Contact center quality assurance that scores every call, chat and email — and follows every promise until it is kept
Every call scored, and every promise an agent makes followed until it is kept.
See one case, screen by screen ↓Today a QA analyst pulls 4–5 random calls per agent per month from the recorder, listens at 1.25× and fills a spreadsheet scorecard: greeting, ID check, empathy, accuracy, close. The duties the regulator cares about — disconnection rights, payment-plan terms, the recording notice, customers in vulnerable circumstances — are a yes/no column, checked only on the calls that happened to be sampled. Chats and emails are not reviewed at all, and analysts disagree by 10–15 points on the same call at the monthly calibration.
So the rare, high-risk call is found by luck. A customer says her husband is on an oxygen concentrator; the agent places a disconnection hold but never records it on the priority services register, and the next disconnection run still includes the account. And the most common reason customers call back — “you said you'd call me and nobody did” — is never seen, because nobody tracks what an agent promises. The repeat call is the first signal, often as a complaint.
Estimated split of one QA analyst's five-day week, by hand and with the solution. With every interaction scored, the analyst's time goes to disputes, auto-fails, calibration and the calls that need a person.
Six agents transcribe, pin, score and check every interaction and follow every promise; QA, compliance and team leads decide.
Lumina Utilities supplies electricity and water. On Mon 12 Oct 2026 at 10:14, Helen Marsh rings about a final disconnection notice for $428.40. Tomas Reyes, Team 3, billing and payments, takes the 6:18 call. Here is what happens next, screen by screen, in the working solution.
Nadia owns quality for Teams 3 and 4. Her list holds every call, chat and email from Wed 7 to Tue 13 Oct scored against rubric v7: 41,860 interactions — 28,410 calls, 7,920 chats, 5,530 emails — averaging 86.4, sorted lowest score first. 148 carry a compliance alert and 61 a missed promise. Helen Marsh's call sits at 82, with a medical-dependency alert and a promise still open.
“Every call, chat and email scored against rubric v7 · 41,860 interactions, average 86.4”
Six agents scored the call a minute after it ended, at 10:21. The transcriber produced 24 turns from two speakers and marked 3 words it was unsure of. The moment finder pinned 10 rubric moments, from the greeting at 0:04 to the close at 5:51, and found a 42-second hold at 4:10. Each pin is coloured by the score it earned: 82 out of 100, one promise open.
“82 · one promise open · Rubric v7 · 10 items · team average 85.2”
Nadia clicks the Plan terms pin. The playhead jumps to 2:55, the moment plays and the transcript line lights up. Tomas offered six monthly payments of $71.40 from the first of November and said what happens if one is missed. The scorer checks the sum — 6 × $71.40 = $428.40, the full arrears — and gives 15 of 15 under Customer Service Code §3.3.
“Amount, dates and the missed-payment consequence all said.”
“My husband's on an oxygen concentrator at night.” Tomas placed the disconnection hold at 10:17, worth 6 points. But the customer was not recorded on the priority services register and the medical certificate form was not offered, so 9 points come off and the item scores 6 of 15. Code §4.2 asks for the register to be updated the same day, so the checker raises alert CA-2291.
“Not on the priority services register · alert CA-2291 open”
Farah's alert list shows every unmet duty under the Customer Service Code, with the quote and the time it was said: 148 this week, 97 confirmed, 18 dismissed, 33 open. 23 are medical dependencies not recorded, including a chat in which Kofi Boateng's mother “can't be without power”. Farah confirms CA-2291 and records Helen on the register. She dismisses others with a reason — §3.1 applies to homes, not to Northgate Bakery Ltd.
Compliance checker: never closes an alert — the compliance lead confirms or dismisses.
Tomas promised to call Helen back once billing had checked the meter reading. The promise tracker records what, who and by when, and turns “tomorrow” into the next working day at 17:00: Tue 13 Oct, 17:00. Until then it watches the outbound call log for any call to account 3004 7718 25, case notes by Tomas or the billing team, and emails and letters sent. Meter-reading request MR-55107 is still with billing.
Rubric item 7: “held open after the call and scored again when the promise falls due: kept with a matching record 10, kept late 6, missed 3.”
At 17:00 the proof search finds no call to Helen's numbers since Mon 10:20, no case notes since the call, and no email since the plan confirmation at 10:21. The promise is missed. Item 7 is rescored from 10 to 3, the call from 82 to 75, and Owen Achterberg, the Team 3 lead, is alerted.
“17:00 · no callback to Helen Marsh was logged — promise missed. Owen Achterberg alerted”
Nadia books the callback from the pin: Tomas rings Helen at 17:20 on his evening shift, and billing is asked for the meter-reading result first. Helen's journey shows why it matters: at 10:31 she rated the app “Neutral” — “Plan is fine but I still don't know about the meter reading.”
“Customers with a missed callback rang again within 2 days 64 % of the time (Lumina, last 90 days).”
Each promise is a bar from the moment it was said to the moment it is due. This week: 1,284 promises made, 1,127 kept with a matching record, 96 open and watched, 61 missed — 5.1 % of those due — and 58 callbacks booked after a miss. Proof came from the outbound call log 598 times, letters and emails 287, field work orders 141 and account credits 101.
Promise tracker: never marks a promise kept without a matching record, and never contacts the customer.
The heatmap shows Team 3 earning 62 % of the promise points: 9 of 23 callbacks promised this week were never booked in the system. Tomas averages 84.1 on 63 interactions. Two themes: book the callback before the call ends — 3 of 9 booked, 2 missed — and record a medical dependency the same day, 3 of 5. His strength: payment-plan explanations, 11 of 12 in full, which Lucia Brandt has asked to use in training.
Coaching writer: never ranks agents for discipline — the team lead decides coaching.
The same call, seen by the five people who own a part of it — what their week looked like, and what it looks like now.
Six agents transcribe, pin, score and check every interaction and follow every promise; QA, compliance and team leads decide.
Turns every recording into a two-speaker transcript with word timings, separates agent and customer, finds holds and silences over 5 seconds, and marks any word below 80 % confidence.
Finds the rubric moments in calls, chats and emails — greeting, identity check, reason, empathy, vulnerability, plan terms, promises, holds, accuracy, close — and pins each to a time or a line. Where an item did not happen, it pins where it should have.
Scores each rubric item using only the pinned words and the account record, checks facts against billing and meter records, and writes the reason for every point taken off.
Lists the Customer Service Code duties that applied to each interaction — disconnection rights, plan terms, medical dependency, recording notice, debt advice — and whether each was met, with the quote and the time. Raises an alert for each unmet duty; a medical dependency not recorded is sent the same day.
Extracts each commitment — what, who, by when — and turns “tomorrow”, “later today” and “within N days” into a due time. Searches the outbound call log, case notes, letters and emails sent, work orders and account credits; at the due time with no record, marks it missed, rescores the promise item and alerts the team lead.
Groups each agent's lowest-scoring items across every interaction in the period, writes up to three themes and one strength in plain words, and picks the clips for each.
Settle scores, confirm alerts. The agents propose; a named person decides.
Ask about a call, a promise, an alert or an agent's coaching — every answer cites its source.
Why did Helen Marsh's call drop from 82 to 75?
At 3:42 Tomas promised to call Helen back tomorrow once billing had checked the meter reading. Rubric item 7 holds a promise open and scores it again when it falls due. At 17:00 today the call log, case notes and emails sent showed no callback, so the item fell from 10 to 3. You can book the callback from the pin.
Which promises are still open and due today?
Of the 96 promises being watched, two are due before the end of today: Marisol Vega's plan agreement (Kwame Asante, by 18:00) and George Amankwah's metering callback (Leah Morgan, by 17:30).
Show this week's calls where a medical dependency was not recorded
23 interactions this week. The Customer Service Code asks for the priority services register to be updated the same day. Two are still open, including Helen Marsh, whose husband uses an oxygen concentrator at night.
What should Tomas Reyes be coached on?
Two themes from all 63 of his interactions this week: book the callback in the system before the call ends (3 of 9 were booked, 2 missed), and record a medical dependency the same day (3 of 5). His payment-plan explanations are a strength worth sharing.
13 screens from the working solution, on its sample data. Pick one to see it large.
What a promise means, where proof is looked for, who confirms alerts and how much is double-scored are settings, approved by the QA manager.
| Setting | Default | Choose from |
|---|---|---|
| When the agent says “tomorrow” | Next working day 17:00 | Next working day 17:00 · Next working day 12:00 · Next day 20:00 |
| Grace after due, before a promise turns red | None | None · 15 minutes · 1 hour |
| Where to look for proof | All six on | Outbound call log · case notes · letters and emails sent · field work orders · account credits and refunds · text messages sent — each On or Off |
| Promise missed — who is alerted | Team lead | Team lead · Agent and team lead |
| Compliance alert — to confirm within 4 working hours | Farah Siddiqui | Farah Siddiqui · Compliance team |
| Medical dependency — same day | Priority register team | Priority register team · Farah Siddiqui |
| Random calibration sample, double-scored each month | 5 % | 5 % · 2 % · 10 % — on top of every auto-fail and dispute |
| Channels and languages scored | Calls, chats and emails · English and Spanish | Calls, chats and emails · Calls only; English and Spanish · English |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample week of Lumina Utilities interactions · “target” = the design goal, measured in the live solution · “estimated” = our estimate: $121K a year = 1,284 promises a week × 9 points fewer missed (1 in 7, measured in Lumina's first fortnight of tracking, to 1 in 20) × 2.1 repeat calls each × $9.54 a call × 52 weeks, before complaint-handling cost; the split of an analyst's week is also our estimate · “typical” = published figures: manual QA covers 1–2 % of interactions (vendor-reported: EdgeTier, Klink); $9.54 per utility call (E Source, utility contact-center costs). Lumina Utilities, the Utilities Regulation Commission, its Customer Service Code and every person, account and call named on this page are fictional — characters and sample data in the working solution.
Software that checks customer conversations against a QA scorecard and the rules a contact center answers to. Contact Center Quality Assurance scores every call, chat and email — not a 1–2 % sample — quotes the words behind every score, raises compliance alerts the same day and follows every promise an agent makes until it is kept.
The call transcriber turns each recording into a two-speaker transcript, the moment finder pins each rubric moment to a time or a line, and the rubric scorer scores each item from the pinned words and the account record, with the reason for every point taken off. Auto-fail items always go to a person.
Yes. The compliance checker checks the duties in your customer service code on every interaction — disconnection rights, payment-plan terms, medical dependency, the recording notice and debt advice — and raises an alert with the quote and the time for each unmet duty. In the sample week it raised 148, 23 of them medical dependencies not recorded.
The promise tracker extracts each promise — what, who and by when — and sets a due time from your promise windows. It looks for proof in the outbound call log, case notes, letters and emails sent, work orders and account credits. With no record by the due time the promise turns red, the score is revised, the team lead is alerted and a callback can be booked from the pin.
Yes, on the same rubric. Chats and emails carry the rubric pins in the margin of the text. In the sample email, the reply answered the leak-allowance question and kept its promise to send the form, but left the customer's second question — whether to pay $612.18 now — unanswered.
In the working solution, item-level agreement with two analysts was 93.8 % on 400 interactions scored by both, with totals within 5 points on 91 %. A random sample — 5 % by default — is double-scored each month, and a rubric change needs a calibration run and the QA manager's approval.
Yes. QA analysts settle every dispute and auto-fail, the compliance lead confirms or dismisses every alert, the team lead decides coaching and callbacks, and the QA manager approves rubric and settings changes. The agents never contact customers and never rank staff for discipline.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your QA scorecard, customer service code, promise rules and a sample of recorded calls, chats and emails — and it goes live once every quality gate has passed. We will walk you through it on your own interactions first.
We'll run Contact Center Quality Assurance on a sample of your own recorded calls, chats and emails.