Automating sales: How AI agents handle our post-call workflow end to end

When it comes to our sales process, we weren’t starting from zero. We’d been using Fathom as our notetaker for a while. It joined every sales call, recorded it, generated a summary, and synced that summary to HubSpot. For a lot of companies, that’s the finish line. Notes in the CRM, call recorded, done.

But we needed more. Fathom’s summaries are great for a quick recap. They’re not deep enough for what we actually wanted to do: analyze calls against our specific sales process, prepare detailed briefs for demos, run coaching across every conversation. More importantly, these transcripts are the key piece of data for other departments, specifically Marketing and Product. We will come back to these use cases later. To do any of that, we needed the full transcripts. Every word, structured, and in a shared space the whole team could access.

This is the story of how we built Transcript Sync Agent – the foundation for everything else we’re automating in Sales.

The real starting point: Good enough that wasn’t enough

The old process was working. Every sales call was recorded. Notes were in HubSpot. Reps could reference past conversations. It was organized.

The problem wasn’t disorder. The problem was depth. Fathom summaries gave us the headlines, not the full story. And when you want an AI agent to analyze a conversation against your specific qualification criteria, or prepare a tailored demo brief, or give a rep personalized coaching, headlines aren’t enough. You need the transcript.

We also had a second problem: we wanted this intelligence to flow across the company. Marketing needs to hear how prospects describe their pain in their own words. Product needs to know what features come up in demos. You can’t build that from summaries. You need the source material.

Step 1: Full transcripts in a shared space

The first thing we built was a pipeline that captures full transcripts from every external client call and stores them in a shared company space.

Two decisions mattered here.

• Only external calls. We filtered out internal meetings, standups, one-on-ones, and team syncs automatically. The system only processes calls with prospects and customers. This keeps the data focused on what drives revenue and avoids drowning the workspace in noise.
• Shared, not siloed. Transcripts don’t live in one rep’s account or one tool’s dashboard. They go to a central location where anyone with the right access can find them. Sales, Customer Success, Marketing, Product. The same source of truth. When a CS manager needs to know what was promised during the sales cycle, it’s there. When Marketing wants to pull real customer language for a case study, it’s there.

This sounds simple. It mostly was. But getting speaker identification right, detecting the call type (e.g. discovery vs. demo), and making transcripts actually searchable took more iteration than we expected.

Step 2: Analysis tailored to our process

Here’s where it gets interesting. We didn’t invent a new way to analyze sales calls. We automated what we were already doing by hand.

Before Transcript Sync, someone on the team would take a call recording, pull the transcript, paste it into ChatGPT, and run it through a set of prompts tailored to our sales process. It worked well. The analysis was good. But it took time, and it wasn’t 100% consistent.

Now the AI agent does this automatically for every call. As soon as a transcript lands in the shared space, the agent picks it up and runs the full analysis:

• From the deal perspective: it pulls out the prospect’s pain points, their timeline, budget signals, decision-making structure, and competitive mentions. It maps what was said against our qualification criteria. Every deal gets the same rigorous analysis, whether the rep is senior or brand new.
• Next steps extraction: the agent identifies every commitment made on the call, both by us and by the prospect. “I’ll send you the security questionnaire by Friday.” “We need to loop in our CTO.” These get structured and flagged so nothing falls through.
• Demo prep briefs: for discovery calls, the agent generates a structured brief for the rep who will run the demo. It highlights what the prospect cares about most, what to emphasize, what to skip, and any red flags to address. The sales rep walks into the demo already knowing what matters to that specific prospect. No 30-minute prep huddle. No “can you give me the quick version of what they said?”

The key insight: the prompts and criteria behind this analysis are ours. We built them around our sales methodology, our qualification framework, our definition of a good deal. The AI didn’t define the process. We did. The AI just runs it consistently, every single time.

Our next step was to create the coaching agent – Bill. We‘ll talk about him in our next part.

Also, we have some additional automation items and agents on our roadmap that we plan to tackle soon:

• Discovery call preparation and research
• Demo deck creation
• Client-specific sample project data creation within Birdview

What surprised us

• The filtering was harder than the analysis. Getting the AI to analyze a transcript well was the easy part. Reliably identifying which calls are external client calls versus internal meetings, filtering out the noise, handling edge cases where an internal call includes a customer, that took more work than we expected.
• Full transcripts unlock things you didn’t plan for. We built this for Sales. But within two weeks, our Marketing team pulled the transcripts and extracted ICP insights we’d never formalized. The exact words prospects use to describe their problems. The job titles that show up on calls most often. The patterns in what makes someone a good fit versus a bad one. They built a sharper ideal customer profile almost immediately, straight from real conversations instead of assumptions. Customer Success started referencing what was promised during pre-sale. Product began reading transcripts to understand feature requests in context. A single source of truth turns out to be useful for everyone.

What we’d do differently

We spent too long debating where the AI output should land. Email summaries? HubSpot notes? Slack messages? We tried different combinations before finding what actually works: Slack for the immediate alert and quick summary, HubSpot notes for the detailed analysis that lives with the deal. Reps check Slack constantly. They don’t go hunting through email for a post-call report. And when they need the full breakdown later, it’s right there in the CRM next to the deal. That combination stuck.

The bigger lesson: we should have asked the sales team from the start. Build with the people who’ll use it, not for them.

The takeaway

If you’re thinking about automating your sales process with AI, don’t start with the fanciest use case. Start with the foundation: get your client conversations into a single, shared, structured space. Full transcripts, not summaries. Accessible to the teams that need them, not locked in individual tools.

Then automate the analysis you’re already doing. If your best reps or managers are manually reviewing calls and pulling out insights, that’s your starting point. Codify their process. Let the agent run it consistently.

Next week, we’re going deeper on AI coaching for sales reps: how we built an agent that remembers every conversation, learns each rep’s style, and gives feedback the way Bill Campbell would. Personalized, direct, and based on who the person actually is.

This is Part 2 of “Becoming AI-Native,” a weekly series from the Birdview PSA team on our AI transformation journey. Follow along here on the Birdview blog, or on Vadim’s LinkedIn.