How Aloware Built a Zoom Transcript Agent That Logs CRM Deals in One Emoji

Originally published on the Aloware blog. Republished here with permission.

Disclosure: Aloware is a Runbear customer. This case study was developed from their production implementation.

When Aloware's sales team came to us, they had a logging problem that no amount of process reminders could fix. I remember sitting with their ops lead as they described the "admin debt" that was crushing their reps' productivity. Aloware is an AI-powered contact center platform used by more than 1,000 B2B sales and support teams. Their reps live in calls. On a typical demo day, a single sales rep might run five back-to-back Zoom sessions. Each one generates follow-up tasks, account notes, and CRM entries that need to be logged before the next call starts. Aloware is a Runbear customer.

The problem was that logging almost never happened.

A Zoom transcript agent is an AI-powered Slack assistant that automatically processes meeting recordings, extracts structured data, and prepares CRM updates for one-click approval.

The Manual CRM Problem

Sales reps at Aloware were expected to log each demo call in HubSpot. This included the deal name, stage, contract details, attendees, pain points, action items, and next steps.

In practice, after a long demo day, nobody wanted to open HubSpot and fill in eight fields from memory. Calls went unlogged. Deals were created late or not at all. Pipeline reports became unreliable. Salesforce research has shown that reps spend only 28% of their week actually selling, with the rest consumed by administrative tasks like manual data entry. HubSpot research confirms this remains one of the top productivity bottlenecks for sales teams.

The usual answer was better process. Aloware tried reminding reps to log, making it easier, and adding templates. We saw the same pattern repeated: the behavior did not change. This was not because reps were lazy, but because the task was genuinely annoying to do after an hour-long call.

The real problem was not discipline. It was that logging a call manually is the hardest thing to do right after you have finished one. I spoke with the Aloware ops team about this, and they confirmed that even their most diligent reps were losing an hour a day to "admin debt."

This maps to a pattern we have written about before called the Actions Gap. This is the gap between having the information and actually executing the update. In our framework, the "Actions Gap" occurs when AI can draft the entry in seconds, but without a system that takes that last step, the draft lives in a chat window while the CRM stays empty.

The Actions Gap: The disconnect between having the necessary information generated by AI and actually executing the required update in the system of record.

What They Built

Aloware built a Zoom Transcript Agent on Runbear. This was the first of two agents they deployed. The second, AloPedia, handles company-wide knowledge retrieval.

The flow for the Zoom agent is simple:

• A Zoom call ends and a transcript is generated automatically.
• The agent processes the transcript. It extracts the deal name, contact, pipeline stage, contract term, and account type. It also creates a structured summary of attendees, pain points, action items, and next steps.
• The agent checks HubSpot for existing deals to avoid duplicates.
• Within minutes of the call ending, the rep receives a pre-filled CRM deal proposal directly in Slack.

There is no new app to learn and no context switch required. The information appears where the rep already is. The entire build ran on Runbear without needing custom code.

The Approval Loop

The Slack message is not just a notification. It is a complete decision interface. The rep sees the proposed deal with all eight fields pre-filled and responds with a single emoji.

They can:

• Approve: The deal is created in HubSpot exactly as proposed.
• Edit: The rep types a correction in plain English. For example, they might say "change the stage to Demo Scheduled." The agent updates the draft and resubmits it.
• Reject: No action is taken and nothing is created.
• Route: The agent sends the deal proposal to a different colleague for review if the primary rep is not the correct owner.

The correction loop is important. The agent does not need to be perfect. It needs to be close enough that fixing it takes five seconds instead of five minutes. In practice, most proposals go through without edits. When we reviewed the logs with Aloware, we found that when a rep does correct something, the agent's accuracy for that specific deal type improves in the very next interaction. This "feedback-to-execution" loop is what builds long-term trust. This is what we mean when we describe ops automation that doesn't create bottlenecks. The human stays in the loop, but the loop takes seconds instead of minutes.

Smart Routing: Sales vs. Support

Not every call needs a CRM deal entry. Aloware's customer experience and support teams also run Zoom calls, but a support session does not belong in the deal pipeline.

The agent handles this with smart routing. Sales calls trigger the full CRM flow. Support calls receive a structured summary only. This includes attendees, issues discussed, and follow-ups.

Results

The deeper change is behavioral. Reps no longer think about CRM logging as a task. It happens automatically and they spend five seconds approving it. Sales managers now have reliable pipeline data without having to chase down incomplete records or deduplicate entries at the end of the week.

As James Chen, Senior AE at Aloware, put it: "Before this, logging a call was the last thing anyone wanted to do after long demo days. Now the bot sends me everything in Slack already filled in. I tap one emoji and it is done."

Key Design Decisions

A few things Aloware got right that made this work:

Human-in-the-loop by default.

The agent never writes to HubSpot without approval. This is a trust mechanism. Reps adopted the tool quickly because they knew nothing would be created without their say-so.

Conservative AI extraction.

The agent extracts only what it can confirm from the transcript. If a field is ambiguous, it flags it rather than guessing. This keeps the correction rate low and prevents bad data from entering the pipeline.

Plain English corrections.

The edit flow does not require reps to navigate fields or click through a form. They type what they want changed, the same way they would tell a colleague.

These decisions connect to the three types of ops requests. We categorize requests into Type 1 (simple retrieval), Type 2 (synthesis and context), and Type 3 (complex judgment). The Zoom Transcript Agent is a Type 2 workflow.

How It Works on Runbear

Runbear is a Slack-native AI platform that connects to your tools and lets you build agents that work inside your existing workflows. For Aloware, that meant connecting Zoom transcripts, HubSpot, and Slack.

Setup took about 10 minutes. There was no code, no integration engineering, and no new app for the sales team to learn.

Ready to build your own? Try Runbear for free today.

The Invisible Cost of Admin Debt

Admin debt isn't just a nuisance; it's a compounding tax on your sales team's momentum. When a rep finishes a demo, they are at their peak level of insight about that deal. They know the customer's hesitation, the specific feature that made their eyes light up, and the exact follow-up needed to close. But as the day progresses and more calls are stacked, that resolution fades. By 5 PM, the nuances of an 11 AM call are lost.

For Aloware, this admin debt meant that their CRM was always 48 hours behind reality. Forecasts were based on old data, and managers couldn't support deals in real-time. By automating the capture and extraction phase, the Zoom Transcript Agent ensures that the data is logged while it is still fresh, eliminating the cognitive load of retrospective entry.

Connecting the Stack: Zoom to HubSpot via Slack

The technical implementation uses Runbear's native connectors for Zoom and HubSpot. When a recording is finalized in Zoom, a webhook triggers the Runbear agent. The agent then retrieves the transcript and uses its context window to identify key entities. Because the agent is connected to HubSpot, it can perform a real-time lookup to see if the contact or company already exists, preventing the common problem of duplicate records.

The most impressive part of the workflow is the "Natural Language Correction" feature. If the agent misidentifies a deal stage, the rep doesn't have to open a form. They simply reply to the Slack message with "Actually, this is still in the Discovery phase," and the agent re-processes the draft before submitting it to HubSpot. This maintains data integrity while keeping the user experience entirely Slack-native.

How to Get Started with Your Own Transcript Agent

Building an agent like Aloware's doesn't require a month-long development cycle. In fact, most teams can have a functional prototype running in under an hour. The process follows three simple steps:

• Connect your tools: Authenticate your Zoom and CRM accounts within the Runbear dashboard.
• Define your schema: Tell the agent which fields you want to extract from your calls (e.g., deal size, next steps, pain points).
• Set your triggers: Choose which calls should trigger the agent and who should receive the approval proposal in Slack.

Aloware's experience shows that the biggest barrier to CRM hygiene isn't a lack of tools, but the friction of using them. By bringing the CRM to where the team already lives, Slack, you can transform your data from a lagging indicator into a real-time asset for your entire organization.

See also: How Aloware Built AloPedia: A Company-Wide Knowledge Agent