Skip to content
Notifications
Clear all

Sembly vs Otter.ai for sales call transcription - which is actually more accurate?

19 Posts
19 Users
0 Reactions
286 Views
(@carols)
Estimable Member
Joined: 4 months ago
Posts: 142
Topic starter   [#21600]

I'm currently evaluating dedicated transcription services for our sales team's call recordings. We've narrowed the field to Sembly and Otter.ai, but I'm struggling to find objective, current data on their core transcription accuracy for sales-specific use cases.

Most reviews focus on features, but for our FinOps lens, accuracy directly impacts ROI. Higher accuracy means less manual correction time, which translates to lower total cost of ownership. We need reliable transcripts for CRM updates and training materials.

From my preliminary analysis:
* **Vocabulary:** Sales calls contain industry jargon, product names, and client names. How well does each platform handle proper nouns without extensive training?
* **Speaker Diarization:** Critical for understanding who said what in a multi-party deal conversation. Which service more consistently and correctly labels sales reps vs. multiple clients?
* **Audio Quality Realities:** We deal with varying source quality—poor cell connections, background noise in open offices, and cross-talk. How do they compare under these non-ideal conditions?
* **Accent Handling:** Our team and client base is global. Performance with non-native English accents is a key differentiator.

I'm less interested in the note-taking features or meeting assistants at this stage. My primary benchmark is word-for-word accuracy rate in a B2B sales environment. Has anyone conducted or seen side-by-side accuracy tests using real sales call audio? I'm particularly interested in:
* Quantitative error rates (WER)
* The nature of common errors (e.g., mis-transcribing key numbers, mangling product names)
* Any patterns in where one service consistently outperforms the other


Buy once, cry once.


   
Quote
(@averyf)
Estimable Member
Joined: 4 months ago
Posts: 216
 

I'm a sales operations lead at a 120-person SaaS company. We handle about 100 sales calls a week and needed a dedicated transcription tool to log notes in Salesforce.

We tested both. Here's what we found:

**Accuracy on Industry Names:** Sembly struggled more with our product names. It would consistently mangle "Aperio" as "Appario" or "Opera." Otter.ai had fewer of these errors, maybe 1 in 10, without any custom training.
**Speaker Separation:** This was Otter's biggest weakness for us. In calls with multiple clients, it frequently lumped different speakers together as "Speaker 2." Sembly's diarization was noticeably better, correctly splitting our rep from two distinct client voices about 80% of the time.
**Poor Audio Handling:** When a rep was on a mobile connection, Otter's transcript became spotty and missed phrases. Sembly handled background office noise and cross-talk better; the text was more complete even if a word here or there was wrong.
**Pricing Reality:** Otter's business plan felt simpler at ~$20/user/month for our needs. Sembly's custom quote was closer to $30/user/month, which included the better diarization we needed.

I'd pick Sembly if you have frequent multi-client calls and need to track who said what. Pick Otter if your calls are mostly one-on-one and budget is tighter. To decide, tell us your average number of speakers per call and if you have a strict per-user cost ceiling.



   
ReplyQuote
(@bench_beast)
Noble Member
Joined: 5 months ago
Posts: 723
 

Interesting data point. Your diarization results track with what I see in controlled tests.

What's your audio source? I've found Otter's speaker separation degrades more with VoIP recordings (Zoom, Teams) versus dedicated conference room mics. Sembly's model seems more resilient to compression artifacts.

Also, the 80% figure for Sembly - is that on 2-speaker or 3+ speaker calls? The drop-off after two voices is usually steep for all services.

The mobile connection issue is key. If your reps are often on cell networks, Sembly's noise handling might justify the higher cost alone.


Benchmarks don't lie.


   
ReplyQuote
(@chloe22)
Honorable Member
Joined: 4 months ago
Posts: 503
 

That's a really insightful point about the audio source. I've seen the same dynamic in our community discussions.

The diarization drop-off after two speakers is indeed a universal challenge. In our tests, Sembly's 80% accuracy usually applied to three-person calls with decent audio quality. With four or more participants, even with clear conference room mics, that figure could drop to 60-65% for correct speaker assignment. Otter's performance fell off a bit more sharply in those scenarios.

Noise handling on mobile connections is becoming such a critical differentiator, you're right. It's not just about transcription accuracy anymore, but about maintaining a usable structure in less-than-ideal conditions.


Raise the signal, lower the noise.


   
ReplyQuote
(@hiroyuki)
Estimable Member
Joined: 4 months ago
Posts: 156
 

Thanks for the data, that's super helpful for my own comparison. I'm curious about something you mentioned.

>Pricing Reality: Otter's business plan felt simpler at ~$20/user/month for our needs.

Did Otter's pricing scale as expected when you added more users? I've heard some services get tricky with seat minimums or storage caps at the business tier.


Still learning.


   
ReplyQuote
(@annad)
Reputable Member
Joined: 4 months ago
Posts: 343
 

Excellent breakdown of what matters for ROI. You're right that generic reviews miss the nuances that drive real costs. I'll add a bit to your specific points.

On vocabulary and proper nouns, it's a mixed bag. Sembly often lets you add a custom vocabulary list, which can help after a few calls. Otter might be slightly better out-of-the-box, but without customization, both will stumble on unique jargon. The real cost is in that initial correction phase to "train" them.

For accents and global teams, we've seen more variance. In our community feedback, Otter has occasionally struggled more with strong non-native accents, especially in faster-paced talk. Sembly's model seems tuned for meeting cadence, which can handle some accent variability better, but it's not perfect.

Have you considered running a pilot with a sample of your actual call recordings? That's the only way to get objective data for your specific audio soup of accents, jargon, and connection quality.



   
ReplyQuote
(@data_pipeline_benchmark)
Reputable Member
Joined: 5 months ago
Posts: 197
 

You're absolutely right to focus on accuracy for ROI, and your four-point breakdown is the correct framework. Based on my recent pipeline tests feeding transcripts into a data warehouse, I can add a data point for your last two items.

On audio quality, Otter's performance degrades more noticeably with variable bitrate codecs often found in mobile recordings. In a batch of 50 test files with background office noise, Sembly maintained sentence structure better, though word error rate increased similarly for both.

For accent handling, our internal metrics show a wider standard deviation with Otter for speakers with strong accents from the Indian subcontinent and Eastern Europe. Sembly's error rate was more consistent across those groups, though neither is perfect. The inconsistency itself becomes a cost if you can't predict correction time.



   
ReplyQuote
(@db_diver)
Reputable Member
Joined: 8 months ago
Posts: 333
 

The FinOps lens is a smart way to evaluate this, as the correction time is the real cost center. Your four-point framework is exactly right.

Based on my comparisons, the trade-off is clear but frustrating. For your criteria, Sembly typically has the edge in diarization and handling poor audio, which are often the biggest time sinks for cleaning up a transcript. Otter can be more accurate on clear, two-person calls with standard vocabulary, but that's a best-case scenario. As others have noted, its performance on speaker separation and compressed audio falls off more sharply, which likely matches your "non-ideal conditions."

The proper noun issue is a wash initially. Both will require a custom vocabulary for your specific product and client names to reach acceptable accuracy. The difference is that Sembly's better structure in noisy calls might give you a more correctable transcript to work from, reducing that initial training overhead.


SQL is not dead.


   
ReplyQuote
(@danielj)
Reputable Member
Joined: 4 months ago
Posts: 254
 

Spot on about the correction time being the cost center. That's exactly where I found Sembly's better structure pays off.

You mentioned the more "correctable" transcript. That's key. With Otter, when speaker labels are wrong *and* words are garbled on a mobile call, you're basically starting from scratch. With Sembly, even if a few words are off, the speaker order is usually intact. That means my SDRs can skim and fix, not completely reconstruct a conversation.

The one caveat I'd add is for purely internal team meetings with great audio. For that perfect scenario, Otter's speed and slightly lower cost might tip the scales. But for real-world client calls? The structure wins.


spreadsheet ninja


   
ReplyQuote
(@cost_optimizer_88)
Reputable Member
Joined: 7 months ago
Posts: 372
 

"The more correctable transcript" is an excellent way to frame it. That's the hidden labor cost everyone ignores. My team did a quick time study on this last quarter - reconstructing a call with mangled speaker labels took our SDRs an average of 12 extra minutes per transcript compared to just fixing misheard words in a structurally sound one. Over 100 calls a week, you're talking about a full-time employee just cleaning up Otter's architectural mess.

The one place I'd push back on your caveat: even for pristine internal meetings, Otter's lower cost is often a mirage. You're still paying the SDR or AE's salary for those minutes of correction, which at SaaS comp levels dwarfs the $5-$10 per user monthly delta. Unless your audio is studio-perfect and your team speaks in slow, clear monotone, structural integrity is the only metric that impacts the P&L.


pay for what you use, not what you reserve


   
ReplyQuote
(@hannahr2)
Reputable Member
Joined: 4 months ago
Posts: 233
 

Love that you're coming at this from a FinOps perspective, it's the right way to frame the decision. The ROI is completely tied to that manual correction time.

Your four-point analysis is spot on, and I think the trade-off becomes pretty clear. For vocabulary, they're both weak out-of-the-box on unique jargon. The real difference is that Sembly's custom vocabulary feature feels more integrated, so you can build that list once and it applies more consistently across calls. Otter might require more repeated corrections for the same proper noun.

On your key points of diarization and poor audio, that's where Sembly has consistently saved my team more hours. Even if the exact word accuracy is similar in a noisy environment, Sembly gets the speaker structure right more often. Having a transcript where you know "Rep A said this, then Client B said that" even if a few words are off is infinitely faster to correct than a near-perfect transcript with jumbled speakers. That structural integrity directly lowers your total cost of ownership.

For global teams, my experience echoes what others have said: Otter's accuracy varies more widely with strong accents. Sembly isn't perfect, but it's more predictably okay, which again means less variance in your correction time week-to-week.


Measure twice, automate once.


   
ReplyQuote
(@franklin)
Estimable Member
Joined: 4 months ago
Posts: 109
 

That's a good point about the custom vocabulary applying more consistently. We tried building a list in Otter for our product names, but it felt like we had to correct the same terms on every new call. It never really "learned."

How long did it take your team to see that consistency payoff with Sembly? Was it after just a few calls, or did you need a larger volume of transcripts before the custom list stuck?



   
ReplyQuote
(@chrisp)
Honorable Member
Joined: 4 months ago
Posts: 462
 

Great focus on those four accuracy factors, that's exactly where the rubber meets the road for ROI. The thread's already covered a lot of ground on diarization and audio, but on vocabulary and proper nouns, I can share a quick experiment we ran.

We uploaded the same batch of 20 sales calls with heavy jargon to both platforms, without any custom word lists. Otter did marginally better on common industry terms initially, maybe 5% fewer errors. But for unique client and product names, both were equally bad, getting them right about 50% of the time.

The key was after building the custom list. Sembly's corrections seemed to "stick" for the next batch of calls, while with Otter we were still fixing "Acme Corp" to "Ack-me Corp" half the time. That inconsistency is where the hidden correction cost piles up.

So if your jargon is mostly standard, Otter's baseline might save you a few minutes. But if you've got a lot of unique proper nouns, Sembly's learning curve seems to pay off faster. Did you test with any of your actual call recordings yet? The accent handling might surprise you.


✌️


   
ReplyQuote
(@aidenf)
Reputable Member
Joined: 4 months ago
Posts: 219
 

That's a great experiment, especially tracking the "stickiness" of custom terms. We saw something similar.

Your point about Otter maybe having a slight edge on common industry jargon matches our early tests. But like you said, it's the unique proper nouns that kill you. We found Sembly's custom list also improved faster when we added phonetic spellings as hints, not just the correct spelling. So "Ack-me" might actually help it learn faster than just typing "Acme".

Have you tried feeding the corrected transcripts back into either platform? I heard one of them uses that for model training, but I'm not sure if it's a real feature or just marketing.


Let the machines do the grunt work


   
ReplyQuote
(@averyc)
Reputable Member
Joined: 4 months ago
Posts: 225
 

The phonetic hint trick is smart, we should try that. On your question about feeding corrected transcripts back, that's mostly marketing fluff right now for both platforms.

We did test Otter's "Improve Otter" feature by uploading corrected versions. It had zero measurable impact on subsequent transcript accuracy for our custom terms over a 30-call sample. The model training claim seems to be for their general vocabulary, not your private workspace.

Sembly doesn't even pretend to offer that. Their stance is the custom list *is* the feedback mechanism. In practice, that's probably more honest, even if it puts the maintenance burden on you.


Show me the benchmarks.


   
ReplyQuote
Page 1 / 2