Articles · Case files
Looking in the wrong place: nine traps we set for our audit agent
12 October 2026
A fitness club sold 277 new memberships in a quarter and did not grow. The owner decided enquiries were being answered too slowly and was looking at an assistant that replies round the clock. The data said something else: 270 memberships ended in the same quarter, and some of them were ended by the system itself — after two failed direct debits, with nobody phoning the member first. This club does not exist: we invented it.
The same goes for the other eight companies in this article. We test our audit agent on companies like these: the right answer is known in advance, and the cause of the loss is deliberately hidden somewhere other than where the owner is looking. The agent wrote the reports — here is the report on the club, and the rest are linked below. None of this is a market statistic: the cause was somewhere else nine times out of nine because that is how the test is built.
They are worth reading for another reason. The traps are all built the same way, and you can use that pattern to check your own company.
In all nine cases the data contradicted the owner
Dental clinic, three chairs. The owner's view: patients have become irresponsible, so the clinic needs a bot and a voice assistant. The data: 13 patients out of 2,430 missed an appointment more than once. Meanwhile fewer than half of appointments — 46.9% — are marked as reminded. With a reminder, 5% of patients fail to turn up; without one, 17%. The dental report
Road haulage, 16 trucks. The owner's view: the drivers take their time, so the firm needs a route planner. The data: 75% of overtime hours fell on days when the truck had a stop at one of two sites with a queue, although such days make up only 33% of the total. And the waiting time hardly ever reaches an invoice: 13 lines in three months, against 379 jobs with a stop at those sites. The haulage report
Repair shop. The owner's view: the staff are glued to their phones and slow to reply, so the firm needs a bot. The data: during opening hours, half of all enquiries get a reply within 9 minutes. What gets lost is website requests: all 494 reached the CRM with no one assigned, and 231 never got a reply. The repair-shop report
Building and renovation. The owner's view: the margin is going on inaccurate estimates, so the firm needs AI estimating. The data: contract work runs 2% over the estimate. The rest is extra work, and it often never gets invoiced: 68 completed changes were closed without an invoice, which is 499.5 hours of work. The construction report
The fitness club from the first paragraph. Of the 173 members whose membership ended at the end of April or May, 48 tapped their card at the gate afterwards and were refused. The owner believed the front desk was ringing members whose debit had failed, working from a Friday list. There has been no list since last year, and nobody at the club knows how many of the 270 memberships the system ended. The club report
Travel agency. The owner's view: the managers sell badly, so the firm needs an assistant to put together tour shortlists. The data: a tour the client has paid for cannot be booked until the accountant marks the payment as received, and she is in the office on Mondays and Thursdays. While the tour waits, the places sell out or the price goes up: 19 enquiries that had already been paid for were lost, and with them 3,586,700 KZT of commission over 13 weeks. The travel agency report
Cleaning company. What the owner was promised: a scheduling agent would cut payroll by 20%. The data: overtime is 8% of what the firm pays its cleaners; the rest is guaranteed hours. Each week is copied from the last, and 60% of overtime hours are matched by short days for the same cleaner in the same week. The cleaning report
Training centre, two branches. The owner's view: the receptionists cannot keep up with enquiries, so the centre needs an AI receptionist. The data: of the 500 people who came to a trial lesson, 213 signed a contract. There is no rule that anyone calls after the trial — "we call when we remember" — and 76 people are on the books as neither a contract nor a loss. The training-centre report
Recruitment agency. The owner's view: CVs are screened too slowly, so the firm needs AI screening. The data: every application is read the same working day. The losses come later: of 881 CVs sent to clients, 280 ended with the candidate withdrawing. No answer date has been agreed with a single client. The agency report
The step the owner suspected is working fine
Nine minutes to reply. An application read the same day. An estimate that is 2% out. Drivers' hours taken from the tachograph rather than from anyone's say-so.
Wherever the suspicion could be checked, it was cleared. At the travel agency there was nothing to check it with: there is no log of calls and messages, and the data are silent about the managers.
An owner looks where everything is in plain view: who is on their phone, who has the most overtime, whose estimate did not add up. The loss sits where there is nobody to look at.
The loss sits in a seam nobody owns
A seam is the place where one person has finished their part of the job and the next has not yet started theirs. The nine traps contain three kinds.
- Between people. The travel agency's manager has sold the tour, and the accountant will mark the payment on Thursday. The teacher has given the trial lesson, and calling afterwards is nobody's job. The recruitment consultant has sent the CV to the client and is waiting.
- Between systems. The website creates a request in the CRM with no one assigned, and it can only be seen through a special filter. The driver app sends nothing to the invoice. The club's system ends a membership by itself, and the reason appears in no report.
- Between "done" and "invoiced". The carpenters did the extra work the same day, and nobody got round to pricing it before the final invoice went out. The truck stood in a queue, and a waiting-time line appears only if the driver phoned in and the planner had time.
At the dental clinic and the cleaning company the seam is one of time. Reminding patients falls to a receptionist who is alone on her shift and rings "when there's time". At the cleaning company each week is copied from the last, and nobody checks the planned day against the cleaner's contract hours.
At a seam nobody is at fault, so nobody complains and no report shows it. What the owner does have is a firm belief that everything there is in order.
The travel agency's owner believed payments were confirmed within the hour. All 189 first bookings were made on a Monday or a Thursday.
The haulier was sure waiting time went straight from the app to the invoice. The club's owner was sure members were being rung from a list. Three beliefs, and not one survived the first check.
The tool the owner came for would have sped up the wrong step
AI estimating would have refined estimates that are already accurate. AI screening would have sped up reading that is already finished the same day. A route planner would have reshuffled routes while the truck stood in the queue at the warehouse.
The first steps in these reports are rules, not the purchase of an agent. When an agent's turn comes, and which work to hand it, is the subject of our article on five signs of "yes" and three of "no". For the travel agency: confirm payments every day, which costs 3,000 KZT a month in bank charges. For the repair shop: put one named person in charge of website requests, free. For the builder: price extra work on the same day every week, also free.
A free rule still costs time. At the building firm, pricing extras goes up from 11.78 to 19.56 hours a month across all the project managers. The report says so plainly: the rule means more work, not less.
In a real company the owner may well be right. How often, we do not know. The traps do not test owners; they test the agent — whether it follows the data or the theory of whoever commissioned the review.
What to check in your own company today
- List your seams. The places where work passes from one person to another, or from one system to another. Ask two questions of each: who is responsible, and by when. "When we get round to it" means nobody owns it.
- Test one belief about timing. "We call back within the hour." "The invoice goes out the day the job is handed over." Open the dates in your system for the past month and see on which days and at what times it really happens.
- Compare "done" with "invoiced". How many extras, waits or call-outs were logged in the month, and how many lines for them reached an invoice. That is two numbers from two of your systems.
- Take your last ten lost customers. Against each one, mark the step at which they dropped out. Not the person — the step. If six out of ten dropped out in the same place, that is where to look.
Want the same look at your own company? The audit is free for now — it is an experiment. Show us how your work runs and get a report: where time and money are lost, and where AI would help and where it would not.
Request an auditSee sample reports
“Audit” here means a review of how the company works, not an audit of financial statements. The companies in the examples are invented; the reports on them were written by our AI agent.