Case studies

The gaps nobody finds until somebody asks

What actually turns up when every call gets reviewed instead of a sample — and what it takes to close the gap. Evidence-led write-ups from the team behind The Heartbeat.

All case studies

  1. One week of calls drawn as a waveform. A narrow slice is selected and burning red — the five percent a manual QA team reviews. The remaining 95% is drawn but never heard.Coverage · Indian contact centresThe 5% Illusion: Why Your QA Numbers Describe a Week You Never HeardA busy floor records eight thousand calls a week and listens to forty. The problem is not that quality is unknown — it is that quality is confidently misreported, from a sliver of the week, with silence about the rest.The Heartbeat Team · 26 Aug 20269 min read
  2. A register of calls. Most records are hollow because nobody reviewed them; a reviewed one opens into a verbatim quote, a timestamp and a link back to the audio.Audit readiness · Collections & lendingWhat You Cannot Prove: The Evidence Gap Nobody Finds Until an Auditor AsksFor the calls nobody reviewed there is no finding, no note, no reviewer, and no record that a review ever happened. That is not a QA problem — it is an audit-readiness problem, and it surfaces at the worst possible moment.The Heartbeat Team · 26 Aug 20268 min read
  3. One code-switched sentence parsed twice: an English-first tool drops the Hindi tokens, while native judgement keeps every one and flags the risky phrase.Language · Hindi, English & HinglishThe Language Tax: Why English-First Tooling Quietly Fails on HinglishIt does not fail loudly. It returns a plausible-looking score built from the part of the call it happened to understand, and nothing on the dashboard tells you that half the conversation was never scored at all.The Heartbeat Team · 26 Aug 20268 min read
  4. A week of automated flags entering a gate. Most are struck through and fall away; the few a human confirmed pass through and burn red.Signal quality · Contact centre QAFour Hundred Flags: Why Automated Call QA Becomes Shelfware in Three WeeksThe flags are technically correct and practically useless. The supervisor cannot check four hundred of them, so they check none, and the tool becomes shelfware that still appears on the licence renewal.The Heartbeat Team · 26 Aug 20268 min read
  5. A column of scored calls beside a set held back behind a gate — excluded rather than scored zero, each carrying the reason it was not judged.Scoring fairness · Agent performanceWhat We Refuse to Score: The Fairness Gap That Ends QA ProgrammesThe customer was furious, so the agent's score dropped — even though the agent handled it perfectly. The call was never answered, so the agent scored zero on every skill. Each produces a defensible-looking number that is wrong.The Heartbeat Team · 26 Aug 20269 min read
  6. Two channel lanes with the opening forty-five seconds bracketed, weighted cues scoring one lane as the agent, and a recorded swap decision.Speaker attribution · TechnicalWhich Voice Is the Agent: The Silent Failure Under Every ScorecardThe standard heuristic is that the agent is whoever talked most. It is right maybe 80% of the time, and the 20% it gets wrong is exactly the population you care about — the calls where the agent went quiet, or the customer dominated.The Heartbeat Team · 26 Aug 20267 min read
  7. The same call scored twice by hand, drifting apart, beside a long run of identical machine-applied rubric marks holding a straight line.Scoring consistency · Agent performanceTwo Supervisors, Two Scores: The Subjectivity Nobody MeasuresThis is not incompetence. It is what happens when a rubric is applied by a human under time pressure at the end of a shift — and it quietly destroys the two things scores exist for: ranking and trend.The Heartbeat Team · 26 Aug 20268 min read
  8. A flagged line of agent speech with the exact words isolated, and a replacement line offered directly beneath it in the same language.Coaching · Agent developmentA Score Is Not an Instruction: The Coaching Gap in Every QA ProgrammeEmpathy: 2 out of 5 is not coachable. Neither is opening score 67%. The translation from score to behaviour happens in a one-to-one, from memory, at the end of a shift — which means it happens late, inconsistently, and only for the worst scores.The Heartbeat Team · 26 Aug 20267 min read
  9. A dimmed dashboard grid beside a five-page report, its final page lit — a list of names, behaviours and due dates.Delivery model · ReportingThe Dashboard Nobody Opens: Why QA Tooling Has a Delivery ProblemBought by a QA head, used intensively for six weeks, then opened once a month before the review meeting. And a dashboard ends in numbers — numbers do not assign work.The Heartbeat Team · 26 Aug 20268 min read
  10. Six storage connectors converging into a single ingestion path, with call metadata read straight out of the filenames the dialler already writes.Onboarding · Contact centre ITTwenty Minutes, Not Two Quarters: The Integration Gap That Kills QA DealsMeanwhile the compliance risk the tool was bought to address accrues daily. Vendors integrate where the value is highest for them — at the dialler, in real time — which is also the hardest place to reach.The Heartbeat Team · 26 Aug 20267 min read
  11. A stereo call drawn as two clean separate lanes beside a mono call split into two copies, each with the other speaker's intervals muted.Recording quality · TelephonyMono, Stereo and the Honest Answer: A Trade-off the Category HidesIn mono, who said that becomes a machine-learning inference rather than a fact. Talk ratio, cross-talk, dead air and per-speaker sentiment all degrade at once — and the industry norm is to say nothing about it.The Heartbeat Team · 26 Aug 20266 min read
  12. Two overlapping keyword sets: defaults plus a workspace's additions, with an explicit allow list subtracted and the subtraction logged.Configuration · Multi-campaign floorsOne Rubric Does Not Fit Every Campaign: The Configuration GapThe collections team is penalised for not closing a sale; the sales team for not verifying identity. Most platforms handle this with a single global configuration, or with per-tenant settings that need a support ticket to change.The Heartbeat Team · 26 Aug 20268 min read
  13. A region boundary holding audio, transcripts and derived scores together, with retention clocks running at fifteen days and twelve months.Privacy & residency · DPDP Act, 2023Where Your Recordings Live: Residency as Architecture, Not PolicyMost conversation-intelligence vendors process in US or EU regions, are vague about where derived data lives, and answer residency questions with a policy document. It is easier to write a policy than to build one.The Heartbeat Team · 26 Aug 20268 min read
  14. An append-only ledger of minutes with a balance gate ahead of submission and refund rows written back for jobs that failed.Commercial model · BPO & outsourcersPaying for QA You Did Not Receive: The Commercial GapBPO call volume swings with campaigns, seasons and client wins. A per-seat annual licence prices the peak and is dead weight in the trough — which is why outsourcers, the natural buyer, resist the standard model.The Heartbeat Team · 26 Aug 20267 min read
  15. A floor-wide grid of agent rows with only the viewer's own row lit, the rest not dimmed but absent entirely.Access control · Agent experienceAgents Are the Last to Know: The Visibility Gap in Your Own QA ProgrammeThey learn their score in a monthly one-to-one, weeks after the calls that produced it, filtered through a team leader's summary. So most platforms resolve the problem by giving agents nothing at all.The Heartbeat Team · 26 Aug 20266 min read
  16. Four stages — connect, process, verify, act — with the verify stage marked as staffed by people rather than automated.Operating model · Service deliverySoftware That Needs a Team You Do Not Have: The Delivery-Model GapThe customer is expected to supply the missing half: someone to configure the rubric, tune the flags, triage the alerts, build the reports, and translate the output into coaching. Most floors have nobody for that.The Heartbeat Team · 26 Aug 20267 min read
  17. A shelf of mostly blank documents with a few filled, and the filled ones marked to separate what was observed from what was inferred.Engineering practice · DocumentationNobody Could Explain the System, Including UsRoughly eighty-nine endpoints the product depended on were undocumented. Design decisions survived only in uncommitted working-tree files. A new column was invented beside an old, empty, already-plumbed one — apparently without anyone knowing the old one was there.The Heartbeat Team · 26 Aug 20267 min read

Curious what is on the calls you never hear?

Book a demo and we will walk you through the platform — how every call gets reviewed, not a sample, and what comes out the other side.

Book a Demo