On this page
Most follow-up fails for a boring reason: it starts over. A lead asks about a Saturday appointment, gets a reply, goes quiet for a day, and the next message that arrives reads like the first one all over again. "Hi! Just checking in about your request." Nothing in it shows that anyone remembers the conversation.
Follow-up that works does the opposite. It continues. It picks up the thread where it stopped, references what the customer already told you, and asks for one small thing. That is the whole discipline, and it is mostly a configuration and workflow problem rather than a copywriting problem.
Follow-up is a continuation, not a restart
Treat the conversation as one long thread that happens to have gaps in it. The customer does not experience your first reply, your second touch, and your third touch as separate campaigns. They experience one exchange with a business that either paid attention or did not.
Two practical consequences:
- Never re-ask a question that has already been answered in the thread. If the address, the trade, or the timing is known, use it.
- Never send a nudge whose only content is that time has passed. "Following up" is not a reason for the customer to reply.
In InstantResponse.AI, follow-up runs inside the same conversation as the first response, and the conversational AI keeps replying with context until the lead responds or opts out, with the full history reviewable afterward. That review history is the tool that tells you whether your follow-up actually continued the thread or quietly restarted it.
Give every touch exactly one purpose
Before a message goes out, you should be able to name its job in four words or fewer. In practice there are only four jobs worth sending:
- Clarify. One missing detail blocks the quote or the visit.
- Reduce uncertainty. The customer is unsure about scope, timing, or what happens next.
- Confirm interest. The thread went quiet and you need a yes or a no.
- Make the next step easy. Offer a specific time, or a single link, or a photo request.
If a message tries to do all four, it becomes a questionnaire. Questionnaires get ignored. One purpose, one ask.
Build cadence around lead state, not a universal best time
There is no single correct hour to follow up, and any cadence built on a borrowed statistic will be wrong for half your job types. An emergency water leak and a spring gutter estimate do not deserve the same rhythm.
Two inputs decide cadence:
- Lead state. Has the customer replied at all? Partially answered? Asked a direct question you have not resolved?
- Job urgency. Same-day emergency, this-week repair, or planned project.
Urgent work compresses everything: a short second touch, then a handoff to a person. Planned work stretches out. A staged cadence such as day 3, day 7, and day 14 is a reasonable default for non-urgent leads and is configurable alongside voice, rules, and handoff triggers, so the schedule matches the trade instead of a generic template.
Seven lead states and what to actually send
The examples below are hypothetical and written to show shape, not to be pasted verbatim.
Never replied. The customer submitted a request and went silent. Restate the specific job in their words, then ask one closed question. Hypothetical: "You asked about a garage door that will not close all the way. Is it still not closing, or has it started working again?" A yes/no is easier to answer than an invitation to talk.
Partially qualified. They gave you the trade and the neighborhood but not the timing. Ask only for the gap. Hypothetical: "Got it on the kitchen faucet replacement in Northridge. Are you hoping for this week or is next week fine?" Do not re-collect what you have.
Asked about price. Silence after a price question usually means the answer felt evasive. Give the honest structure of your pricing, then offer the step that produces a real number. Hypothetical: "Repairs in that range usually depend on the part. A quick look on site gets you an exact number before any work starts. Want me to check availability?"
Requested a day or time. They named a window; your follow-up should confirm or counter with one alternative, never a menu of six. Hypothetical: "Thursday morning works. Does 9 to 11 fit, or is afternoon better?" Full calendar booking is currently in Beta, so many teams confirm the preferred window in conversation and let a dispatcher place it on the schedule.
Needs to send a photo. Photo requests stall because customers forget, not because they refuse. One reminder with the reason attached is usually enough. Hypothetical: "A photo of the panel label is all we need to confirm the right part before the visit."
A human promised a callback. The worst follow-up failure is an automated nudge landing on top of a commitment a person made. When a teammate has taken over, automated touches should stop and the reminder should go to the team, not the customer. Notifications over WhatsApp, SMS, or email exist for exactly this: the human owes the call, so the human gets the ping.
After-hours lead. An overnight message deserves an immediate acknowledgment with an honest next step, then a real follow-up when the office opens. Google's guidance for Local Services Ads message leads points the same direction: keep the inbox active and respond promptly, because response behavior is visible to customers. Fast acknowledgment plus a specific morning follow-up beats an overnight sales pitch every time.
Stay on the channel the lead came from
Continuity matters more than reach. A Google LSA message lead, a Yelp Request, and a Thumbtack message lead each carry the customer's expectation of where the conversation lives. Jumping channels mid-thread reads as a different company making contact, and on platform inboxes it can also cut you out of the record the customer sees.
Keep customer-facing follow-up in the source thread where it is supported. Use team-notification channels—WhatsApp, Twilio SMS, or email—for internal alerts and handoffs rather than starting a second customer-facing conversation.
Write down the stop rules before you write the messages
Cadence without stop rules is how good follow-up turns into a complaint. Decide the exits in advance and configure them, so no one has to remember them at 7pm on a Friday:
- The customer opts out, by keyword or plain language.
- An explicit no.
- Booked elsewhere.
- Wrong fit for the trades you cover.
- Outside your service area.
- Maximum attempts reached.
- Flagged as spam.
- A teammate takes over the conversation.
Keyword handling is operational as well as editorial. Twilio Messaging Services can process opt-out keywords such as STOP and block future sends on that number, which is a useful control for a separately configured SMS workflow. Treat it as one control among several, not as a compliance program. You are responsible for following applicable laws, the policies of each platform you receive leads from, and every opt-out choice a customer makes. That is an operating requirement, not legal advice, and if you are unsure, ask a qualified professional.
AI mode for conversation, scripted mode for confirmations
The two modes solve different problems, and mixing them up is what makes follow-up sound robotic.
Use AI mode when the next message depends on what the customer said: open questions, scope that is still fuzzy, hesitation about price, anything where a fixed script would answer the wrong question. Use scripted mode for messages whose wording should never drift, such as appointment confirmations, reminders, and opt-out acknowledgments. The comparison of AI and non-AI mode lays out the same split, and most teams end up running both: AI for the middle of the conversation, scripts at the edges.
A simple operating checklist
Run this before you turn any cadence on:
- Every touch has one named purpose.
- No message re-asks something already answered.
- Cadence differs by urgency tier, not by a single global timer.
- Each message contains exactly one ask.
- Customer-facing follow-up stays on the source channel where supported.
- All stop rules are configured, including human takeover.
- Confirmations, reminders, and opt-out replies are scripted.
- A named person reviews a sample of threads weekly.
That last item is the one teams skip, and it is the only one that catches drift.
Measure the conversation, not a vanity number
Follow-up quality shows up in a handful of signals you can read directly from your conversation history:
- Reply rate by touch number. If touch two and touch three produce nothing, they are probably nudges rather than continuations.
- Resolved conversations. Threads that reached a clear outcome, positive or negative.
- Confirmed preferred times or bookings, where booking applies.
- Handoffs to a person, and how quickly the person responded after the handoff.
- Opt-outs, watched as a trend rather than a single number.
- Reasons for non-conversion, captured as short tags: out of area, price, timing, chose another provider, no response.
Do not chase benchmarks you cannot verify. Your own baseline from last month is a more honest comparison than any industry average, and the non-conversion reasons will tell you more about what to fix than any single rate.
The trade-off is real in both directions
Too few touches abandons people who were genuinely interested and simply got busy. One message and silence is not restraint, it is a leak. Too many touches, or touches that ignore what the customer already said, damages something harder to rebuild than a lost job. Context-blind repetition is what teaches customers that your business does not read its own messages.
The balance point is not a number of messages. It is whether each message earns its place by adding something: a specific time, a clarified scope, an easy next step, or a graceful exit.
Conclusion
Follow-up stops sounding automated when it stops behaving automatically. Give each touch one purpose, reuse what you already know, set cadence by state and urgency, keep the conversation on the channel it started on, and write the stop rules before the messages. Then read your threads. The conversations themselves will tell you which touches are working and which ones are just noise.