Articles · Money

Three months with an agent: "better" is not the same as "paid off"

12 October 2026

At a three-chair dental clinic, the patient neither turns up nor gives notice for 11.4% of appointments. Among those reminded the day before, 5% fail to turn up; among the rest, 17%. The conclusion writes itself: install a reminder bot and no-shows will fall by two thirds. But under those numbers the report adds a caveat: "this proves a link, not a cause". An appointment is marked as reminded only when the patient was actually reached, and those may simply be the more reliable people. The clinic is invented — it is one of our sample reports; we test our audit agent on companies like it, and the agent wrote the report.

Three months after any agent goes live, an owner is looking at the same kind of pair: before and after. The temptation is the same too — to put the difference down to the agent. Below are four traps in that comparison and a way round each.

Trap one: comparing people who were different before the agent arrived

5% and 17% are not "with a reminder" and "without one". They are "those who were reached" and "everyone else". The second group holds both the people who did not pick up and the people who were never called. On Mondays 28.6% of appointments get a reminder and 18.4% are no-shows: those patients need to be called on Saturday, and on Saturdays the reminder calls never get done.

After an agent goes live, the trap looks like this: on the enquiries it handled alone, the result is better than on the rest. Our own website assistant answers from knowledge cards, and whatever the cards do not cover it passes to a person. If we compared its questions with the ones that went to a person, the person would always look worse: by design, the hard ones go to the person.

The way round. Judge on one number for everybody, before launch and after. At the dental clinic that is the no-show rate across all appointments, 11.4%. If the bot works, it will fall. If it has not fallen, no amount of "5% against 17%" will help. And a second point from the same report: in the reminder field, distinguish "not called" from "no answer", so that you can see whom the reminder failed to reach and who was never given one.

Trap two: the total hides two flows running in opposite directions

A fitness club (also invented) sold 277 new memberships in a quarter. In the same quarter, 270 ended. The owner saw the total — the club is not growing — and decided people were giving up before their enquiries were answered. Sales were fine. The club was losing at the other end.

277 came, 270 went. On the bottom line that is "plus seven", and it shows neither the sales nor the losses.

Now take the same club with an enquiry assistant, three months on. Membership is up, and the assistant gets the credit — though it may simply be that fewer memberships ended this quarter. Membership is flat, and the assistant is written off as useless — though it may have added sales while departures rose by the same number. The difference between two flows tells you nothing about either of them.

The dashboard may also be counting the wrong thing. The club's showed 274 lost memberships where 270 had ended: members who had only frozen their membership were counted as lost.

The way round. Before launch, write down which flow the agent is supposed to move, and watch only that one. An enquiry assistant answers for new memberships, not for the number of members. Losses go on a separate line, by reason: ended by notice, ended by the system, frozen.

Trap three: the season does the agent's work for it

A training centre (invented; here is the report) gets 71 enquiries in a normal week. From 17 August to 13 September it got 165 a week. Of the enquiries from those four weeks, 22% are still marked as new; the rest of the time the figure is 3.5%.

Put an agent in here in mid-September. A month later its report will carry the line: share of stuck enquiries down from 22% to 3.5%. It would have fallen without the agent — enrolment was over.

In three months the season has time to change: holidays, enrolment, public holidays, the year end. Compare one quarter with the quarter before and you are comparing the agent and the calendar together, with nothing to separate them afterwards.

The way round. Compare with the same months last year. If you have no figures for last year, leave part of the work without the agent for the same period: one branch, one channel, one shift. Both parts live through the same season, and the difference between them is more honest than the difference between spring and summer.

Trap four: the mix of enquiries changed, not the quality of the work

A travel agency (invented; the report is here) switched on Instagram advertising, and the owner saw more enquiries and no more sales. The owner's conclusion: the managers sell badly. The data say otherwise: of the people who come in through Instagram ads, 3.2% buy; from all other sources, 16.2%. Of the 1,578 enquiries in the period, 662 came from Instagram and produced 21 sales.

The overall conversion rate would have fallen whoever the managers were: what changed was who was writing to them. There were more people who ask "so how much is it?" and vanish. With an agent this cuts both ways. Switch on advertising at the same time, and the overall rate sags — the agent gets the blame. Switch off a weak channel, and the rate rises — the agent gets praise it has not earned.

The way round. Work out the rate within a single source: compare Instagram enquiries with Instagram enquiries, referrals with referrals. If nothing has moved within each source, the agent has changed nothing, whatever the overall total shows.

The same report has a simpler problem. The owner's claim — "twice as many enquiries, the same sales" — could not be checked at all: the firm had kept no figures from before the advertising. No "before", no comparison.

A clean difference is still not payback

The traps strip out false wins and false failures. What is left is set against the bill: set-up, three monthly payments, and your own people's time spent checking the agent's work — on a line of its own. What that bill consists of in full is in the article on the five lines in an agent's price.

Set-up is not earned back in the first good month. All the months together earn it back, and the count starts from the beginning: how much has gone out since day one, and how much has come in. The agent has paid for itself when the second sum catches up with the first, not when a monthly report first shows a plus.

What to do today, even if you have no agent yet

1. Write down the "before". The one number you will judge by, for the last three months and for the same months last year. Note where it came from and who worked it out. Three months from now there will be nothing to rebuild it from.
2. Split the total into flows. In and out, sold and refunded, each source separately. Underline the one flow the agent is supposed to move.
3. Start a "what else changed" list. Advertising, prices, a new member of staff, holidays, the rota — one line each, with a date. It will explain the jumps that would otherwise be credited to the agent.
4. Decide which part of the work you will leave without the agent. A branch, a channel or a shift. The season cannot spoil that comparison.

All articles