Must read stuff · Essay
The card taped to the clinic monitor
A clinic's front desk spent twenty minutes each morning reading tomorrow's appointments off one screen and typing them into another. I automated the step and it worked. What the automation also removed was the reading that caught a known quirk before it reached a patient, and nobody had ever written that part down.

The front desk at the clinic had two monitors and one chair, and the chair was occupied by a woman named Ingrid who had been there for eleven years. I had been called in about the two systems on those monitors, which had never quite agreed with each other.
They had never agreed on their own. What they had was Ingrid. Every morning she opened the reminder service in one browser tab and the booking system in another, read the list of tomorrow’s appointments off the first, and typed confirmations into the second. It took her about twenty minutes. She did it before the first patient arrived, with the radio on low and a cup of coffee going cold beside the keyboard.
The twenty minutes nobody had budgeted
When I mapped the workflow, the twenty minutes were not in any document. They were not in the contract with the reminder service, which promised to send appointment notifications. They were not in the booking system’s manual, which assumed the front desk entered appointments directly. The twenty minutes were the seam between two systems that each believed they were the only one, and Ingrid was the seam.
The good part of what she did was invisible because it worked. Patients got their reminders. The schedule matched reality. When a patient called to move an appointment, Ingrid moved it in both places before lunch, and the two systems stayed in step because a person walked between them.
I proposed an integration. The reminder service had an interface, the booking system had an interface, and the two could be made to talk. It was a small project with a clear seam and a measurable result. The clinic approved it, I built it, and it worked. Reminders went out automatically. Confirmations landed in the booking system without anyone typing them. Ingrid got her twenty minutes back.
What the seam was doing besides typing
I did not think about the seam again until Ingrid took a two-week leave in the spring, and the clinic called me because patients were arriving on the wrong days.
The integration was running. Reminders were going out. The problem was upstream of the integration, in a place I had not looked. When a patient called to reschedule, the front desk changed the appointment in the booking system, and the reminder service was supposed to pick up the change. It usually did. But the reminder service had a quirk: a change made within a certain window before the appointment did not propagate, and the reminder went out for the old time. Ingrid had known this. She had learned it years ago, probably by being embarrassed once, and she had handled it by checking the reminder list against the booking system every morning and fixing the mismatches by hand. The twenty minutes were not just typing. They were the place where a known quirk got caught before it reached a patient.
When she was gone, nobody caught it. The integration did exactly what it was built to do. The quirk did what it had always done. The difference was that the person who knew about the quirk was on a beach somewhere, and the knowledge had never been written down, because it was not a procedure. It was a habit.
The case for writing it down
The strongest argument against my own hand-wringing is that this is a documentation problem, not a technology problem, and documentation problems have solutions. Ingrid’s knowledge should have been a checklist. The quirk should have been reported to the reminder service, or worked around in the integration, or at minimum written on a card taped to the monitor. The clinic had a single point of failure and it was a person, and the right response is to remove the single point of failure, not to romanticize it.
That argument is correct, and I have made it myself to other clients. The integration was still the right call. The twenty minutes were still worth returning to Ingrid, who used them to call patients back about test results, which is a better use of a person than retyping dates.
What I got wrong
What I got wrong was the assumption that the seam was only typing. I looked at the workflow and saw a manual step, and I removed the manual step, and I did not ask what else the manual step was doing. Ingrid was not just moving data. She was reading it, and the reading caught things the integration could not catch because nobody had told the integration to look.
I could have asked her. I did not, because the workflow diagram made the answer look obvious, and because asking a busy person to explain her habits feels like wasting her time. It would not have been a waste. It would have been the whole project.
The card on the monitor
I went back in June to add the quirk handling to the integration, which took an afternoon. While I was there, Ingrid showed me the card she had taped to the monitor after she got back, a handwritten note about the rescheduling window, in her handwriting, with the date she had learned it. She had written it down herself, without being asked, because she had come back from leave to a mess and did not want to come back to another one.
The card is still there. The integration handles the quirk now, so the card is out of date, but she has not taken it down. I asked her about it once, and she said it was a reminder of the two weeks, and then the phone rang and she answered it, and I let it go.
#Essay #smallbusiness #workflow #automation #workflow automation #tacit knowledge #small business systems #integration #single point of failure