Improving Team Success via AI Meeting Feedback
In January 2024, I found myself overhearing a video call at a cafe—what sounded like a budding relationship playing out on screen. The person leading the conversation clearly meant well, but I couldn’t help thinking: He could really use a coach. Someone—or something—to guide him toward deeper listening, better questions, and more presence.
That moment sparked an idea: What if we could build tools that help people communicate more effectively in real time and reflect more thoughtfully afterward? Given how much of our communication now flows through smart devices—and how far natural language processing (NLP) has come—it seemed worth exploring.
Initially, the idea focused on social coaching. But momentum really picked up at Savas Labs, where we were already using Fathom, a video intelligence tool that integrates with Zoom and Google Meet to record, transcribe, and summarize meetings. We started recording nearly every meeting and quickly saw the benefit: easier follow-ups, clearer records, and more space to focus during the moment.
That got us thinking—what if we layered on AI to analyze how we communicate, not just what was said?
We launched an internal R&D effort under Savas Labs’ Labs initiative (read more here) and partnered with Matters to build a lightweight tool. Our goal was simple: start with post-meeting feedback based on the transcript alone. No complex setup. No need for training the model with deep context. Just: Was this an effective meeting? Where could we improve?
To stay focused, we started with three meeting types:
Sales prospecting calls, where coverage of key questions is crucial.
Client kickoff meetings, where preparation and clarity of materials matter most.
Manager 1:1s, which involve performance and presence from both participants.
What emerged was fascinating: the metrics of a “good meeting” differ wildly depending on the type. For example, kickoff meetings aren’t just about what’s said—they hinge on prep and slide quality, which prompted us to build a slide submission and evaluation feature. In 1:1s, it’s not just the manager’s performance that counts; the health of the relationship is co-created.
These insights began shaping the product in real-time as Savas teams started using early prototypes. We’re now just a few sprints away from a functional MVP that works across all three use cases.
Progress also leapt forward during a recent Savas Labs Hackathon, where a small team fast-tracked key functionality. We’re excited to keep building—and to help make meetings not only more productive, but more human.
Want to learn more or help shape the future of feedback? Reach out—we’d love to talk.