From $0 to $400M: The Playbook I'd Run Differently Today (With AI)

In 2010, I walked into Groupon and started building a sales team from nothing. Two years later, we had 400+ sellers generating $415M in revenue across 23 markets.

It was chaotic, fast, and built almost entirely on human intuition, hustle, and luck. We made a lot of money and a lot of mistakes.

I've been thinking about what I'd do differently if I were running that play today — with the AI tools that exist right now. The answer changes almost everything about execution, while the core strategy stays the same.

What Wouldn't Change

The fundamentals of building a high-velocity sales org are not disrupted by AI. They're amplified by it.

You still need:

None of that changes. AI doesn't replace the strategic clarity required to build a sales organization. If anything, it makes strategic clarity more important — because AI amplifies whatever direction you're already moving.

What Changes Completely

Market Mapping and ICP Development

At Groupon, we built our ICP by having reps call everyone and figure out who bought. It worked, but it took six months to develop the pattern recognition that told us where to focus.

Today, that analysis happens in days. You feed your closed-won data into an AI model, it identifies the firmographic and behavioral patterns that correlate with fast closes and high retention, and you have a refined ICP before you've made 100 calls.

We would have known in week three what we figured out in month six. That's not a marginal improvement — it changes your entire ramp trajectory.

SDR Productivity

At Groupon's scale, a significant portion of our team's time was spent on research, data entry, and list building. We had people whose primary job was to find phone numbers. That's not an exaggeration.

With today's Clay + AI stack, one ops person can do what a team of researchers did in 2010. Your SDRs spend their time on conversations, not prep work. The unit economics of outbound change fundamentally.

Forecasting and Pipeline Management

I made a lot of forecast decisions on gut feel in 2010. Sometimes I was right. Often I wasn't. When you're scaling at that speed, a bad call on where to add headcount or where to pull back costs you months.

AI-native forecasting gives you signal you can actually act on: which markets are overperforming, which teams are seeing deal velocity slow, which rep cohorts are underperforming relative to their pipeline. Real-time intelligence, not end-of-quarter surprises.

Enablement at Scale

When you hire 400 people in two years, enablement is a crisis response. You build content, you run trainings, you hope enough of it sticks. The reps who figured it out faster were the ones who had the most coaching time with the best managers — which was random and inequitable.

Today, you build AI-powered coaching tools. Every rep gets immediate feedback on their call quality, their deal qualification, their email copy. The best manager's judgment gets embedded in a tool that scales to every rep simultaneously.

That's not a nice-to-have. That's the difference between your 50th percentile rep performing like your 70th percentile rep. Across 400 people, that's hundreds of millions of dollars.

The One Thing That Doesn't Change

Here's what two decades in this business has taught me: the quality of your people still determines the ceiling.

AI raises the floor dramatically. A mediocre rep with great AI tools can outperform a good rep without them. But the great reps — the ones with judgment, resilience, strategic thinking, and the ability to build genuine trust with buyers — are still the difference between good and exceptional.

What AI does is make it clearer, faster, who those people are. The signal-to-noise ratio on performance improves when you remove the tasks that let average performers hide.

That's uncomfortable for some people. It's an enormous advantage for the ones who are genuinely excellent at the human parts of the job.

If I were building Groupon's sales team today, I'd hire fewer people, deploy them with better tools, and set a higher bar for what "good" looks like. The output would be the same. The efficiency would be unrecognizable.

That's the playbook I'd run today.