-reply rather than reply rate, because the speed literature is brutally consistent — the well-worn finding that contact within five minutes makes qualification something like twenty-one times more likely has survived a decade of re-testing. I have written up the mechanics of that window separately in the five-minute speed-to-lead build.
Finally, listen to ten of these calls. Not a summary, the actual recordings. In every engagement I have run, the owner discovers at least two calls that were obviously bookable jobs, which is usually the moment the project gets approved.
Six steps, in n8n, sitting between the voice platform, the messaging channels and the CRM. The governing rule is the one that keeps this safe: the automation may explain price structure freely, and may state a specific number only where that number is fixed, published and already true.
Step one: a price-answer library, not a prompt. For each service you sell, write the four fields the answer needs — starting point, what moves it, diagnostic or call-out fee and what it covers, and typical range if you are willing to publish one. This is a structured record the workflow reads, not instructions in a system prompt. It is also the actual work of the project; the automation on top of it is straightforward. If a service has no honest structured answer, that is a pricing problem the build will expose rather than solve.
Step two: detect the intent on any channel. Classify the inbound contact — call transcript, SMS, profile message, chat — into a price-enquiry intent with a service type attached. One classifier, many inlets, so a typed question and a spoken one enter the same pipeline and appear in the same reporting.
Step three: answer first, then qualify. The agent gives the structured answer for that service, then asks the three or four qualifying questions. The order is not stylistic. Callers tolerate questions after they have been given something and resist them when the questions come first, and reversing these two steps is the single most common reason a deployment underperforms. Cap it at four fields: job, location, timing, source.
Step four: score and branch. Inside the service area, defined job, timeline within the booking horizon — that is a booking attempt, offered a real slot now, the same way a same-hour quote flow works. Outside the area or outside the horizon, or explicitly comparing, and it becomes a written summary plus a slow nurture. The branch is a rule in the workflow, not a judgement call in a prompt, because it decides how often a human being's phone rings.
Step five: send the written breakdown. Every price enquiry, regardless of branch, gets a text or email with the structure that was just explained, built from the library fields with the model writing phrasing and never figures. Any draft containing a number absent from the source record gets rejected before it sends. This document is what the caller compares against the other two quotes, and it is the only artefact your competitors are probably not providing.
Step six: write everything back. Contact, service type, quoted structure, branch taken, outcome. Price enquiries become a measurable population with a conversion rate instead of a shrug. Without this step you cannot tell in three months whether any of it worked, which is how these builds quietly get switched off.
Three, and the first is non-negotiable. Never let the model produce a job price. Fixed and published figures are injected as validated variables; everything else is structure. A model estimating a repair cost has created a commitment the business will be held to by a customer who has it in writing, and the reputational cost of walking that back is far larger than the booking was worth.
Cap the qualification. Four fields. Every additional question measurably increases abandonment, and an agent that keeps probing after a caller has signalled they are done reads as evasive — the exact impression the build exists to remove.
Treat the nurture branch as outbound. A comparison shopper who did not book is a prospect, not a customer, and the follow-up sequence sits under the same consent and opt-out rules as any other outbound programme, against the same suppression table every other workflow reads. Nothing about having called you first changes that.
Running cost lands between 40 and 150 USD per month for a typical local business, scaling with inbound contact volume rather than customer count, on top of an existing voice or messaging platform. Setup is dominated by step one — getting honest price structures written down per service — which is usually a couple of sessions with whoever actually sets prices, and is uncomfortable in a useful way.
Baseline four numbers before go-live: price-enquiry volume as a share of all inbound contact, identity-capture rate on those contacts, booking rate from price enquiries, and time-to-first-reply on typed enquiries. The second one moves first and fastest, usually within a fortnight, and it is the one that makes the rest measurable.
The honest caveat: this build makes a price conversation structured, fast and recorded. It does not make an uncompetitive price competitive. If the real problem is that the business is priced thirty percent above the market with nothing to justify it, this automation will surface that fact much more clearly and much sooner — which is genuinely useful, and not at all what the owner asked for.
Gideon Wafula builds custom AI automation systems, n8n, WhatsApp, Voice AI, and more.
See Services →