Automate unsold estimate follow-up for contractors
Automate unsold estimate follow-up for HVAC, plumbing and electrical contractors: the estimate-state trigger, five touches of real copy and stop-on-reply.
What you get from this guide
- Why the trigger is estimate state, not a timer, and what that fixes
- How to detect acceptance from the field service system instead of guessing from replies
- A five-touch cadence segmented by estimate value and job type
- Real message copy for every touch, including the close-out
- Every stop signal, and the pre-send re-check that catches the ones you miss
Why estimate state is the trigger
Search for this and you will find a page selling you the outcome. Recover the revenue sitting in your unsold estimates. Turn dead quotes into booked jobs. The pages are not wrong about the opportunity, and the vendors behind them - Hatch and the agencies that resell this as a service - are selling a real product. What almost none of them show is the thing you actually need, which is the mechanism: what fires the sequence, what stops it, and how the system knows the customer already said yes.
Here is the distinction that makes a contractor build different from a generic drip. A marketing sequence is driven by time. You entered the list, so on day three you get message two. An estimate follow-up sequence is driven by state. The estimate is an object in your field service system with a status, an amount, a job type and a customer, and every one of those changes what should happen next. Treat it as a mailing list and you will eventually text a customer to ask about an estimate they signed last Tuesday - and that single message costs you more trust than the whole sequence earns.
So the rule for the entire build is this: the field service system is the source of truth, and the automation reads it, never assumes it. Every design decision below follows from that.
This is the specific version of a general guide
Detecting acceptance from the system
This is the section the vendor pages skip, and it is the one that decides whether your build is safe to turn on. There are four ways to learn that an estimate is no longer unsold, and they are not equally trustworthy.
| Signal | How you get it | Trust it? |
|---|---|---|
| Estimate status read at send time | An API call for that estimate immediately before the message goes out. | Yes. This is the authority. Everything else is an optimisation on top of it. |
| Estimate status-change webhook | The system pushes an event when the estimate is approved, declined or converted. | Yes, as a fast path. Not as the only path - events get delayed, dropped and replayed. |
| A job or invoice created from the estimate | A job webhook, or a scheduled query for jobs linked to the estimate. | Yes. Often arrives before anyone updates the estimate status by hand. |
| The customer replied 'yes' | Inbound SMS or email, possibly classified by a model. | Only as a stop, never as a confirmation. Pause the sequence and get a human to check. |
Housecall Pro is a useful concrete example because it publishes both halves. Its help documentation describes estimate statuses changing to Customer Approved or Customer Declined when the customer acts, and Pro Approved or Pro Declined when a pro does it manually. Its webhooks documentation lists estimate events including estimate.sent, estimate.updated, estimate.copy_to_job and estimate.option.approval_status_changed, and states plainly that webhooks are only available on the MAX plan, returning a signing secret when you save a webhook URL. All of that is as documented in August 2026; confirm it in the live help centre before you build, because plan gates move.
That plan gate is the pattern rather than the exception. Housecall Pro puts webhooks behind MAX, ServiceTitan puts API access behind a plan tier - covered in detail in the ServiceTitan automation API guide - and Jobber gates its native quote follow-ups by plan too. Find out what your plan actually includes before you design anything, because the answer decides whether you are building on webhooks or on scheduled polling. The polling versus webhooks entry covers the trade-off in general terms.
If your system publishes no estimate events at all, you are not blocked. Poll the estimates endpoint on a schedule, diff the statuses against what you stored, and treat any transition into a closed status as a stop. You lose freshness, not correctness - and because the pre-send read is the real safety net, a polling build is just as safe as a webhook build. It is only slower to notice.
Segmenting before you send
One cadence for every estimate is the second most common mistake in this build, behind not checking state. A contractor's estimate book is at least three different sales processes wearing the same label.
| Segment | Typical work | Treatment |
|---|---|---|
| Small repair | Drain clear, minor electrical, a single fixture. | Short ladder, two or three touches inside ten days. The decision is fast or it is not happening. |
| Mid-size job | Repair with options, partial replacement, multi-day work. | The full five-touch sequence. This is where automated follow-up earns its keep. |
| High-value replacement | System replacement, re-pipe, panel upgrade, anything financed. | One automated touch at most, then a task for the person who walked the job. Automate the reminder to call, not the call. |
The threshold between mid-size and high-value is a business decision, not a technical one. Pick a number you would be comfortable defending to the customer if they found out a machine was chasing them about it. Most trades businesses land somewhere that is roughly twice their average ticket. Above that line, the automation should be creating work for a human, not replacing them.
Segment on job type as well as value where you can. An estimate presented with financing options attached has a different decision cycle to a cash-price repair, and an estimate raised after an emergency call has a different urgency to one from a planned assessment. If your system carries the job type as a field, use it - it is free segmentation that you would otherwise pay a person to apply.
The five-touch sequence
The cadence below is for the mid-size segment. It is not a magic set of intervals, and anyone who tells you their specific day numbers are optimal is selling something. What the gaps are actually doing is covering the distinct phases of a real decision, and that is the part you should keep even if you shift every date.
| Touch | Timing | Job of the message |
|---|---|---|
| 1 | Same day, a few hours after the visit | Confirm the estimate arrived and ask how the visit went. A service question, not a sales question. |
| 2 | Day 3 | Add information rather than pressure: what the options mean, current lead time, what happens if it waits. |
| 3 | Day 8 | Ask the direct question and invite the objection. Price, scope, or timing - which one is it? |
| 4 | Day 18 | Change the frame. Availability, a seasonal reason to act, or an offer to re-scope to a smaller version. |
| 5 | Day 30 | Close it out. Say you will stop chasing, state when the pricing expires, keep the door open. |
Touch one being a service question rather than a sales question matters more than the timing. "How did the visit go?" is answerable by someone who has not decided anything yet, which is most of your list on day one, and an answer of any kind puts a human back in the conversation. "Ready to book?" on day one is answerable only by the people who were always going to book.
The gap between touch three and touch four is deliberate. After a direct ask that got no reply, another message three days later reads as pestering. Ten days later, with a genuinely different angle, reads as professional persistence. Touch five is the one owners want to cut and the one that most reliably produces replies, because permission to say no is easier to act on than another request for a yes.
Vendor guidance on cadence is worth reading with your eyes open. Hatch, whose whole product is this use case, describes its highest-responding HVAC campaign as seven messages - five texts and two emails - across seven days, in an analysis published on its blog and read in August 2026. That is far denser than the ladder above. It is also measuring response, and a dense seven-day burst optimising for any reply is a different objective from protecting a relationship over a month-long replacement decision. Decide which you are optimising for before you copy anyone's intervals.
Message copy for every touch
Fill the braces from the estimate record. Keep the amount, the job description and the estimate number as literal fields so nothing gets paraphrased into something you did not quote - a real risk if you let a model write the whole message rather than personalise one line of it.
Touch 1 - same day (SMS)
Hi {first_name}, {tech_name} here from {business}.
Thanks for having me out today. I have sent the
estimate for {job_description} through to
{email} - can you let me know it arrived?
Any questions on the options, just reply here.Touch 3 - day 8 (SMS)
Hi {first_name}, following up on estimate
{estimate_number} for {job_description}
({amount}).
Are you looking to go ahead, or has something
changed? If it is the price, the scope or the
timing, tell me which and I will see what I can
do. A no is completely fine - it just helps me
plan the schedule.
{sender_name}, {business}Touch 5 - day 30 (email)
Subject: Closing out your {job_description} estimate
Hi {first_name},
I will stop chasing you on this one - I know it is
not the only thing on your list.
Estimate {estimate_number} at {amount} holds until
{expiry_date}. After that I would need to re-check
it against current parts pricing, which I am happy
to do any time - just reply to this email.
If anything changes with the {job_description},
you know where we are.
{sender_name}
{business} - {phone}Four copy rules that matter more than the wording. Send from the person who walked the job, by name, because an estimate is a personal transaction and a message from "the team" reads as a machine. Keep every message shorter than the estimate. Never open two messages in the same sequence with "just following up". And thread emails onto the original rather than starting new subjects, so the customer can see the numbers without hunting.
If you are sending automated SMS from a ten-digit US number, registration comes before copy. The A2P 10DLC guide covers brand and campaign registration, and the A2P 10DLC glossary entry is the short version. Unregistered automated texting gets filtered by the carriers rather than delivered, and you will not necessarily get an error telling you so.
Stop-on-reply and stop-on-acceptance
Everything above is copy. This is the engineering, and it is the difference between an automation your salespeople trust and one they quietly ask you to turn off.
There are two families of stop. Stop-on-reply means any inbound signal from the customer pauses the machine and hands the conversation to a person. Stop-on-acceptance means the estimate's state in the source system has moved out of unsold, so the sequence ends whether or not anyone replied. You need both, and they behave differently: a reply is a pause with a human owner, an acceptance is a permanent close.
- Inbound SMS on the number the sequence sends from - stop, assign to the salesperson, and never auto-resume.
- Inbound email reply on the thread - same treatment.
- Estimate status moving to any accepted, approved, declined or expired value - close the record with that reason.
- A job booked or an invoice raised that references the estimate - close, because this often happens before anyone updates the estimate by hand.
- A manual stop button the salesperson can hit after a phone call, which is the most common real-world stop and the one people forget to build.
- An unsubscribe or STOP keyword - close, and write the opt-out back to the customer record so no other sequence picks them up.
The keyword handling is not optional and not just etiquette - see stop handling and opt-in for what carriers and regulators expect. The part builders get wrong is scope: an opt-out has to stop every sequence you run, not the one that happened to send the message.
Now the important bit, which almost no page on this topic states. Checking the stop condition when the event arrives is not enough. Events race. A customer signs at 9:58, your hourly sweep picks up their record at 10:00, and the approval webhook lands at 10:01. If the only check happened at event time, the message went out. So the sweep re-reads the estimate from the API for every record it is about to message, and skips any that has moved. That is one extra API call per due record, a handful per hour, well inside any sane rate limit, and it is the single highest-value line of code in the build.
The build, step by step
The steps below are platform-neutral. They work in n8n, Make, a scheduled script, or whatever you already run. What matters is the shape: a state table, an hourly sweep, and a read-before-send.
Write down what unsold means in your system
Open your field service software and list every estimate status it can hold. Mark which ones are unsold, which are closed-won, which are closed-lost, and which are ambiguous. The ambiguous ones are where the bugs live - a status like "pending" that your office staff use for two different things will produce messages to the wrong people forever. Fix the process before you automate it.
Trigger on presented, not created
Subscribe to the estimate-sent event if your plan includes webhooks, or poll for estimates whose status changed since your last run. Trigger on the estimate reaching the customer. Triggering on creation means asking someone about an estimate that is still a draft on a tablet in a van, which is the fastest way to lose confidence in the system.
Create the follow-up record
One row per estimate in Postgres, Airtable or a sheet: estimate id, customer id, name, mobile, email, amount, job type, presented timestamp, status = open, step = 0, next_touch, and a segment field from the value rules above. Key it on the estimate id so a replayed webhook updates the row instead of creating a second sequence - see idempotency for why this is the first thing to get right.
Route the big ones to a human
Before the record enters the ladder, check the amount against your threshold. Above it, create a task for the person who walked the job, attach the estimate, and set the record to status = human. The automation still does useful work here - it remembers, it schedules, it nags the salesperson rather than the customer. That is human-in-the-loop applied where it actually pays.
Build all six stop listeners
Inbound SMS webhook, inbound email, estimate status-change webhook or poll, job-created event, manual stop, and STOP keyword handling. Each finds the record by estimate id or customer id and writes status, a reason, and a timestamp. Build these before the first outbound message exists, and test each one by actually replying to a test estimate from a real phone.
Sweep hourly and re-read before sending
A scheduled trigger every hour selects records where status = open and next_touch has passed. For each one: call the API for the current estimate, and if it is no longer unsold, close the record and move on. If it is still unsold, apply the quiet-hours guard in the customer's timezone, send the touch, log it, then increment step and set the next next_touch. Use a scheduled trigger rather than multi-day wait nodes, which do not survive restarts or redeploys.
Write everything back to the source system
After each send, post a note or activity onto the customer or estimate record in the field service system recording what was sent and when. This is what makes the salesperson trust it: they open the customer in the software they already use and see the whole follow-up history, instead of having to remember a second system exists. Handle failures here with retry and backoff so a transient API error does not lose the log line.
Close records and report weekly
After the final touch, set status = closed_no_response. Then report: estimates entered, messages sent, replies, estimates that moved to sold while in sequence, and value of that work. Attribution here is genuinely fuzzy - some of those customers would have booked anyway - so report it as "estimates that converted while in sequence", not as revenue the automation created. Overclaiming is how these projects lose credibility six months in. The cost per booked job framing is the honest way to size the value.
Where these builds go wrong
- Time-driven instead of state-driven, so the sequence keeps running after the customer signed.
- Checking estimate status only when a webhook arrives, never immediately before sending.
- Triggering on estimate created rather than presented, so customers hear about drafts.
- One cadence for a two hundred dollar repair and a fourteen thousand dollar replacement.
- Multi-day wait nodes instead of a state table, so a redeploy strands or duplicates sequences.
- Email and SMS run as separate sequences with separate state, so a text reply does not cancel the queued email.
- Quiet hours applied in server time rather than the customer's timezone.
- Never writing the messages back to the field service system, leaving salespeople blind.
- Sending automated SMS in the US without A2P 10DLC registration, so messages are filtered silently.
- Reporting every converted estimate as revenue the automation created, which nobody in the business believes.
If your business runs on presented estimates, this is usually the highest-return automation available to you, because the trigger is clean and the judgement stays with a person. Sector context is on the HVAC and plumbing automation page, the money side is on AI automation cost, and if you want us to scope it, the free automation audit is the place to start. What we do and how handover works is on automation services.
FAQ
- What counts as an unsold estimate?
- An estimate that was presented to the customer and has not been accepted, declined, or expired. The precise definition has to come from your own field service system's status values, because that is what your automation will read. In Housecall Pro, for example, the help documentation describes an estimate status changing to Customer Approved when the customer approves it and Customer Declined when they decline, with Pro Approved and Pro Declined for the manual equivalents, as documented as of August 2026. Your unsold set is everything that has been sent and has not landed in one of the closed states.
- How is this different from a generic quote follow-up sequence?
- Two things. First, it is driven by estimate state read back from the field service system rather than by a timer that assumes nothing changed. Second, contractor estimates carry a job type and a value that should change the cadence, because a two hundred dollar drain clear and a fourteen thousand dollar system replacement are not the same sales process. Our generic quote follow-up guide covers the platform-agnostic cadence and templates; this page is the contractor-specific, state-driven version.
- How many follow-up touches should an unsold estimate get?
- Enough to cover the period in which the decision is actually made, which for replacement-sized work is weeks rather than days. Five touches across roughly a month is a defensible default for contractor estimates, with the last one being an explicit close-out. What should drive your own number is your data: pull your last two hundred unsold estimates, find the date of the last contact on each, and see how many died with no contact after the first week. That count is the size of the problem.
- Are the response-rate numbers vendors publish reliable?
- Treat them as vendor-reported and not as a benchmark you should expect to hit. Hatch publishes an analysis of what it describes as 163,212 HVAC estimate follow-up campaigns over six months, reporting an average response rate of 60 percent, a best campaign at 90.06 percent and a worst at 17.82 percent, as published on its blog and read in August 2026. That is a real dataset with a real method described, but it is Hatch's own customers, running campaigns on Hatch, measured by Hatch, and response is not the same as revenue. Use it to argue that follow-up is worth building, not to forecast your own numbers.
- How do I stop the sequence when the customer accepts?
- Do not guess it from a reply. Read the estimate status back from the field service system immediately before every send, and treat an accepted status, a job created from the estimate, or an invoice raised against it as a hard stop. A webhook telling you the estimate was approved is a fast path, but the pre-send read is the safety net, because webhooks get delayed, dropped and replayed. The failure everyone eventually hits is texting a customer who signed forty minutes ago.
- Should the follow-up be text or email?
- Both, and the text is the one that does the work in the trades. Text gets read, email carries the estimate document and the numbers. If you send automated SMS from a ten-digit US number you must register the brand and campaign under A2P 10DLC first, or your messages will be filtered rather than delivered - that step blocks more of these builds than any code problem does.
- What should the last message say?
- That you are going to stop chasing. An explicit close-out message consistently produces replies from customers who felt awkward about the silence, it protects the relationship for the next job, and it gives you clean data: an estimate that passes the close-out is genuinely dead rather than sitting in a follow-up list forever. Say when the pricing expires and offer to re-check it later.
- What does it cost to have this built for us?
- The cost driver is not the sequence, it is the state of your estimate data and whether your system exposes an API on your plan. If estimate statuses are used consistently and the API is available, this is one of the cheaper builds with a clear payback. If half your estimates sit in a status nobody maintains, the cleanup is the project. We scope it in a free consult and quote a fixed price in writing rather than publishing a package price that would only be a guess at your setup.
Done for you
Want your unsold estimates chased properly, without a message going to someone who already signed?
We build estimate follow-up against your field service system's API: the state table, every stop signal, the pre-send re-check, the write-back, and reporting you can defend. The audit comes first, and it is free.
The consult is free, and you keep whatever we build.
Keep building
The rest of the cluster
Automate Quote Follow-Up (Build Guide + Templates)
The four-touch cadence, copy-paste templates, stop-on-reply logic, and the n8n or Make build.
Read the guideServiceTitan Automation API Without Zapier
The plan gate that stops most people, client-credentials auth with tenant ID and app key, and the documented rate limits.
Read the guideJobber Quote Follow-Up Automation Past 2 Reminders
The documented two-reminder limit, and the four-touch sequence you build on top of Jobber's API.
Read the guideFree AI Automation Audit Template
Score every recurring task and rank the top three worth automating before you build anything.
