How Embark Built an AI Early-Warning System for Sales Calls

Watch This Episode of Building For Others

A finance director with minimal coding background built an AI system that monitors 100+ daily sales calls and surfaces critical issues to executives. Scott Graham proves that when every employee becomes a builder, value creation accelerates, turning months of manual work into automated insights delivered to Slack every morning.

Scott Graham isn't your typical AI builder. He's a finance guy by training who has found himself Director of Data and Solutions at Embark, a 500-person consulting firm. But when he looked at the mountain of unstructured call transcripts piling up in their system, he saw an opportunity that traditional data tools couldn't solve.

The result? An executive call elevation system that automatically reviews every sales and client call, identifies the ones that matter most, and delivers concise briefings straight to leadership's Slack channel at 6 a.m. every day.

What It Does (And Why It Works)

Think of it as an early-warning system for revenue opportunities and client risk.

Here's the problem Scott solved: Embark's sales team conducts 50-100 calls daily. Their head of sales can't possibly listen to all of them, but buried in those conversations are signals that matter. A major deal accelerating, a client expressing frustration, or a competitive threat emerging could easily get lost in the noise. Before Scott's system, these signals slipped through undetected.

His solution processes every transcript through a simple workflow:

• Chorus captures and transcripts calls
• Alteryx pulls transcripts via API into Snowflake for light data prep
• A workflow automation tool orchestrates the process: rule-based automation combined with AI analysis
• OpenAI's GPT reads each transcript, summarizes it, and decides whether leadership should see this
• Slack delivers the filtered results, typically 4-5 calls out of 100, with direct links to dive deeper

The beauty? It cost virtually nothing to build, runs in 15 minutes daily, and requires zero ongoing human effort. The stack is surprisingly lean—just 7-8 nodes in N8N doing the heavy lifting.

The "A-Ha" Moment

For Scott, the breakthrough came from realizing that AI's true value emerges when it's embedded in automation, applied through chatbots and other interactive tools.

"I've been using chatbots since ChatGPT launched," Scott explains. "Research assist is great, but what I wanted to figure out was: how do I erase manual labor and still get good outputs?"

The answer came when he discovered he could combine traditional rule-based automation (filtering, routing, formatting) with AI calls (summarizing, sentiment analysis, decision-making) in a single workflow. 

"We had tons of long-form data," Scott says. "This was the first time I was able to get that long-form text into a process that created value while requiring zero go-forward human effort."

What Made This Actually Get Used

Scott's system has been live for two months, and Embark's head of sales checks the Slack channel every morning. Why does it work when so many AI projects fail to gain traction?

The system delivers pure alpha. Unlike AI systems that replace human judgment and introduce new risks, this one simply catches things that would have slipped through the cracks. Salespeople still escalate deals they care about. Executives still use their intuition. But with his system, a safety net surfaces conversations they would never have had time to review.

Scott's also iterating based on feedback. He recently switched from GPT-4o to GPT-4.1 and noticed it's flagging more transcripts than before. That increase might mean adjusting the prompt to be more selective. The system operates as a living tool that gets refined as the team learns what's most valuable.

Three Lessons for Building AI That People Actually Use

1. Solve a Painful, Specific Problem

Build around a concrete need rather than a general-purpose AI tool, which rarely stick. Scott's advice: "Dive into an actual need. Really think about: What do I dislike doing? What do I want to take off my plate?"

At Embark, the surprising AI champions were the sales reps rather than the business transformation consultants. Why? Because AI solved their most tedious problems: meeting prep, account research, and crafting narratives. They wanted to spend less time researching and more time building relationships.

The takeaway: Start with a task that genuinely burdens someone's day. If it saves time or prevents mistakes immediately, adoption follows.

2. Deliver Where People Already Work

Scott's mantra: "Don't ask people to go somewhere new."

His executive call system delivers to Slack because that's where leadership already spends their day. Not a dashboard. Not a new app. Not even an email. Slack.

"I know everyone's already in Slack," Scott says. "I want it to show up easily in front of you rather than asking people to go to a link to find these things."

When evaluating vendors, one of Scott's first questions is: "How integrated is this with Salesforce?" Because managing business across four tools guarantees failure.

The takeaway: Match your workflow to existing habits. Adoption requires aligning with how people already work, not asking them to change their behavior.

3. Build a Quick POC, Then Iterate

Scott's executive call system went from idea to working prototype in just a few hours. Then he spent two weeks refining it before sharing with leadership.

"Get that POC up and running," he advises. "Get the basics. We spun up an idea in a few hours, then iterated for a month before I felt ready to share this with people."

He's not done. Next on his list: using transcript analysis to inform pricing strategy, improve competitiveness, and increase win rates. The system functions as a foundation rather than a finished product.

The takeaway: Build something valuable quickly, even if it's just for yourself. Let real-world use drive your iterations rather than waiting for perfection.

Want to Learn More?

Scott Graham shares his AI experiments and lessons learned on LinkedIn. If you're interested in building custom AI assistants for your team (like the sales prep bots Embark's reps love), check out elvex to explore how you can enable your employees to become builders too.

The future of work centers on giving every employee, even finance directors with no coding background, the tools to automate the tedious and focus on what matters most.