3,061
Measured1,216 inbound (1,165 deduplicated), 1,730 guarding-desk outbound, 115 officer briefings. Test, internal, silent and never-connected calls excluded.
Evidence
The numbers.
And what they count.
Measured from a live security control room, 1 September – 14 October 2026. Every call and case checked against its trace, failures shown, and the human comparison figures are the operator’s own.
A work item is a case. A call is a channel event. The figures below keep their original labels and denominators.
volume
3,054
MeasuredGuarding-desk cases with a recorded outcome.
speed
10–30 s→2 s
Measured vs operatorPlatform pickup P95 1.0 s across 1,216 inbound calls; stated as 2 s because a client review saw 1–3 s. Before: the operator's figure for its previous control room.
5–15 min→67 s
Measured vs operatorMedian over 108 dispatches on the current agent. An earlier dashboard shows 68 s over 121 including the legacy agent; the full-desk review shows median 71 s, P95 140 s. Before: the operator's previous control room, 5 min typically and up to 15.
5–12 min
operatorOperator's step-by-step benchmark for the same job done by hand.
4–13×
derivedThe ratio of the two figures above. Both inputs must be visible on the same page for this to be shown at all.
Derived from the operator’s 5–15 min and the measured 67 s ring-to-dispatch median above.
accuracy
34 s
MeasuredMedian from ring to the first exact site match.
2%→0.8%
Measured vs operatorShare of location checks that failed, latest two weeks. Before: 2%, and the operator's own figure was 98% verified.
96%
MeasuredThe dispatch reference read back to the caller: 104 of 108.
13 s
MeasuredOne client call, 23 September 2026: the caller gave the wrong postcode, caught at 13 s, dispatched at 77 s, 4 actions at dispatch.
follow-up
33%→98%
Measured vs operatorAptic 98.3–98.4% (98.3% with the test site and placeholders removed). Before: the operator's own 33%.
10%→100%
Measured vs operatorBefore: the operator's own 10%.
2%→0%
Measured vs operatorAptic 0.04%. Before: the operator's own 2%.
7–21 a week→0
Measured vs operatorBefore: misses on the staffed desk, operator-reported, 7–21 a week.
2×
ModelledAbout 2,086 completed against about 1,030 at the operator's old rates (2.02×). Modelled on those rates.
oversight
The two 94% figures below are different numbers over different denominators. One is all workforce cases; the other is book-on cases only.
94%
Measured94.1% of 3,054 workforce cases closed by Aptic alone.
4.4%→1.9%
MeasuredWeekly 3.9, 5.0, 4.8, 1.9, 1.9% for the weeks of 7 Sep – 5 Oct. Early (7–20 Sep) against late (28 Sep – 11 Oct) is significant. The drop follows the 21 September routing fix.
94%
Measured39 of 665 book-on cases escalated: worker not answering 17, system or app 9, a manual shift-system step 6, cover 5.
audit
100%
MeasuredEvery call run through rule checks; flagged calls read by a person.
100%
Measured697 of 697 guarding calls tie to a case. On the alarm desk 452 of 454 have a call-log row, and the other two have a dispatch or a note on record.
31
Measured31 KPIs in the catalogue, reviewed weekly.
improvement
3.5%→1.4%
MeasuredCalls with a hard system error, early in the period against late.
5→1
MeasuredPhantom promises on the current agent after a full re-check: five early on, one since. A rule now blocks the agent from saying it before a response is dispatched.
capacity
Nothing in this group was measured. Every real call and follow-up was replayed at human handling times, inbound first, with breaks and admin excluded — which makes it the best case for a staffed desk, not the worst.
8
ModelledThe most jobs in hand at one moment if the same work were done by hand. Range 6–11. Replay of every real call and follow-up at human handling times.
~1 hour
Modelled4.4% and 1.2% of time. Bursts on 44 of 45 days, and 30 days hit 4+. The range of 2+ time is 2.9–7.4%.
13–14
ModelledA rota at the 80th percentile of hourly workload, 30% shrinkage, 42-hour weeks. Central case 4 seats and about 13 FTE; slow handling 5 seats and about 14.
33% / 7% / 1.5%
ModelledRanges 18.7–57 / 1.3–18.4 / 0.5–4.1% across fast to slow handling. Over five minutes with two people: 0.3% central, up to 4.6% slow. Shift changes 06–08 and 18–21.
18% / 2% / 0.3%
ModelledStarted more than five minutes later than Aptic made them.
cost
18–27% lower
EstimateDepends on desks, volumes and set-up; sized with the prospect during the pilot.
calls
25%
Measured158 of 637 after setting aside 60 passed to the alarm desk, judged from what they said. 122 of them in their first sentence.
49%
Measured522 alarm-desk calls by keyword classification: alarms 49%, keyholder and access 15%, intercom 9%, officers 5%, office and patrol 5%, brief or voicemail 16%. Approximate.
Real calls by UK hour Measured