The UK Driving School has a sales channel that cannot be staffed. Its ads make the phone ring, and the people who could answer it are instructors in a moving car with a learner at the wheel. A missed call is rarely a message left; it is a pupil ringing the next school on the search results. The school’s real product is a full diary for each instructor, so every unanswered call is income lost twice.

When we asked the person who answers the phone today what worried him, the answer was not automation. It was accuracy: a machine confidently quoting the wrong price, promising an area nobody covers, or failing a caller with limited English whom a person would have understood. That fear became the brief. The receptionist had to be right, had to know when to stop, and had to let the owner check its work in a few minutes on a phone between lessons.

The school set the line itself. Prices, whether an area is covered, weekend availability and taking booking details are fine for the receptionist. Reschedules where the instructor is not free, a pupil who has just failed a test, complaints and refunds must reach a person. Those rules are written into the receptionist in plain language and enforced as rules rather than suggestions, so a caller who asks for a refund is put through, not reasoned with.

Prices were the hardest part to get exactly right, because a voice assistant that can do arithmetic will happily invent a total. The receptionist quotes the school’s per-lesson prices and its discount threshold as written and is forbidden from multiplying them into a package price nobody agreed. Area coverage works the way callers speak: by district name as well as postcode, across the dozens of neighbourhoods the school covers. Content guardrails are set to rephrase rather than end the call, so an ordinary question about declaring a medical condition to the DVLA does not get a paying caller hung up on.

Most call dashboards lead with volume. This one leads with the calls that need a person, at the top of the first screen, with the reason and how long the caller has waited, because that queue is the owner’s actual job. Every handled call is shown with its transcript beside what the receptionist quoted, checked and refused, so “what did it say, and was it right” takes one tap. A caller who switched language is labelled as an outcome, not hidden as an error.

The rest of the system grew from the same principle. Instructors mark themselves free or busy on a simple mobile calendar and move lessons with a tap, which replaces the paper diaries the receptionist could never read. Pupil records hold lessons, payments, balances and progress against the driving test competencies, and booking confirmations go out by email. A failed transfer cannot lose a caller: the handover is written down before the call is put through, so if nobody picks up, the caller hears that someone will ring back and the owner sees their name, number and words.

Every change to the receptionist is checked against a suite of 174 scripted test calls before it goes live: callers asking for prices, areas, availability, refunds and people, in the ways people actually ask. A change that breaks one does not ship. That is the practical answer to the owner’s worry about a machine getting things wrong; not a promise, but a test that runs every time.

Call recordings and transcripts are personal data, so a retention period is set, deletion runs on a schedule, and access is by role so an office user sees less than the owner. The same pattern is what we apply to any client whose callers are members of the public.

Any business whose people are too busy doing the work to answer the phone has the same problem: trades, clinics, salons, tutors, letting agents. The pieces are the same too: a receptionist with written rules about what it may and may not say, a live handover to a person, a calendar it can actually book into, and a console where the owner can check its work. We built this one for The UK Driving School and run it alongside them.