Using AI for estimate and lead follow-up texts

Why follow-up is where jobs are won

Most home service companies are good at the first contact and weak at the second. Estimates go out and nobody checks back. Web leads get one callback that goes to voicemail. The research on lead response is old but consistent: in an audit of US companies, firms that contacted web leads within an hour were nearly seven times as likely to qualify them as firms that waited even an hour longer^[3], yet among companies that responded at all within 30 days, the average response time was 42 hours^[4].

Texts fix part of that problem because people actually read them. About 98% of US adults own a cellphone and 91% own a smartphone^[1], while eight in ten say they generally do not answer calls from unknown numbers^[2]. A text from a business the customer recognizes lands in a thread they can answer when they are free.

What to automate and what to keep human

Review requests deserve a caution. The FTC's guidance treats asking only happy customers for reviews as potentially misleading, so send the same request to everyone you serve^[5]. Our guide to drafting review replies with AI covers the reply side.

Consent and timing rules

Consent is the foundation. Ask for it plainly when the customer first contacts you, record exactly what they agreed to, and keep service messages separate from marketing. A customer who asked for an estimate has a reasonable expectation of hearing about that estimate; that does not mean they agreed to receive monthly promotions.

• Consent line at booking or inquiry: Is it okay if we text you at this number about this request? You can reply STOP at any time.
• Store the date, the channel, the exact wording and the number for each consent.
• Honor STOP, UNSUBSCRIBE and similar replies immediately and confirm once.
• Register your business texting with your messaging provider and carriers before sending at volume; see our guide to A2P 10DLC registration.
• Identify your company by name in the first message of every thread.

Timing matters too. Federal rules bar telephone solicitations to residential subscribers before 8 a.m. or after 9 p.m., local time at the called party's location^[6]. Many companies apply that window to all automated texts as a safe default, even service messages. Identification matters as well: nine in ten US adults think people often or sometimes impersonate others by phone or online to steal information^[7], so a text with no company name looks like a scam. This is general information, not legal advice; check marketing programs with counsel.

A sample estimate follow-up sequence

1. 01

   Day 0, within an hour of sending the estimate

   Hi Maria, this is Alex at [company]. Your estimate for the new heat pump is in your email. Happy to walk through the options by phone if that helps. Any questions so far?

2. 02

   Day 2

   Hi Maria, [company] here. Did the estimate answer your questions about the two system sizes? If timing is a concern, we can hold an install date while you decide.

3. 03

   Day 6

   Hi Maria, one more note from [company]: the estimate is valid until [date]. If anything has changed, just reply and I will update it.

4. 04

   Day 14, closing the loop

   Hi Maria, [company] here. I will close out your estimate for now so we stop messaging. If you would like to pick it up later, reply any time and we will take it from there.

Each message is short, uses the customer's name and the company name, asks one question and gives the customer an easy way to answer. The last message closes the loop politely, which often prompts a reply from people who meant to answer earlier.

Writing rules for every message

Whether a person or an AI tool writes the message, the same rules keep follow-up helpful rather than pushy. Put them at the top of your prompt or template library so every message is checked against them.

• One message, one purpose, one question. If you need to say two things, you probably need two messages on different days.
• Use the customer's first name and your company name. Never use a vague sender such as Hi, it's me.
• Keep it under about 300 characters so it reads in one screen on a phone.
• No fake urgency. Only mention a deadline if it is real, such as an estimate expiry date or a held install slot.
• No links unless the customer expects them, such as the estimate they asked for. Unexpected links look like phishing.
• No pressure after a no. Thank the customer, close the thread and stop.
• Write the way your best CSR talks on the phone: plain, warm and specific.

Read the sequence aloud before switching it on. If any message would sound odd coming from a person at your front desk, rewrite it.

Where AI helps, and its limits

AI is useful in follow-up in three places. First, drafting: turning estimate notes into one personalized line, such as a reference to the noisy outdoor unit the customer mentioned, so the sequence does not read like a template. Second, sorting replies: classifying answers as ready to book, has a question, not now, not interested or stop, and routing each one. Third, summarizing: giving the person who takes over a two-line summary of the conversation so far.

Its limits are just as important. AI should not invent discounts, deadlines or financing terms, should not answer price questions the company has not approved, and should not argue with a customer who says no. If AI replies in real time, it should be honest about being automated when asked.

Hand-off rules

1. Any reply that asks a price question, mentions a competitor's quote or asks for a discount.
2. Any complaint, frustration or mention of a previous bad experience.
3. Any reply mentioning safety, such as a smell of gas, water near electrical or a carbon monoxide alarm. Those also get your safety instruction and an on-call alert.
4. Any request to change scope, such as adding a second system or a different service.
5. Any reply the AI cannot classify with confidence.
6. Any customer who asks for a person.

Example: unsold landscape estimates

Example: a fictional landscaping company sends about forty design and install estimates a month and has never followed up systematically. It sets up a four-message sequence like the one above, with AI drafting one personal line from each estimate's notes and sorting replies. The office manager approves every first message for a month. Replies split into three groups: questions about timing, requests to phase the project over two seasons and polite no-thank-yous. The company adds an approved answer about phasing, routes timing questions to the scheduler and leaves pricing questions to the owner. After two months the owner reviews which messages get replies and drops the day-six message, which produced almost none.

What to measure

Track reply rate per message, jobs booked from follow-up, opt-out rate and complaints. A rising opt-out rate means messages are too frequent or too salesy. Look at follow-up results by job type too; maintenance plan renewals and large installs behave very differently. Feed what you learn back into your call summaries and CRM notes, so the next person who speaks to the customer knows what has already been said.