Alright, let’s cut through the marketing fluff for a second. I’ve been running Sembly for meeting transcriptions and AI summaries piped into HubSpot for about eight months now, which is practically a lifetime in my book. The core functionality works—meetings get logged, the summary pops into the meeting notes field. Fine.
But the entire value proposition of stitching this into a real revenue stack hinges on mapping those juicy AI-generated insights (like “discussed pricing tier Enterprise,” “flagged competitor X,” “next step: send security review”) into actionable, filterable **custom fields** in HubSpot. Which, for the first few months, it did. Now? It’s becoming a clown show.
Over the last three weeks, I’ve documented a systematic degradation. The integration is still *running*, no obvious errors in the logs they provide, but it’s selectively forgetting my field mappings. It’s not all of them; it’s the more valuable ones.
Here’s the corpse I’m currently examining:
* **“Meeting Sentiment” field:** Mapped to a HubSpot dropdown (Positive/Neutral/Negative). Sembly analyzes this, I can see it in the Sembly interface. For roughly 30% of recent meetings, this field is simply left blank in HubSpot. The data is generated but never arrives.
* **“Key Topics Discussed” field:** A multi-select property in HubSpot. Sembly’s summary will list topics, but the integration now dumps them as a comma-separated string into the notes field instead of populating the actual property. This breaks any automation or list building based on those topics.
* **“Decision Maker Identified” field:** A simple checkbox. Sembly’s AI is prompted to flag this. When it does, the checkbox remains stubbornly unchecked in HubSpot about half the time.
I’ve gone through the ritualistic troubleshooting dance, of course:
- Re-authenticated the HubSpot connection twice.
- Deleted and re-built the field mappings from scratch in Sembly’s settings.
- Confirmed the HubSpot property API names haven’t changed (they haven’t).
- Tested with different meeting types (Zoom, Teams).
The result is a CRM slowly filling with incomplete data, which is worse than no data at all because it gives a false sense of insight. My sales team is starting to ignore the fields because they can’t trust them.
So, my question isn’t just “is this happening?”—it’s “what’s the underlying architecture failure here?” Is this a silent API permission change on HubSpot’s end? A borked queue processor on Sembly’s side that fails on certain data types? Or the classic “we updated our AI model and broke the downstream data mapping” maneuver?
If you’re using this stack, check your historical data for the last month versus the last week. I’d bet a lukewarm coffee that your custom field completion rate is dropping. What are you seeing?