[ CASE STUDY · PHASE ONE DISCOVERY AUDIT ]
We sat at the front desk and counted.
A multi-site private healthcare group asked one question — what is our phone-only front door actually handling? In July 2026 Automatedge spent a week on site answering it: 689 telephone events logged and classified across four working days, alongside an in-person observed sample, a staff survey across every reception site, and a structured interview with the front-desk trainer.
Every figure on this page is labelled by how it was captured. Findings are aggregated and redacted for publication — no individuals, sites, or systems are identified.
[ 01 · THE METHOD ]
Every figure is labelled by how we know it.
The audit replaced opinion with evidence. Each claim carries its capture method and evidence class — and projections are marked as projections, never quoted as measurements.
Counted and timed in person by the Automatedge consultant at the front desk — 41 events in a single observed session, one call every 3.5 minutes sustained.
Contemporaneous call-by-call logs kept by eleven front-desk staff — roughly three quarters of the team — producing twenty-four logs across four working days.
Inferred under a stated rule — free-text call types classified from their notes under one keyword rule set, with every convention listed in the full report.
Extrapolated from measured data with methods and assumptions shown. Never quoted as a measurement — that is the whole point of the labelling.
[ 02 · THE BASELINE ]
The telephone is the front door, and it is saturated.
In the observed session: one call every 3.5 minutes sustained across the late morning, 22 calls in the busiest hour. Group-wide, the morning peak concentrates into the first hour of the day. These are answered calls only — the true demand is at least what was counted, never less.
689
telephone events logged and classified across four working days
43%
of everything logged was booking or rescheduling — the single biggest reason the phone rings
2.0 min
average handling time across 652 timed events
81%
of calls resolved at first contact by the front-desk team
105
calls landed in the 08:00–09:00 hours of the logged days — the morning peak
0.3%
of events were clinical escalations — 2 in 689. The load is administrative, not clinical
[ 03 · THE CALL MIX ]
Booking is the biggest single reason the phone rings. Equal-biggest is everything else.
Booking and rescheduling account for 43% of all logged events. General administration — transfers, requests for a specific team, callers for other sites, letter queries, status checks — accounts for another 38%. Exactly two events in 689 were clinical.
- Admin — transfers, team requests, letters, status checks38%
- Booking38%
- Prescription10%
- Results query7%
- Cancel / reschedule5%
- Type not recorded — logged under pressure2%
- Payment0.4%
- Clinical, escalated0.3%
689 events, processed under one uniform rule set. Free-text call types classified from their notes; every data convention is listed in the full report.
[ 04 · WHAT THE CALLS ACTUALLY WERE ]
The strongest evidence is the calls themselves.
Aggregated from twenty-four contemporaneous staff logs. Each pattern below is a demand the current front door has no other channel to absorb.
One letter, at least twenty calls.
A single administrative letter produced call clusters in four separate logs across three days — eight calls on one desk in a single day. A fully predictable communications event, with no message channel available to absorb it.
The confirmation gap, measured.
Calls that existed only because appointment details were unknown to the patient: callers who could not remember a time, who did not know which practitioner they were booked with, and patients ringing back after booking online just to check it worked.
Notification by telephone, one patient at a time.
Ten results-related entries in one desk’s morning. Seven "informed patient of…" calls in another log — three of them in three consecutive minutes. Repeated failed outbound attempts logged as "unable to contact": telephone-tag runs in both directions.
The switchboard effect.
Requests to reach a specific practitioner’s team recur across every log. Callers for other sites ride every queue. The phone is simultaneously the booking channel and the switchboard.
The online channel leaks back into the phone.
Online bookings cancelled by phone. Wrong-practitioner online bookings corrected by phone — "a few of these a day". Accidental cancellations at the confirmation step. An unguided online journey generates its own calls.
The long tail must reach a person — faster.
A 20-minute billing query. An eight-minute call from a relative. A seven-minute accessibility query. The calls that must always reach a human are exactly the ones a saturated line answers slowest.
[ 05 · THE SCALE ]
Two independent models bracket the true demand.
The audit captured demand only through the staff who logged, in partial windows. A participation-scaled model and a site-rate model now bracket what the whole group answers per day — and with four days of data, they converge.
340–680
answered calls per working day across the group — roughly 1,700–3,400 per week, which at the measured 2.0 minutes per call is approximately 57–113 staff-hours per week of pure telephone handling. The two models overlap at roughly 480–495 calls per day.
[PROJECTION]29%
phone-queue reduction demonstrated in NHS England’s Sussex general-practice trial of an AI triage front door — the external anchor. At only that deflection rate, roughly a fifth to a third of the group’s weekly handling time returns to patient-facing work.
[PUBLIC RECORD]Models, inputs and known biases are stated in the full report. Projections are never quoted as measurements.
[ 06 · WHAT ANY SOLUTION MUST RESPECT ]
The constraints came from the evidence — and from the staff.
The audit produced design constraints before it produced designs. Several came directly from the front-desk team, and are adopted in full.
Confirmation for every booking.
Whichever channel a booking came through, a confirmation message follows. The absence of this single layer was measurably generating daily calls.
Notify instead of dialling.
Results-ready and prescription-ready notices, and planned communications events, go out as messages with a reply path — instead of one outbound call per patient.
The phone stays open — and gets faster.
Elderly patients and carers were named by the team as permanent phone users. Automation absorbs the digitally able majority so the line answers sooner for everyone else. This constraint came from the staff, and it is adopted in full.
Humans in the loop, always.
Complex bookings keep a person in the flow, every workflow has a named clinical escalation route, and recall automation preserves a human sign-off before anything sends.
No single-person bottlenecks.
No clinic type should be bookable by exactly one named person. Process rules like that dissolve into the routing layer.
Patient data stays under the practices’ control.
Practice-approved extracts to practice-controlled infrastructure, under the applicable data-protection law, with a data processing agreement in place before anything moves.
[ 07 · WHAT THIS REPORT DOES NOT CLAIM ]
Measured before claimed. Labelled before quoted.
The purpose of Phase One was to replace opinion with evidence — which also means stating plainly what the evidence does not say.
Projected figures are labelled extrapolations with methods shown — replaced by telephony analytics and raw exports as those become available.
Staff-logged records are contemporaneous but not independently verified, and are labelled as such throughout.
Staff judgements about what "could be done online" answer a configuration question, not a capability question — they are treated as a floor, not a ceiling.
Automation does not create clinical capacity. It returns administrative hours and smooths routing; it does not add appointments.
No third party can write appointments into the group’s clinical platform today — so no vendor can honestly promise fully hands-off rebooking. Workable designs route patients to the group’s own booking pages, and are built to take advantage of a scheduling interface the day one ships.
[ 08 · THE COMMERCIAL MODEL ]
Break-even at 6.7% admin deflection.
Against a £1.2m annual admin benchmark, a modelled £80k year-one programme represents a deliberately modest hurdle. The point is not to promise savings — it is to make the operational value measurable. The audit above is what measurement looks like.
The benchmark and programme figures are modelled, and labelled accordingly. Every percentage point after break-even compounds into capacity, faster response times and clearer patient flow.
[ 09 · NEXT ]
Your front door has numbers like these. Nobody has counted them yet.
A Workflow Audit does for your operation what Phase One did here: measures the demand, labels the evidence, and shows what it would take to capture what you’re losing — before you commit to anything.
Book a Workflow Audit