Every score, coaching note, and revenue number Overtone produces traces back to the same diarized, timestamped transcript. That's what makes the output arguable in a good way — you can always play the second of audio it came from.
Post-call or live-stream, pulled from your recorder or platform by API. Audio is never copied into a store you don't own.
Contact-center-tuned ASR with speaker diarization, timestamps, and PII redaction before anything is persisted.
Sentiment, phases, entities, intents, compliance checks, and your rubric — applied identically to every call.
Dashboards by role, coaching plans with evidence, alerts on risk, and metrics pushed into WFM and BI.
A single "negative" label for a nine-minute call tells you nothing you can coach. Overtone tracks sentiment per speaker, per turn, so the useful question becomes answerable: where did this go wrong, and did the agent bring it back?
Most conversation intelligence ships with a fixed scorecard and asks you to adopt it. That's backwards. Your quality program encodes years of regulatory exposure, brand voice, and hard-won operational knowledge. Overtone runs your rubric.
Because the rubric is yours, disagreements become productive. When an agent disputes a score, you open the dimension definition and the cited utterance side by side. Most customers see disputes drop by half in the first quarter.
A score says an agent lost twenty points on objection handling. That is not coaching. Coaching is: here are the three calls, here is the timestamp, here is what the top quartile said instead, and here is what it was worth.
You delivered the regulatory disclosure after quoting price on 46% of eligible calls. On calls where disclosure came first, your close rate was 17 points higher — customers who hear terms up front raise fewer objections at signature.
When a customer names a competitor you acknowledge and move on. J. Whitfield asks one clarifying question first and retains 23% more of those calls.
Verification greeting compliance, measured on the 60 calls following the session.
The engine that decides whether anyone renews the contract. Overtone joins every scored conversation to what actually happened in your CRM, then answers the only question that matters at budget time.
Attribution down to the individual dimension and, where the signal supports it, the individual phrase pattern. Segmented by product, channel, tenure, and customer type — because what works on a renewal call does not work on a first-contact quote.
Model the revenue effect of moving any dimension by n points for any cohort, before you spend a quarter of coaching capacity on it. Prioritize the lever with the largest modeled return rather than the largest score gap.
Cohort comparisons with control groups, confidence intervals on the lift estimate, and an exportable methodology appendix. Built to survive the meeting where someone asks whether correlation is doing all the work.
Every transcript is indexed for full-text and semantic search, and sits behind a natural-language query interface. Analysts stop filing tickets for one-off reports and just ask.
Across 68,412 sales conversations in Q2, three objection types account for 71% of lost-close volume:
Agents who ask a clarifying question before responding to a price objection retain 23% more of those conversations.
| Capability | Detail |
|---|---|
| Audio formats | WAV, MP3, OPUS, μ-law/A-law, G.711, G.729, AMR-WB · mono, stereo, and dual-channel |
| Transcription | 97.2% average accuracy on contact-center audio · 31 languages · custom vocabulary and acronym adaptation |
| Diarization | 99.1% on dual-channel, 94.3% on mixed mono · speaker-role attribution, not just speaker turns |
| Latency | Post-call scoring in under 4 minutes at p95 · live-stream mode surfaces sentiment within 8 seconds |
| Throughput | Benchmarked to 40,000 concurrent calls per deployment · horizontal scale-out on commodity CPU |
| PII handling | PAN, SSN, DOB, policy and account IDs, addresses · masked before persistence and before indexing |
| Retention | Governed entirely by your policy · configurable per data class, with hard-delete verification |
| Authentication | Entra ID, Okta, Ping, SAML 2.0, OIDC · Active Directory and AD FS for on-premises |
| Access control | Role-based down to queue, site, and dimension level · full read-audit trail |
| Export | REST API, webhooks, scheduled CSV/Parquet drops, Snowflake and Power BI connectors |
| Deployment | On-premises (Windows Server or Linux, no Kubernetes required), private cloud, or managed SaaS |
| Availability | 99.9% SLA on Enterprise and above · active-passive DR with documented RPO/RTO |
Send us a week of audio and your existing scorecard. We'll score every call and walk you through what your QA sample never surfaced.