How to follow up on Facebook leads without losing them

The ad worked. The form was submitted. The row is now sitting in a tab somebody opens on Thursdays. Most of what small businesses call a lead quality problem is actually this moment, and it is the cheapest thing in the whole account to fix, because it costs no money at all. This article is about what happens after the lead exists: who touches it, in what order, what they say, how you tell the real ones from the noise, and what you owe the ad platform once you know the answer.

The lead is colder than your ad report makes it look

Start with what the person actually did. They were scrolling. Something held them for four seconds, they handed over a name and a number that were already on the screen, and they went back to scrolling. They were not shopping, they did not compare you against anybody, and there is a reasonable chance they did the same thing to two of your competitors inside the next minute. Why a form inside the feed produces that state, and why a landing page produces a different one, is a separate argument with a separate answer and it is written up at aevin.dev/blog/facebook-lead-ads-vs-landing-pages. What matters here is the state itself, because everything below is a response to it.

An hour later, consider what is left. Not your business name, which they never read. Not the wording of the offer, which they skimmed. What survives is a picture and a want: the photograph of the finished kitchen, the phrase about same-week appointments, the fact that they wanted the thing. Your job on the first contact is not to be remembered. It is to be recognized quickly enough that the person does not hang up while they work out who you are.

There is a second force working against you and it has nothing to do with memory. Your number is not in their phone. In the United States a call from an unrecognized number arrives wearing whatever label the carrier's analytics decided to give it, and a small business number that has just started making a burst of short outbound calls is precisely the pattern those systems exist to flag. So you are not only racing a fading memory, you are arguing with a screening habit. That, rather than any preference about channels, is the real argument for texting alongside calling.

What is actually known about response speed, and what is just repeated

Nearly every article on this subject rests on the same three numbers. Answer within five minutes and you are twenty-one times more likely to qualify the lead. Seventy-eight percent of customers buy from whoever responds first. It takes eight touches to reach a prospect. They travel without dates, without sample sizes and usually without a link. Before repeating any of them here, we went looking for where each one came from. This is what came back, checked on September 12, 2026.

Two of the three are real research and the third appears not to exist. That does not make the two of them usable. The 2007 study describes six business to business companies buying dialer software in the year the first iPhone shipped, and the document quotes its own author warning that the pattern is only visible once the six are pooled together, because a request for a price quote and a whitepaper download behave nothing alike. The survey half of the same document is blunter still: asking marketers when to call back produced, in its authors' own words, no statistically significant answers at all. And the Harvard Business Review article that people cite as the grown-up version of this work sits behind a paywall, with no numbers anywhere in the part that is free to read.

None of that means speed does not matter. It means the multiplier is borrowed, and a borrowed multiplier is worth nothing in a planning conversation. Nobody has measured your business, your offer or your city. If you want a defensible number, it is a timestamp exercise on your own records rather than a citation, and the two articles this one sits beside already describe how to run it.

The mechanism, without the multiplier

Three things are working in that first hour, and all three can be described without a single statistic. The 2007 document has a name for the first one, and the name is a better contribution than its multiplier was: presence. The person filled in the form on a phone they are still holding, so a call now reaches a hand and a call tomorrow reaches a voicemail box. The second is exclusivity. If they enquired with three businesses, the first real conversation gets to define the problem, frame the price and take the slot, and the other two arrive at a decision already half made. The third is identity, and it compounds: every hour that passes makes your call harder to explain and more like a cold call, until eventually that is exactly what it is.

That is the entire case, and it is enough to act on. Sooner is better, the gain is front loaded, and where your own curve falls off is a question about your own records rather than about anybody's research. One floor is worth knowing before you promise anybody five minutes. Meta's developer documentation describes lead webhooks as real time pings that occur on events with a delay of up to a few minutes, on a page Meta dates to May 21, 2026 and which was read here on September 12, 2026. Part of your five minutes is spent before you are even told that the lead exists, and no tool of any kind changes that part.

The order of touches, and how many are worth making

Call first. Not because calling is more polite, but because it is the only channel that can finish the job inside the window where the interest exists. An answered call can qualify, price and book in four minutes, and no message thread has ever done that. The objection that nobody answers the phone any more is largely true of cold calls and much less true of a call arriving minutes after somebody asked to be called.

If the call is not answered, leave a voicemail and then send a text from the same number, in that order and inside the same minute. The text is not a second try at the same thing. It does two jobs the call could not do. It converts your number from unknown to identified, so that your second call is answered by somebody who knows who is ringing. And it hands a busy person a way to reply at a traffic light. A missed call with no text behind it is a wasted attempt, because it leaves nothing behind.

Email is the third channel and the slowest, which is exactly what makes it right for everything speed cannot carry: the quote, the photographs, the availability, the thing they will forward to a partner before deciding. Send it after the first call attempt rather than instead of one, and treat a reply to it as a live lead rather than as correspondence.

1. 1.Day one, as soon as anybody is free: call, voicemail, text. Three touches, about ninety seconds of work, one entry in the log.
2. 2.Day one, later: a second call at a different hour. The first attempt tested one hour of one person's schedule and nothing else about them.
3. 3.Day two: a call, and a short text that says something new. Repeating your first message teaches the recipient that ignoring you is free.
4. 4.Day four: an email carrying the thing they actually wanted, a price range or an opening in the calendar, plus exactly one question.
5. 5.Day seven: one final attempt that says it is the final one, and then stop and mark the lead. The sentence that says you will close the file gets more replies than the four before it, and it is also simply true.

Five or six attempts across a week is a defensible starting shape rather than a rule, and the table above is the reason it is phrased that way: the eight touch figure in circulation was measured on cold outbound and describes somebody who never asked for your call. What matters more than the count is the distribution. Front loaded and finite beats even and open ended every time, because the same number of attempts spread over three weeks costs the same and reaches fewer people, and a sequence with no ending turns into a queue nobody works.

What the first message says, and what it leaves out

Open with the ad rather than with the form. Somebody who does not remember submitting anything hears "you filled in our form" as an accusation, and hears "you tapped our ad about same week bathroom fittings" as a reminder. That one piece of advice already appears at aevin.dev/blog/facebook-lead-ads-not-showing-up, where it belongs to a larger argument about keeping the ad, ad set and campaign ids on the record in the first place. It is repeated here in a single sentence because everything else in this section depends on it.

• –Who you are and the name of the business, inside the first five words, because the listener is still deciding whether this is a scam call.
• –What they responded to, described the way the ad described it rather than the way your price list does.
• –Why you are calling now. "You asked us to get in touch a few minutes ago" is a complete and honest reason, and it removes the sense of intrusion entirely.
• –One question, and a question that is easy to answer out loud. A person will say a zip code or a day of the week to a stranger. They will not say a budget.
• –A specific next step with a time attached. "Let me know" is not a next step, it is the work handed back to them.

The list of what to leave out is the same length and it matters more, because every item on it is something people write without thinking.

• –Prices nobody asked for. A number with no context is something to compare you against, and you are not there to defend it.
• –A booking link as the opening move. It asks a stranger to do administration for you before you have given them a reason to.
• –The words "just following up" and "checking in". Both announce that you have nothing new to say, which is usually accurate and never useful.
• –Anything that assumes they remember a previous message. Every touch has to stand on its own, because most of them are the first one the person actually reads.
• –More than one question. Two questions in a text message reliably produce zero answers.

The voicemail deserves its own fifteen seconds of planning, because most of them are deleted before the second sentence. Name, business, the ad they responded to, your number spoken slowly, and one sentence saying a text is coming too. That last clause is the entire point of leaving the voicemail, because the text is the thing they will actually act on.

Telling a junk lead from one you simply never reached

Every account collects leads that were never going to be anything, and confusing those with leads you merely failed to reach is the most expensive bookkeeping error in this entire process. At the end of the week the two look identical, because neither one answered. They are opposites, and which of the two you write down decides what you will later teach the ad platform.

Real junk has visible shapes. A mistap produces a submission with no intent behind it, which is why some of them arrive seconds apart from the same person. A submission from an ad preview is your own test, and Meta's developer documentation names it as one of three reasons a lead can arrive with no ad id attached, the other two being a lead generated from organic reach, where the export sets is_organic to 1, and a person requesting the leads who has no advertiser privileges on the ad account; that page was read on September 12, 2026 and Meta dates it to May 21, 2026. A prefilled email that bounces is usually a real profile address that went stale years ago rather than a fake one, which is a property of the instant form itself and is covered in the comparison article. And somebody outside your service area who wants something you do not sell is not junk at all. That is a targeting result, and it should be counted as one.

The action for junk is to mark it, never to delete it. A deleted row is a lead you paid for and then hid from yourself, and by the end of the month it has quietly removed the evidence that one ad set is producing all of them. Mark it, leave the ad id on it, and let it sit in the same report as everything else. Junk concentrated in one ad, one placement or one hour of the day is a finding, and the finding is worth considerably more than a tidy board.

Record the reason as a short fixed list rather than as free text: wrong area, wrong service, no budget, duplicate, test, never reached. Six options somebody can pick in one tap will get filled in. A text box will not, and a lost reason nobody filled in is the same as no reason at all. Keep never reached separate from every other option on that list, because it is a fact about your week rather than about the lead.

A pipeline that shows you where leads get stuck

Stages are not decoration and they are not a report. A stage is a queue with a rule for leaving it, and the whole point of having any is that a lead which has stopped moving becomes visible without anybody going looking. Counting leads by stage in order to work out whether the advertising pays for itself is a different exercise with a different answer, and that one is at aevin.dev/blog/cost-per-lead-vs-cost-per-acquisition. The table below is about the working board rather than the arithmetic.

Five stages is enough, and the restraint is the point. Every stage you add is a stage somebody has to drag a card into, and boards with eleven columns finish the month with everything still in column two. What makes the whole arrangement work is not the number of stages but a clock on each of them, because a stage without a time limit is a place to put things rather than a step in a process.

Why the outcome has to go back to the ad account

Meta optimizes toward the last event it can see. If the last event it can see is a form submission, it will go and find you more people who submit forms, and it will get better at that every week it runs. This is not a malfunction. It is the system following the only instruction anybody gave it, extremely well.

The missing instruction is the outcome. Telling the platform that this lead qualified, or that this one became a customer for $2,400, changes the target from volume to the thing you actually wanted. How that transmission works, which identifiers get hashed, how events are deduplicated against the ones your website already sent and what to check after it is switched on, is a full article of its own and it is at aevin.dev/blog/facebook-conversions-api. What belongs here is the half that depends on your follow-up rather than on your plumbing.

And that half is uncomfortable. The quality of the signal you send back is capped by the discipline of the work above it. If a third of last month's leads were never actually called, then every one of them you close out as unqualified is a falsehood you are now teaching to a machine learning system with your budget behind it. It will learn to avoid people who resemble the ones you never got round to phoning, which is a description of your Tuesday rather than of your market. Outcome data out of a sloppy pipeline is worse than none, because no data leaves the platform optimizing for volume, while wrong data sends it somewhere specific and wrong.

There is one more reason not to send only the wins. A closed sale is the end of a long road and there are very few of them, which leaves delivery with almost nothing to learn from; the stage before it happens far more often and arrives weeks earlier. Sending both the qualified event and the purchase gives the system a frequent signal and an accurate one, and the frequent one is what it can actually act on in the first month. Below a certain weekly volume none of this will move delivery at all, and that threshold is the same one that decides whether a campaign can be optimized for purchases in the first place. Send the outcomes anyway. They are worth having as your own record long before they are worth anything to the auction.

What you are actually allowed to do, in the United States

Advice about speed runs into telephone regulation sooner than most people expect, and the relevant rules are short enough to read. Under 47 CFR 64.1200(c)(1) no telephone solicitation may be made to a residential subscriber before 8 a.m. or after 9 p.m., local time at the called party's location. The exemption matters as much as the rule: the same section defines a telephone solicitation so as to exclude a call made to a person with that person's prior express invitation or permission, which is normally what a submitted form is. So the late evening callback is not usually the violation people assume it is. It is still a bad call to place at eleven at night, for reasons that have nothing to do with the regulator.

The rule with real teeth is the one about stopping. 47 CFR 64.1200(a)(10) says a called party may revoke consent by any reasonable method; it lists stop, quit, end, revoke, opt out, cancel and unsubscribe as revocations in themselves when sent in reply to a text; it says other words count too if a reasonable person would understand them to convey the request; and it requires that a revocation made in any reasonable manner be honored within a reasonable time not to exceed ten business days. "Please do not text me again" is not on the list and is plainly a revocation. Build the sequence so that one person can stop it in one action, because the alternative is an automated cadence that keeps running after somebody asked it not to.

There is exactly one five minute rule in this subject with a citation you can check, and it is not the famous one. 47 CFR 64.1200(a)(12) permits a single text confirming that somebody has opted out, and states that if that confirmation is sent within five minutes of receipt it is presumed to fall within the consumer's prior express consent, while a slower one obliges the sender to show the delay was reasonable. The five minutes that is actually written into law is about how fast you confirm that a person wants you to stop. These paragraphs were read on eCFR on September 12, 2026, against a Title 47 the site dated current to September 10, 2026. They are the text of the rule rather than legal advice, and automatic dialing systems, prerecorded voice and bulk texting sit under separate paragraphs of the same section.

None of this requires you to buy anything

Everything above works in a spreadsheet. Two timestamp columns, a stage column, a reason column and the discipline to fill them in will beat most software, because the failure this article describes is almost never a tooling failure. It is a nobody-owns-it failure, and a tool nobody opens is exactly as useful as a tab nobody opens.

Where software earns its place is the part a person cannot do, which is being awake at the moment the form is submitted. Aevin, our own product at $97, $197 and $497 a month, subscribes to your lead forms directly, so a lead lands on a board rather than in an inbox, with the ad, ad set and campaign ids already on it. Its pipeline is the five stages in the table above, they can be renamed and reordered, and the rules read the stage flags rather than the stage names: a lead moved onto a stage marked qualified sends a qualified lead event, and a lead moved onto the won stage sends a purchase with a value and a currency, with email and phone hashed before they leave. Two limits worth saying plainly. Nothing changes in a connected ad account without a person approving it, so what a full pipeline buys you is a proposed change and not an executed one. And the tool will not phone anybody for you: the sequence in this article is still a person, and the product's job is to make sure that person is looking at the right lead first.