A decibel is the smallest change in sound a person can reliably notice. That's the whole thesis: the difference between a conversation that closes and one that doesn't is usually small, specific, and completely invisible unless someone is measuring every call.
Our founders met at a regional insurance carrier, on opposite sides of the same problem. One ran a quality program scoring six calls per agent per month on a form built in Excel. The other was building the speech infrastructure that could, in principle, have scored all of them.
The gap between those two facts was the company. The technology to analyze every conversation had existed for years. What didn't exist was a product that a regulated enterprise could actually deploy — because every vendor in the category required shipping customer recordings to someone else's cloud, and the security review died there every single time.
So we built the opposite. Overtone runs where your data already lives. The system of record stays inside your walls, audio never persists off your network, and the analysis is applied to every call rather than a sample chosen mostly by convenience.
We've been deliberately quiet while we proved that out with early customers in insurance, banking, and healthcare. This page is the first thing we've said publicly.
No launch party, no category invention, no manifesto. Just a long stretch of getting the hard part right before talking about it.
Two colleagues at a regional carrier conclude that sampling-based quality management is a technology artifact, not a methodology, and start prototyping on nights and weekends.
Founded with a seed round from investors and angels who had run contact center operations at scale, and knew exactly which promises to be skeptical of.
An auto and home insurance carrier goes live analyzing quote calls. Coverage moves from 1.8% of conversations to 100% in a single quarter, and the first surprise is a top performer skipping a required disclosure.
After the third enterprise evaluation stalls on data residency, we rebuild the system of record to run entirely inside the customer's network with stateless inference. It becomes the reason most customers choose us.
Nine deployments across insurance, banking, healthcare, and telecom. More than forty million conversations analyzed, and the team roughly doubling.
Every score the product produces cites the utterance it came from. We hold ourselves to the same standard internally — no roadmap decision gets made on a story about one customer call, and no benchmark goes on this website that we can't reproduce.
It would be faster and cheaper to build SaaS-only and tell regulated buyers to get comfortable. We decided early that we'd take the harder architecture instead, and we've turned down deals that would have required loosening it.
Analytics pointed at people who never see the output becomes surveillance. We design for the agent first — visible scores, visible rubric, visible evidence, and a dispute path that works. If a feature only makes sense when agents can't see it, we don't ship it.
The interesting part of this product is the analysis. The deployment, the upgrade path, the failure modes, and the audit trail should all be aggressively boring. Nobody has ever thanked a vendor for a surprising database migration.
Seattle has an unusual concentration of two things this product needs: people who have built production speech systems at scale, and people who have run large regulated contact centers for insurers, banks, and health systems. We're in Pioneer Square, and about a third of the company is remote across North America and Europe.
AlphaDecibel, Inc.Small team, real customers, unusually hard deployment constraints. If enterprise on-premises software sounds like a punishment rather than an interesting problem, this probably isn't the right fit — and that's fine.
Seattle or remote (US/EU) · Own diarization and ASR adaptation quality across 31 languages and some genuinely terrible telephony codecs.
Seattle or remote (US) · Build the on-premises processing pipeline and query API. SQL Server depth valued; Kubernetes experience explicitly not required.
Seattle preferred · Design for three audiences with opposite needs — a frontline agent, a supervisor mid-shift, and a CFO in a board deck.
Remote (US) · Live in customer security reviews and on-premises deployments. Comfort with Windows Server, AD, and change advisory boards is the job.
Remote (US) · Sell into carriers and large agencies. You've run a regulated enterprise sales cycle and you don't flinch at a twelve-week security review.
Nothing quite right? Write to careers@alphadecibel.com and tell us what you'd want to work on.
Whether you're evaluating, curious, or want to work here — there's a real person on the other end of every one of these.