How I Built an AI-Powered Sales Call Assistant MVP in One Weekend
Technically, in one long weekend...
It’s not often that Atlanta gets snow, and when it does, the whole city shuts down. That’s exactly what happened last Friday. With nowhere to go and three full days ahead of me, I decided to use the unexpected downtime to solve a problem I’ve seen sales teams struggle with for years: the tedious task of updating Salesforce after a call.
By Sunday night, I had built a working MVP (Minimal Viable Product) for an AI-powered sales call assistant. It wasn’t perfect, and turned out to be more of a POC (proof-of-concept) but it was enough to prove that an AI Assistant can help Sales people spend less time updating CRMs, and more time actually selling!
The Problem: Sales Reps Are Buried in Admin Tasks
Picture this: You’re a sales rep wrapping up a Zoom call with a high-potential prospect. During the call, you discussed:
Their pain points and how your product solves them
A follow-up demo to loop in their VP of Operations
Sending over pricing options and a case study
The call ends, and now you have to:
Summarize the discussion
Enter the notes into Salesforce
Log follow-ups and tasks in your to-do list
These admin tasks are repetitive, time-consuming, and frankly, the kind of work reps hate doing. Worse, details can get lost or forgotten, potentially jeopardizing the deal.
The Hypothesis: I Can Use AI Tools To Build an AI Assistant for Sales in One Weekend
I thought, What if there were an AI-powered assistant that could sit in on calls, take real-time notes, and handle the admin work automatically?
Here’s the vision I started with — the AI Sales Assistant would:
Take notes in real-time during sales calls
Categorize key points like next steps, decisions, and follow-ups
Automatically update Salesforce, eliminating manual data entry
Send proactive reminders to ensure nothing slips through the cracks
It seemed simple enough, so I gave myself a challenge: Build a working prototype in one weekend.
How I Built the AI Sales Assistant MVP in 3 Days
Day 1: Integrating Zoom and Google for Near Real-Time Notes
The first step was connecting to Zoom’s API to access call recordings. I paired this with Google Cloud Speech-to-Text for near real-time transcription. By the end of Friday, I had basic functionality:
Although I was not able to get calls transcribed in real-time, I was able to use the Zoom recording after the call to parse out basic keyword/phrase detection.
The system could recognize key phrases like “Follow up” or “Next step.” Of course, a real conversation with multiple people and dialects would add complexity, but the purpose of this exercise was to prove out the concept.
Day 2: Categorizing Notes and Updating Salesforce
On Saturday, I used a Salesforce free trial instance to simulate how the tool would integrate with a live CRM. Using Zapier, I automated the process of pushing categorized notes into the appropriate Salesforce record. For example:
“Send the pricing proposal” was logged as a task under the opportunity.
“Schedule a demo next week” created a calendar event.
For this one, the Zoom call had to generated from the actual opportunity, otherwise the system struggled with where to place the records.
Day 3: Adding Slack Reminders and Polishing
Sunday was all about usability. I used Zapier to integrate with Slack, so follow-ups could automatically notify reps and managers.
By Sunday night, the MVP could:
Transcribe and categorize notes during a Zoom call
Automatically update Salesforce with structured notes and tasks
Notify the sales rep in Slack about upcoming deadlines
The AI Sales Call Assistant in Action
Here’s how it works:
You hop on a Zoom call with a prospect.
The assistant quietly listens, transcribes, and categorizes key points in real-time. (future). For the MVP, the call is transcribed but the categorization and Salesforce updates occur after the call, based on the recording.
Minutes after the call ends, Salesforce is already updated with:
A clean summary of the conversation, logged as call activity for the Contact
Tasks for follow-ups, like sending documents or scheduling a demo, logged at the Opportunity level
Events with deadlines created on the user’s Gmail calendar
Slack reminders set one day before each due date
The result? No more manual data entry, no missed action items, and more time to focus on selling. Although this solution is not production-ready, as a proof-of-concept, it shows just how much potential there is to free up sales reps from tedious admin work. These limitations aren't roadblocks; they're really just to-do list items for the next iteration.
Why I Think This Could Change the Game
Sales reps shouldn’t have to spend hours on admin tasks. This AI assistant simplifies workflows, saves time, and ensures nothing gets missed.
For Reps: Spend less time on CRM updates and more time closing deals.
For Managers: Get better data and improved pipeline visibility.
What Do You Think?
If you’re in sales, would this save you time?
Drop your thoughts in the comments or DM me—I’d love to hear your feedback as I continue refining this tool.