> but for our FinOps lens, accuracy directly impacts ROI
Exactly. So run the real numbers on correction time, not the marketing accuracy % they quote. We clocked our team.
On your points:
* Vocabulary: Both terrible out of the box. Sembly's custom list "sticks" better, but that's a low bar.
* Diarization: This is the hidden cost sink. Otter falls apart with >2 speakers or crosstalk. Sembly gets the structure right even if words are wrong, which is way cheaper to fix.
* Audio Quality: Sembly handles our terrible cell recordings better. Otter's accuracy tanks.
* Accents: Our team in Singapore gave up on Otter after a week. Sembly wasn't perfect but usable.
Don't pay for a slightly lower subscription if it means paying your sales rep to rewrite transcripts.
show the math
>Our FinOps lens
Good, because the accuracy percentages on their marketing pages are basically fiction. They're measured on pristine, academic datasets. Your sales calls aren't that.
You've nailed the right criteria. On diarization and poor audio, Sembly's error is just words, which is cheap to fix. Otter's error is architectural, which means you're paying a sales salary to rebuild the transcript from scraps. The subscription price difference is noise compared to that labor cost.
For vocabulary, they're equally bad out of the box. But Otter's custom vocabulary feels like a suggestion box that gets ignored. Sembly's actually works, which is a depressing thing to praise.
Your stack is too complicated.
Exactly. So test them on your actual worst call, not their marketing demo audio. The quoted accuracy percentages are meaningless.
You're right to focus on those four points. For global accents, Otter was unusable for our APAC team. Sembly managed UK and Australian fine, but struggled with thicker Singaporean and Indian accents. Still, it was correctable. Otter's output was just garbled.
Forget the price difference. Calculate the hourly cost of your team fixing speaker errors in Otter versus just correcting words in Sembly. That's your real TCO.
Trust but verify.
You're missing the actual test. Their "accuracy" claims are useless.
Upload the same three real calls from your team: a clean Zoom recording, a bad cell connection with crosstalk, and one with strong accents. Time how long it takes a sales rep to get each transcript CRM-ready.
Sembly will win on the diarization and audio quality, which is where the real hours are lost. Otter's custom vocabulary feature is basically broken, so even if word accuracy is similar, maintaining it is a hidden labor cost.
The price difference between plans is irrelevant. The labor difference is the ROI.