Skip to content
Notifications
Clear all

Sembly after 6 months - honest review from a 40-person SaaS company

3 Posts
3 Users
0 Reactions
0 Views
(@data_analytics_rover)
Reputable Member
Joined: 4 months ago
Posts: 150
Topic starter   [#10752]

Our team adopted Sembly six months ago as our primary meeting transcription and AI summarization tool. We're a 40-person SaaS company with a fully remote engineering and product team, averaging 20-25 recorded meetings per week. The goal was to improve meeting note accountability and create a searchable knowledge base from discussions.

**Initial Setup & Core Functionality**
The onboarding was straightforward. We connected our Google Workspace calendar, and Sembly began joining and recording designated meetings automatically. The transcription accuracy is generally good for clear English speakers, though it struggles significantly with technical jargon and non-native accents, often requiring manual correction of key terms. The AI-generated "Meeting Brief" with bullet-point summaries, decisions, and action items is the primary value driver. It saves our project managers an estimated 3-4 hours per week.

**Integration & Workflow Gaps**
Our main pain point is the lack of deep integration with our core tools. While it can post summaries to Slack, the data feels siloed. We attempted to use the API to pipe structured outputs (decisions, action items) into our data warehouse for tracking in our internal dashboards, but the JSON schema is inconsistent. For example, the key for action items varies between `action_items` and `actionItems` across different API responses.

```json
// Example of inconsistent API response structure
{
"meeting_id": "123",
"summary": {...},
"action_items": [{"owner": "John", "task": "draft spec"}] // sometimes present
}

{
"meeting_id": "456",
"summary": {...},
"actionItems": [] // sometimes present with different key naming
}
```

This inconsistency makes building a reliable data pipeline complex and forces us to add significant data cleaning logic in our dbt models.

**Performance & Cost Benchmark**
We are on the "Teams" plan at $20/user/month. For our active users (approx. 25), this is a $500/month investment.
* **Transcription Speed:** 5-7 minutes for a 60-minute meeting.
* **Accuracy Rate:** We manually sampled 100 technical sentences; ~85% were correct for intent, but key product names and code concepts (e.g., "idempotent retry logic") were wrong ~40% of the time.
* **Storage:** No issues with the 1000 hours/month limit.

**Conclusion after 6 Months**
Sembly provides good baseline transcription and summary capabilities for general business meetings. For a non-technical team, it would likely be a strong recommendation. However, for a technical SaaS company, the value is diminished by:
* Poor handling of specialized vocabulary.
* Lack of reliable, structured data output for integration into the modern data stack.
* Inconsistent API, which increases the engineering burden for automation.

We are continuing our subscription for now due to the time savings on general meetings, but we are actively evaluating competitors that offer more robust APIs and better support for technical terminology. The search for a truly "analytics-engineer-friendly" meeting intelligence tool continues.



   
Quote
(@devops_rookie_2025)
Reputable Member
Joined: 2 months ago
Posts: 203
 

Thanks for sharing this, really helpful to hear from a team that's actually using it at scale. I'm just starting to look into meeting transcription tools for our small team, so this is gold.

The part about the API and siloed data caught my eye. We're also trying to build a searchable knowledge base and I'm worried about getting locked into yet another tool that doesn't talk to our stack. Have you tried any workaround like pulling the raw JSON from the API and feeding it into something like a Notion database or a simple Airtable? I'm curious if the structured outputs are actually usable or if they're just as messy as the transcriptions.



   
ReplyQuote
(@cloud_cost_auditor)
Estimable Member
Joined: 3 months ago
Posts: 106
 

The API angle is the real trap. They'll sell you on the structured JSON, but you're still paying their API call rate to get *your own data* out. That's the lock-in.

We tried piping the summaries into Confluence for a client. The structure was usable, but you end up building and maintaining the entire integration pipeline yourself. The cost wasn't just the tool's subscription, it was the engineering hours to make the data vaguely useful elsewhere.

Did you run the numbers on what it would actually cost to extract, transform, and store that volume of transcript data on your own infra monthly? I'd be curious to see that break-even point versus just paying for the convenience.


Show me the bill


   
ReplyQuote