About Valence AI
We built the tool we needed on the call floor
Valence AI started with one question: why can a skilled call-handler hear that a customer is about to explode, but their tools cannot?
Where it started
Chloe Duckworth spent a summer embedded in a large contact center operation, working alongside supervisors and floor managers. She watched skilled agents recognize caller frustration in the first 30 seconds of a call and adjust their approach instinctively: slowing down, shifting tone, acknowledging the delay before the caller had a chance to name it.
She also watched other agents miss those same cues entirely. The calls spiraled. Supervisors intervened too late. The difference between the two groups was not training level, not script adherence, not tenure. It was attentional bandwidth. The skilled agents had developed an almost automatic emotional read on the caller. Everyone else was managing the conversation consciously and had nothing left for the subtext.
The question that followed: could that perception be made available to every agent, automatically, from the first spoken word? Valence is that answer.
Founded 2023. Headquartered in Palo Alto. Independently funded.
The team
Built by people who have worked in contact center operations, speech AI research, and enterprise SaaS, and who have seen firsthand what a missing signal costs.
Chloe Duckworth
CEO and Co-Founder
Previously led operations technology at a large contact center outsourcing firm. Spent years on contact center operations floors studying how frontline teams process emotional cues before co-founding Valence. Focused on making the product legible to operations leaders, not just developers.
Marcus Webb
CTO and Co-Founder
Background in affective computing and prosodic feature extraction, with years of experience building real-time audio inference pipelines for broadcast and telephony systems. Joined Chloe to solve the latency problem that makes voice emotion useful in production. Responsible for the core detection architecture and API design.
Priya Srinivasan
Head of Customer Success
Background in enterprise SaaS implementation and contact center workflow consulting. Ran the early-access program and designed the integration playbooks that got pilot customers to first signal in under four hours. Now leads all customer onboarding and expansion.
How we operate
Principle 01
Real signal over proxy metrics
CSAT surveys, NPS, and post-call ratings are useful but delayed. We measure what actually happened in the call: the acoustic signal in the moment, not a retrospective score that averages away the detail. When we report accuracy, we report it per emotion class, not as a blended aggregate.
Principle 02
Built for the agent, not just the supervisor
Most contact center AI is designed for the people who read reports about calls: supervisors, QA analysts, managers. The person on the live call gets nothing. Valence is designed so the agent on the call has access to the same signal the supervisor sees, in a form they can actually use without breaking their flow.
Principle 03
Honesty about accuracy limits
We publish our F1 scores per emotion class. We describe the conditions under which accuracy degrades (low audio quality, overlapping speech, non-native accents) rather than quoting a single headline number. We are improving the harder cases, and we prefer that our customers know which those are.
Come work with us
We are building a small, focused team. If the problem we are solving resonates with you, we want to hear from you: developer, operations professional, or someone who has spent time on a contact center floor.