How Fireflies AI Skills Work: Setup, Costs, Governance (Full Transcript)

[00:00:02] Speaker 1: Transcripts and summaries are useful, but what if Fireflies could automatically score sales calls, generate interview feedback, or extract competitor mentions without any manual work? AI skills are automated workflows that analyze your meetings and generate specific outputs. Think of them as custom instructions. After every sales call, extract budget, authority, need, and timeline. Or, after every interview, generate a scorecard based on our hiring rubric. Some skills are pre-built templates. Vant analysis, action item extraction, meeting summaries. Enable them, and they run automatically on matching meetings. You can also build custom skills from scratch. Define what you want extracted, how it's formatted, and which meetings it applies to. AI skills run after meetings are processed. The output appears in the meeting recap and can be pushed to integrations like HubSpot, Slack, or Notion. As an admin, you control two things. Who can create AI skills, and who can access them? If you allow everyone to create, you'll get experimentation, but less consistency. If you restrict to admins, you maintain control but become a bottleneck. Most teams land in the middle. Admins and team leads create, everyone else uses what's built. Access controls who can see and use existing skills, not who can build new ones. Our recommendation. Let everyone access AI skills, but restrict creation to admins and power users. You get the benefits without the chaos of 50 overlapping skills. Individual skills also have visibility settings. Only me, teammates, or shared via link. Use this to test skills privately before rolling them out. AI skills consume AI credits, a usage-based resource separate from your Firefly subscription. Credits power AI skills, ask Fred queries, magic soundbites, and custom summary sections. Standard transcription and basic summaries don't consume credits. Firefly's AI features used dynamic credit pricing. You're charged per request, so the credits used depend on how much work it takes to generate the output. Simpler requests use fewer credits. More complex ones use more. More skills on more meetings equals more credits. This is why governance matters. You don't want 20 skills running on every meeting if only three are useful. You can purchase additional credits as needed. Monitor usage monthly and disable skills nobody looks at. Those are the biggest waste. Something that catches admins off guard. AI skills can only analyze meetings they have access to. If you create a skill set to run on all workspace meetings, but half your team has privacy set to only owner, the skill won't see those meetings. A workspace-wide skill produces incomplete results if privacy is inconsistent. If you want AI skills to work across your entire workspace, you need consistent privacy. Set team defaults to teammates or broader, or accept that skills only analyze meetings with open privacy. Configure privacy defaults before rolling out AI skills. Start with one high-value skill per team. Expand based on what's actually being used, not what sounds cool. As adoption grows, governance becomes important. Three questions to answer. Who can build new skills? What skills are running and on which meetings? How much is it costing in credits? Review credit usage regularly. Identify which skills consume the most. A skill that costs 50 credits a week but nobody reads should be disabled. You can enable or disable skills anytime. Disabling stops them from running on future meetings without deleting past outputs. Assign an owner to each skill. Someone responsible for monitoring whether it's useful. Review your skill inventory quarterly, like integrations or user groups. Control who builds AI skills. Monitor credit usage. Align privacy so skills have data access. And start with high-value workflows per team.