Hear every call.
Not 2% of them.
Caller transcribes and scores every call your contact center takes against your own quality criteria, and tells you which ones need a person. In 39 languages, processed in the EU.
Free during the beta, no card. Upload recordings, or let your phone system send them through the API.
In · raw recordings
MP3, WAV, OGG, M4A. Uploaded by hand or sent through the API.
Hi, I'm Kalle.
I listen to all of them.
Out · scored call
“Let me check the router for you.” The agent started troubleshooting without confirming who was calling.
See it work.
Press play. I'll wait.
The problem
Manual QA is a guess.
A reviewer samples a handful of recordings and hopes they speak for the rest. Coaching, compliance checks and client reports all get built on that sample.
Manual review
Fewer than 2 in 100 calls get heard.
With Caller
Every uploaded call is transcribed and scored.
How it works
From recording to coaching in five steps.
- 01
Upload
Drop in your recordings, or let your phone system send each one. Agent and date are read from the file name.
- 02
Transcribe
Agent and customer are split by voice, even on mono phone recordings.
- 03
Protect
Optional redaction masks emails, phone numbers, Spanish ID numbers, IBANs and card numbers before anything is stored.
- 04
Score
Your scorecard grades the call and quotes the transcript as evidence.
- 05
Coach
A review queue, channel alerts and dashboards show which calls and which agents need you first.
Call analysis
One call, fully taken apart.
A transcript by speaker, a summary, how the call ended and how the customer felt. A score for each criterion, with the line that justifies it and the moment to listen to.
Sample data
Summary
Connection issue fixed with a remote line reset. The agent skipped identity verification before troubleshooting.
- Topic Connection drops
- Outcome Resolved
- Customer negative, then positive
The agent never asked for a name or account details. Evidence is highlighted in the transcript.
Review queue
It tells you which calls need a person.
Scoring every call is half the job. Caller puts the ones worth your time in one queue and says why each one is there.
- Mandatory criterion missedIt earns less than half its points, however good the rest is.
- Critical scoreThe call scores below 60.
- Upset customerThe customer ends the call unhappy.
- UnresolvedNo solution and no agreed next step.
- Long silenceA quarter of the call or more is silence.
Sample data
- Mandatory criterion missed75Router reset, identity never checkedMarta R. · Broadband support · 06:12
- Critical score52Asks to cancel over the priceJon A. · Mobile support · 09:40
- Upset customer64Third outage this monthIker S. · Broadband support · 04:51
- Unresolved71Refund asked for, nothing agreedLaia P. · Billing questions · 07:03
A call leaves the queue when someone opens it and marks it as reviewed.
Alerts · Slack and Microsoft Teams
A missed mandatory step reaches your channel in minutes.
Mark what can never be skipped as mandatory: the legal disclosure, the identity check. When a call fails it, Caller posts in your team's channel with the criterion, the call, the agent, the score and a link to open it.
- Only when it matters. A call that meets its mandatory criteria sends nothing, however low it scores.
- No code. Paste the channel's incoming webhook address and save.
- Send a test notice before the first real one, and see how the last one went.
- For your own systems, every result also goes out as a signed webhook.
I only ping when it matters.
Sample data
Mandatory criterion not met: “Customer Verification”
Router reset, identity never checked · Marta R. · score 75
Open the call
The same notice, formatted for each. In English or Spanish, following your account.
Scorecards
Your criteria, written the way you'd say them.
Each criterion is a plain instruction with a point value, scored yes or no or on a scale you describe. Start from one of eight examples or write your own.
Mandatory criteria
Mark what can never be missing. A call that fails it goes to review and to your channel, whatever its total.
Try it before saving
Run one criterion against a real call of yours and read what the AI understood. Fix the wording, try again.
Data to note down
What you want from every call without scoring it: why the customer wants to leave, which product came up.
Knowledge
Graded against your playbook, not generic advice.
Upload scripts, policies and FAQs as PDF, Word or text. Add them to a criterion and Caller checks agents against what you actually told them to say, and names the document it relied on. A document that reads like a procedure becomes a draft scorecard.
- Retention scriptlinked
- Refund policylinked
- Mandatory legal disclaimerlinked
Languages
39 languages, including the ones other tools skip.
Let Caller detect the language, or set it per campaign when two are easy to confuse.
- Català
- Euskara
- Galego
- Español
- English
- Português
- Français
- Deutsch
- Italiano
- Nederlands
- Polski
- and 28 more
Kalle, the assistant inside Caller
Ask Kalle. It has heard every call.
Ask about your numbers, about what was said in your calls, or how anything works. Kalle answers in the language you write in, links what it found, and never changes anything in your account.
Go on.
Ask me.
Your figures
Which agent needs most support?
Marta R., with an average of 61 over the last 30 days, the lowest on the team. Most points are lost on Customer Verification: 22% of its points. First in Team.
What was said in the calls
Show me calls where the customer threatens to leave
Two calls where something like that was said: Asks to cancel over the price (Jon A., score 52) and Third outage this month (Iker S., score 64).
How something works
How do I send recordings through the API?
Create a key in Integrations and send each recording with a POST to api.caller.ee/v1/calls, the audio in the field file. The examples are in the docs: Send a recording
- Walks a new account through its setup
- Works from your last 7 and 30 days
- Reads the documentation and links the page
- Explains and points. You make the change.
Sample conversation
Analytics
Know who to coach, and on what.
Filter by campaign or agent and compare with the period before. The worst results come first. Export in one click.
I'd start with verification.
Weakest criteria
Average % of its points per criterion. Where the team loses most points.
Agents to coach
Average score per agent, lowest first.
- Marta R.61
- Jon A.68
- Laia P.74
- Iker S.82
- Noa V.89
What your customers talk about
Topics come from what is said in the calls. Nobody has to define them.
For the next session with Marta R.
Customer Verification is where the gap with the team is widest: 22% of its points against 48%.
To correct
Router reset, identity never checked
Good example
Billing address change, by Noa V.
Sample data
Search
Find the call where it was said. In any words.
Type what you are looking for and Caller also finds the calls where the same thing was said differently. Each result opens at that moment of the recording.
Where something like it was said
API and webhooks
Stop uploading by hand.
Your phone system sends each recording when the call ends. Agent, date and phone number are read from the file name, and every result goes back to your systems as a signed webhook.
curl https://api.caller.ee/v1/calls \ -H "Authorization: Bearer $CALLER_API_KEY" \ -F "file=@AG1143_2026-10-09_10-42-11.mp3" \ -F "external_id=pbx-000123"
For BPOs and multi-client teams
Built for many clients at once.
- Campaigns grouped under each client.
- A scorecard and a language for each campaign.
- Performance tracked per agent.
Client
Retail bankCampaign
Card disputesIts own scorecard · EspañolAgents
Marta, Jon, Laia…Data & security
Your calls stay in the EU.
Audio, transcripts and analysis are stored and processed only in the European Union, with a short list of European providers we name one by one. No US AI APIs in the data path.
I don't travel.
- 01Stored and processed only in the EU.
- 02Three providers, each named with what it receives and where.
- 03Nothing is used to train models, ours or anyone else's.
- 04Optional PII redaction before storage and analysis.
- 05Sentiment is read for the customer only, never for the agent.
- 06Strict isolation between accounts. Recordings have no public address.
- 07Delete a call or the whole account yourself, at once.
Who it's for
For the people who answer for quality.
COO
Run QA on evidence, not anecdotes.
One quality score across clients, campaigns and teams. Trends you can take to the board.
QA & Ops manager
Coach where it matters.
Weakest criteria and lowest-scoring agents, ranked. Every point backed by a quote.
BPO director
Every client on one platform.
Clients and campaigns kept apart, each campaign with its own scorecard and language.
Pricing
Free while we're in beta.
Try it with real calls before you spend a euro. Paid plans, priced per agent, are on the way.
Free
Available now€0 / month
- 45 minutes of audio a month
- Analysis included in the minutes
- Calls up to 15 minutes and 25 MB
- 3 agents, 1 user, 10 scorecards
- 250 MB storage for the life of the account
- All languages, alerts, the API and privacy features
Team
Coming soon€19per agent / month
- From 3 agents
- 10 hours of audio per agent a month
- 2 users, 3 scorecards
- Manual upload and the API
- Webhooks, Slack and Teams alerts, PII redaction
Quality
Coming soon€39per agent / month
- From 3 agents
- 22 hours of audio per agent a month
- 5 users
- Unlimited scorecards
- Everything in Team
Business
Coming soon€69per agent / month
- From 20 agents
- 39 hours of audio per agent a month
- Unlimited users and scorecards
- Clients, with a read-only user for each
- Audit log, corporate sign-in, data retention per client
Enterprise
Talk to us- Seats and audio agreed with you
- Everything in Business
- On an isolated instance
Need more volume during the beta? Write to caller@sumgrey.com. We're generous with early partners.
Questions we get first.
Where is my call data processed?
In the EU. The application runs in Germany, recordings are stored and analyzed in France, and transcription happens in the provider's EU region. The three providers are named in our Privacy Policy and Data Processing Agreement, and nothing is used to train models.
Is it really free right now?
Yes. The Free plan is the live plan during the beta: 45 minutes of audio a month with the analysis included, calls up to 15 minutes, 3 agents. Enough to judge it on real calls.
Do I need to connect my phone system?
No. To start, upload recordings by hand (MP3, WAV, OGG or M4A). When you want them to arrive on their own, your phone system sends each one through the API and the results come back by webhook.
How do scorecards work?
A scorecard is a list of criteria. Each one has an instruction in plain language, a way of scoring it and a point value. Caller evaluates every call against them and cites the line of the transcript it relied on.
Which languages are supported?
39, including Spanish, English, Portuguese, French, German and Italian, plus Catalan, Galician and Basque.
Can it warn my team when something goes wrong?
Yes. When a call fails a criterion you marked as mandatory, Caller posts a notice in a Slack or Microsoft Teams channel with the call, the agent and a link. You paste the channel's webhook address once. No code.
What can I ask Kalle?
Your figures for the last 7 and 30 days, calls where something was said, and how any part of Caller works, with a link to the page of the documentation it took the answer from. It explains and points; it does not change anything in your account.
How is sensitive customer data handled?
Turn on PII redaction and email addresses, phone numbers, Spanish ID numbers, IBANs and card numbers are masked in transcripts before they are stored or analyzed. It works by patterns: it does not detect names or postal addresses, and it does not change the recording.
Send us your first call.
Start free today, or talk to us. During the beta you speak directly with the people building the product.