About Sermo

We build agents that sound like the people they answer for.

Sermo started in a support queue. We were the ones answering at 2am, watching careful answers get flattened into templates — and hearing how different every company sounds when it is doing its best work.

A small team building voice agents that keep a company sounding like itself.

Conversations resolved

41M

Every one handled end to end, without a person stepping in.

Conversations resolved

41M

Every one handled end to end, without a person stepping in.

Support teams onboard

380+

From ten-person startups to enterprises running follow-the-sun.

Support teams onboard

380+

From ten-person startups to enterprises running follow-the-sun.

Languages live

Trained on each team's own transcripts, not a translation layer.

Rate

+7.1%

Speed

+12.8%

Languages live

Trained on each team's own transcripts, not a translation layer.

Rate

+7.1%

Speed

+12.8%

A day in the queue

Refund issued · 0:38
CSAT 5.0 · voice
Escalated to Priya
Address updated · 0:52
Callback booked · 0:26
Plan downgraded · 1:14
Invoice resent · 0:41
Delivery rescheduled · 1:03
Card retried · 0:33
Handed to billing · 0:47

Our story

Every company already has a voice. Most software talks over it.

Fifteen people across Rotterdam, Lisbon and Toronto.

In 2021 we were running support for a logistics company with three products and one very tired team. The scripted assistant we bought sounded nothing like us. Customers noticed before we did.

So we stopped writing scripts and started reading transcripts. Ten thousand of our own calls, tagged by hand, looking for the moments where a person said the thing a template never would.

That became the whole product. Sermo trains on the conversations a team has already had, so the agent inherits the phrasing, the caveats and the judgement that were there all along.

Four years on, the test hasn’t changed. If a customer can tell which replies came from us and which came from the team, we haven’t finished the work.

2021

Ten thousand transcripts

We tag our own support calls by hand for six months.

The pattern nobody had written down turns up in the margins.

The first prototype answers a refund question correctly.

2023

First agent goes live

A logistics team ships an agent to real customers.

Nobody flags it as a bot in the first week.

We stop calling it an experiment.

2025

Composer arrives

Teams start shaping agents without writing a script.

Rehearsals run against real transcripts before launch.

Policy changes ship in an afternoon, not a quarter.

Today

Fourteen languages

Voice and chat now share one transcript store.

Agents run in three regions with data staying put.

Same test as day one — you should not be able to tell.

How we work

Three things we don’t trade away.

Written down in year one, argued about ever since.

Listen before generating

No agent ships until it has read the transcripts. A model that hasn’t heard your team is just a confident stranger.

Ridgeline · support inbox

Reading

CALLER

It says delivered but nothing arrived.

AGENT

Let me pull the courier scan for you.

10,482 read

Listen before generating

No agent ships until it has read the transcripts. A model that hasn’t heard your team is just a confident stranger.

Ridgeline · support inbox

Reading

CALLER

It says delivered but nothing arrived.

AGENT

Let me pull the courier scan for you.

10,482 read

Voice is not a text feature

Pauses, corrections and interruptions carry meaning. We build for the call first and let the transcript follow.

Live call · barge-in

On air

1:04

Caller interrupted at 0:12

Voice is not a text feature

Pauses, corrections and interruptions carry meaning. We build for the call first and let the transcript follow.

Live call · barge-in

On air

1:04

Caller interrupted at 0:12

Your transcripts stay yours

Customer conversations train your agents and nothing else. No shared pools, no silent retention, no exceptions we quietly make later.

Ridgeline workspace

Encrypted

Card ending

refunded

Sent to

Never used for shared training

Your transcripts stay yours

Customer conversations train your agents and nothing else. No shared pools, no silent retention, no exceptions we quietly make later.

Ridgeline workspace

Encrypted

Card ending

refunded

Sent to

Never used for shared training

The team

The people your agents learn from.

Most of us answered tickets long before we built anything.

Noor Haddad

Co-founder, CEO

Ran support at two fast-growing marketplaces before starting Sermo. Still reads transcripts every week.

Priya Raman

Head of Voice

Built call-center automation for a decade. Believes a good voice agent should sound like it’s actually listening.

Emil Rask

Co-founder, CTO

Speech researcher turned engineer. Spent six years on latency so a spoken reply lands before the pause gets awkward.

Dario Sant

Speech Infrastructure

Keeps the voice pipeline fast under load. Obsessed with shaving milliseconds off every hop.

Marit Okonjo

Customer Engineering

Sits with support teams during rollout and doesn’t leave until the first week’s numbers look right.

Yusuf Adeyemi

Design

Designs the interfaces agents and humans both have to trust. Previously at two developer tools startups.

Get started

Ready to hear what your support could sound like?

We will build you a working agent on your own transcripts — and show you the numbers before you commit to anything.

Talk to us

Refund issued · 0:38

Escalated to Priya

CSAT 5.0 · voice