Self-Service Booking Conversion and the Drop-Off You Cannot See

Why this matters

A shop turns on self-service booking, watches the confirmation rate come back high, and concludes the booking flow works. The rate is real and the conclusion does not follow, because everyone who opened the form and walked away is not in it. The measured funnel starts after the hardest step, so a high number is a statement about people who already committed, not about the booking experience that decides whether they commit. Meanwhile the one genuinely diagnostic read in this data, who cancelled and how long after booking, is usually sitting unread beside it.

What the rate is made of

• The numerator is bookings created through the form in the window that reached confirmed.
• The denominator is bookings created through the form in the window. Created, not attempted.
• Cancellations sit outside the numerator and inside the denominator, which is what makes the two figures complements rather than independent readings.
• Everyone who reached the form and left is in neither. No record exists, so the drop-off cannot dilute the rate, appear as a row, or show up in a total.
• A confirmed booking is confirmed forever. A booking that was confirmed and then nobody was home still counts as a success here, because the field it depends on stopped moving at confirmation.

The fourth point is structural and the fifth is a definition you can tighten. Take them in that order.

The funnel starts one step too late

The full path from interest to a job on the truck has at least five steps: somebody reaches the page, reads enough to try, starts entering details, submits, and then the booking is confirmed. Your rate measures the last one. Everything that decides whether a stranger becomes a submission happens before the first record exists.

That has a specific consequence worth stating plainly. A high confirmation rate is not evidence that the booking flow is working, and a low one is not evidence that it is broken. Confirmation is mostly decided by whether you can honour the slot, which is a scheduling question. The flow is decided by what happens on the page, which is invisible in this data by construction.

It also means the rate is self-selecting in a way that gets worse as the form gets harder. Ask for more detail, more required fields, a longer address block, and fewer people finish, but the ones who do are more committed, so the confirmation rate can rise while the number of jobs falls. A shop optimising for this number would be led, step by step, to make the form harder.

Three ways to see the step you cannot measure

None of these need anything beyond what an office already does. The first one is the only one that produces a number.

Ask every new phone booking one question. "Did you start to book online first?" It costs three seconds and it turns an invisible population into a countable one. In the month below it produced the entire finding.

Watch the job mix by channel. If bookings through the form are almost all one job type while the phone carries the range, the form is not serving your demand, it is filtering it. That is a content problem on the page rather than a flow problem, and no confirmation rate will ever surface it.

Watch the clock on submissions. Bookings that cluster in the evening against calls that cluster during the day tell you the form is doing after-hours capture, which is genuinely valuable and also means the daytime abandonment you are worried about may be small.

Here is the month. The form created 84 bookings, of which 71 confirmed and 13 cancelled, so the reported figure is 71 of 84, 84.5 percent.

Then the question at the phone. Of 96 new phone bookings that month, 22 said they had started online and given up, 22.9 percent of the phone bookings. So the observable top of the funnel is 84 submitted plus 22 known abandoners, 106 attempts:

• Submitted, out of the attempts you can see: 84 of 106, at most 79.2 percent.
• Confirmed, out of the attempts you can see: 71 of 106, at most 67.0 percent, against the 84.5 percent reported.

Both of those are ceilings and both are one-sided. Anybody who abandoned and then called a different shop is still nowhere, so the real figures are lower than these and there is no way to find out by how much. The useful move is not the precise number, it is that 67.0 and 84.5 are 17 points apart and every decision was being made off the higher one.

The cancellation column is the read you do have

Cancellations get reported as a share and then ignored, which is backwards. The share alone tells you almost nothing. The composition tells you which department has a problem.

Two cuts do the work: who cancelled and how long after the booking. Same month, the 13 cancellations:

Six of the 13 cancellations were the shop, 46.2 percent of cancellations and 7.1 percent of the 84 bookings created. Those are two different bases for the same 6 bookings and they support different conversations, so say which one you mean every time.

Five of those 6 shop cancellations landed more than 24 hours after the customer booked, and that is the dominant cell of the four. The timing is the signature of a slot that was offered and then could not be held, which is a capacity and availability problem rather than a booking-page one. The shop is writing cheques against capacity it does not have, and the customer experiences that as being told a time and then told it is gone, which is worse than never having been offered a time at all.

The other three cells are small and ordinary. Second thoughts inside 24 hours, or a second shop confirming first, is answered by confirming faster and putting the arrival window and the technician's name in the confirmation. A customer cancelling later is life, and needs nothing unless it grows. The one shop cancellation inside 24 hours says published availability is not being refreshed often enough.

Set a working gate on the dominant cell: shop-caused cancellations above about 5 percent of bookings created means stop publishing that much availability. This month is 6 of 84, 7.1 percent, which is over it. Below the gate, cancellations are mostly the customer's and mostly ordinary.

One more tightening while you are here. Of the 71 confirmed bookings, 5 were no-shows or could not be worked on arrival. So bookings that became completed work are 66 of 84, 78.6 percent, against the 84.5 percent the confirmation rate reports, and 66 of the 106 observable attempts, at most 62.3 percent. If the reason you track this number is to plan the schedule, 78.6 is the figure to track, and it is a different question from whether people can book.

What the wrong fix costs

An owner reading only the 84.5 percent has two plausible reactions and both are wrong in the same direction.

The first is to conclude the page is fine and stop looking. The second, if the rate dips, is to spend a month rebuilding the page. Either way the 5 unheld slots keep happening, and a better page makes that worse, because it sends more people into the same inventory.

Order it the other way. Fix what you publish before you touch what you show. Bring shop-caused cancellations under the gate first, because that is the failure the customer feels and it is the one you can see. Then work on the invisible step, starting with the one question at the phone, since a month of those answers costs nothing and produces a number.

There is a real cost to leaving this alone that the rate never shows: the 22 people who gave up online and phoned are counted in your records as phone leads. The online channel produced that demand and gets no credit for it, so the source table under-reports the channel while the confirmation rate over-reports its quality. Both errors point the same way, and a shop can defund the thing generating its demand on the strength of two numbers that are individually correct.

Where an abandoned booking shows up, and where it does not

The last thing worth having is a map of which of your numbers can see each of these people. Most of them appear nowhere, and the ones that appear somewhere often appear in the wrong place.

Read down the first column and the shape of the problem is obvious: the confirmation rate sees two of these five, and it sees them as the same thing, a cancellation. Read down the last column and you have the work list. Two of the five are only reachable by asking somebody, which is the honest answer for a shop without an analytics setup and a better answer than most analytics setups produce, because a person will tell you why and a record never will.

References

• See related: Lead Source Performance: Rank by Conversion, Not Volume, for the source table that mis-credits the last row above
• See related: The Satisfaction Score and Who Actually Answers a Survey, which owns the response and selection bias pattern
• See related: Lost Reasons Are Only as Good as the Field Being Filled, for recording why a customer cancelled
• See related: Collecting the Right Info at Booking, for what the form should and should not ask for