The Review Funnel: Sent, Clicked, and What Happens After

Why this matters

Your review funnel has an honest hole in it, and most shops do not notice where. You can count how many review requests went out and how many customers clicked through. You cannot count what happened next, because the review lands on somebody else's platform and your records end at the link. Every figure you compute from your own data therefore stops one stage short of the outcome you actually care about, and the stage it stops short of is the one that loses the most people. Treating a click as a review overstates the result by a factor of three or four and hides the fact that the last stage is where the leak is.

Three stages you can see, and what sits below the line

Lay the funnel out with the boundary marked, because the boundary is the point of the article.

completed jobs
     |
     v
  eligible after the quality gate     you decide this
     |
     v
  asks actually sent                  you record this
     |
     v
  clicked through                     you record this
     |
- - - - - - - - - - - - - - - - -     your records end here
     |
     v
  review posted                       somebody else's platform
     |
     v
  review visible to a buyer           the platform decides this

Two stages sit below the line. Nothing in your own records reaches them, and no amount of care with the three above will tell you what happened there. That is not a gap to be closed with better tracking, it is a property of where the review lives, and the only thing that closes it is counting by hand.

Stage one: who gets asked, and the gate before it

Every shop puts some gate in front of the ask, and there is a line running through the middle of that decision.

A sequencing gate is fine. A job with an open complaint, an unresolved rework visit, or a disputed invoice does not get a review request this week, because asking somebody to praise you in the middle of a fight is a worse idea than not asking. The job goes back into the pool once the complaint is closed, and it gets asked then. Nothing is excluded permanently, and over a quarter the eligible pool converges on the completed-job count.

A selection gate is not fine. Excluding the jobs that went badly, permanently, or leaving it to the tech's judgement on the day, is a different rule with a different result: it raises the score and removes the population that had something to say. Operationally it also blinds you, because the customers you stop asking are the ones whose experience you most need in the record. See related: The Satisfaction Score Held While the Repeat Rate Fell.

Two rules to hold while you write your own gate, each one attached to what triggers it:

• Offer nothing of value for a review, no discount, no gift, no draw entry, because incentivised reviews and the suppression or selective display of reviews you have collected are addressed in the United States by the Federal Trade Commission's Rule on the Use of Consumer Reviews and Testimonials, 16 CFR Part 465, which took effect in 2024 and binds businesses directly.
• Asking only the customers you expect to be happy is review gating, and in the United States that is a federal question before it is a platform one: the FTC treats it as deceptive under Section 5, and its Rule on the Use of Consumer Reviews and Testimonials, 16 CFR Part 465, in force since 2024, reaches the suppression and selective display of reviews and binds businesses directly. The major review platforms prohibit it in their own terms as well, which is the consequence most shops notice first and the smaller of the two.

Whether a specific incentive or a specific selection rule crosses a line is a question for your own attorney. What you can bank operationally is narrower and enough to work with: an ask that carries no offer and goes to every eligible customer is defensible on both counts, and it is also the version that produces a usable number.

Stage two: when the ask lands

Timing is the highest-leverage variable in the whole funnel and it costs nothing to change. Send within about 24 hours of the work being finished, while the tech is still a person the customer can picture. Tune it if your work runs long, but commit to a number; "when we get round to it" is how a batch goes out three weeks late with the customer wondering who you are.

Two exceptions, and both are about the work not really being finished. A job with a second visit booked gets asked after the second visit, not after the first. A job whose result the customer cannot see yet, where the system has to run through a season before they know, gets asked when they can judge it, and that is a genuinely later ask rather than a delayed one.

Stage three: what the ask says

Three things, and the third is the one shops skip.

• Name the work and the tech. "About the water heater replacement on Tuesday, and Danny who did it" gets read. A generic request from a company name gets archived.
• Name the platform and make the link do one thing. A customer offered three platforms picks none of them.
• Say what it is for, honestly. Something close to "it helps people nearby find us" is true and it is the only motivation available to you now that you cannot offer anything.

Keep it short enough to read on a phone without scrolling, and send it from a person rather than from a no-reply address, because a customer who hits reply with a problem instead of leaving a review has given you something more valuable than the review.

A quarter, counted with its bases named

One shop, one quarter, one platform. Every figure below names the base it is a share of, because this funnel produces four different denominators and mixing them is how a shop talks itself into believing it is doing well.

Two more ratios fall out, and both are worth writing down because they answer different questions:

• 11 of 151 asks sent, 7.3 percent. This is the one to use when you are deciding whether to send more asks.
• 11 of 214 completed jobs, 5.1 percent. This is the one to use when you are forecasting how many reviews a year of work at this volume produces.

Notice how far apart 25.2 percent and 7.3 percent sit. A shop reporting "a quarter of the customers we asked responded" is quoting the click rate and calling it a review rate, and the real figure is under a third of that. The gap between them is entirely below the line.

The hole at the end, and what lives in it

Between the click and a posted review sits a stage you do not control and cannot see. It is worth knowing what happens in there, because the causes call for different responses and only one of them is your fault.

• The platform wants a sign-in. The customer taps through, meets an account wall, and stops. This is usually the largest loss, though you cannot confirm that from your own records, because it happens below the line where your data ends, and there is nothing you can do about it except pick the platform your customers most likely already have an account with.
• The app wants installing. Same outcome, worse.
• They meant to and did not. Ordinary life. A single reminder, at most one, a week later, recovers some of these.
• They wrote it and the platform filtered it. Automated filtering removes a fraction of genuine reviews, often the short ones from new accounts. You will never be told.
• It posted and nobody noticed. More common than it sounds, and the reason the reconciliation below exists.

Because a customer who meant to write one often gets to it days or weeks later, a posted review can lag its click well past a month end, which means a month-against-month ratio will move for timing reasons alone. Compare on a trailing three-month basis. The same effect has a name and a fuller treatment elsewhere. See related: Estimate Conversion Rate and the Cohort Problem.

The one thing below the line you can still move

You cannot make a review post and you cannot make the platform show it. You can respond to it, and that is the only lever left after the boundary. It is worth using, because the reader of a review page is not weighing one review, they are reading how a shop behaves.

Commit to a number here as well: respond to every review, good or bad, within two working days. Keep the response short, thank the good ones by name without a sales line, and on a bad one name what went wrong, say what you changed, and offer to take the rest of it offline with a direct contact. Never argue the facts in public, and never reveal anything about the customer's property, their account or their payment that they did not put in the review themselves.

Track one number off it: the share of reviews carrying an owner response. That is a count over a base you fully control, unlike everything else below the line, and it is usually the first thing to slip when the shop gets busy.

Closing the loop by hand

The reconciliation is a monthly count you do yourself, on the platform, with your own eyes, and it takes about ten minutes. It is the only thing that connects your three visible stages to the outcome.

Record the date you counted and who counted, because the platform's own total drifts (reviews get filtered out later, and a review posted late appears in a month you already closed). Without a dated count you cannot tell a genuine change from a recount.

Then read the two columns against each other, and only these four movements mean anything:

• Clicks rise, posted reviews flat. The ask is working and the platform step is eating them. Check which platform you are sending to and whether your customers have accounts there.
• Clicks flat, posted reviews rise. Somebody is asking in person and it is working. Find out who, and what they say, and teach it.
• Asks flat, clicks fall. The ask itself stopped working. Something changed in the wording, the sender, or the timing, and it is usually the timing.
• Asks fall and everything below falls with them. The gate got wider without anybody deciding to widen it. This is the one that creeps: a sequencing gate quietly becomes a selection gate one judgement call at a time, and the funnel looks healthy on every ratio while the pool shrinks underneath it.

That last row is why the eligible count sits in the table at all. Ratios below it stay flattering while the top of the funnel closes, and the only number that catches it is the plain count of jobs you decided not to ask about.

References

• Federal Trade Commission, Rule on the Use of Consumer Reviews and Testimonials, 16 CFR Part 465
• See related: The Satisfaction Score and Who Actually Answers a Survey
• See related: The Satisfaction Score Held While the Repeat Rate Fell
• See related: Estimate Conversion Rate and the Cohort Problem