I Returned as CEO Just as AI Took Off. These 4 Lessons Kept Our My Spending Honest

I picked an interesting time to return to my company as CEO.

I came back in early 2024 with a mission: rebuild our tech business around profitable growth. That shift was part of a broader trend in which companies from Netflix to Meta to Peloton shifted focus from growth and revenue at any cost to profitability.

My return, after stepping back for two years, also coincided with the rapid rise of large language models. The business world had changed during my absence, and it was changing faster every day.

I’d founded the company in 2005, so I was no stranger to tech upheavals. I knew that AI would have to play a central role in our transformation. But I also saw a disconnect. Too many companies were throwing money at AI, with little apparent regard for the bottom line.

We had to invest. But we needed discipline. How could I stay fiscally responsible, lead the business to profitability and make sure AI created real value for our customers and us?

Here’s what I’ve learned about finding the right balance.

1. Put profitability first

Few companies have pockets as deep as the Magnificent 7, which can absorb a failed AI bet. If you don’t, every investment must be targeted and financially defensible.

When I returned, we were approaching a billion dollars in revenue but with limited adjusted EBITDA. That shaped the agenda for everything that followed, including our AI spending. A year after I came back, we set a goal: $100 million in free cash flow within three years, proof of the business’s underlying value and durability.

It all started with finding cost savings. This bought us time to conduct a full operating review, see the full investment picture and focus our resources rather than funding every AI initiative.

Financially defensible means holding each AI commitment to the same bar as our free cash flow target. If a bet doesn’t clearly protect or advance that number, it doesn’t get funded, no matter how promising the technology looks.

That bar is easiest to miss on the cost side, where multiyear contracts for coding tools like Codex, Cursor and Claude can lock a company into tens of millions in spending. And unlike, say, personal computers or many cloud services, AI doesn’t necessarily get cheaper as the technology matures. Larger, more powerful models can cost much more to run.

Just ask Uber, which burned through its entire budget for AI coding tools in just four months, leaving its COO questioning whether the return justified the cost. At our scale, where every million matters to adjusted EBITDA and free cash flow, a move like that could erase the progress we need to show shareholders.

The solution: match the AI model — and its cost — to the task. We chose OpenCode, an open-source, model-agnostic coding tool, over locked-in frontier subscriptions. This lets us switch to cost-effective LLMs for relatively simple tasks, like crunching retail data.

2. Balance urgency with a clear strategy

Stepping back into the CEO role also meant determining how big a threat AI posed to our business. Almost overnight, the market had decided that every SaaS company was potentially under threat.

By early this year, software companies faced growing pressure to rebuild their pricing models around AI consumption. Spoiler alert: we didn’t go that route. Instead of racing to adopt a new strategy, we considered where AI fit naturally into our existing one.

FOMO can be an expensive way to make business decisions. For example, a third of US hiring managers who cut a role because of AI have since rehired for that job or a similar one. That pattern extends well beyond hiring to any AI-related investment made in haste.

Companies need a sense of urgency in the AI era, but it has to serve the strategy, not replace it. Before following a trend, ask two questions: Does it move the business in the direction you’ve already chosen? And can you explain how it will create measurable value?

3. Understand what AI means for your customers

Wondering why we didn’t rebuild our pricing around tokens? Our business model already aligned our incentives with customer success.

Most of our revenue comes from transaction fees, which grow when our customers’ businesses grow. Switching to AI-usage pricing would have weakened that alignment by tying revenue to how much technology customers consume, rather than how much value they gain from it.

If your business model rewards results, that’s worth defending rather than abandoning it for the latest pricing fad.

The more important question is what customers actually need AI to do. For example, the people who use our products wear many hats: marketer, social media manager, inventory manager, analyst, in-store experience expert — sometimes all in the same day. AI can help them juggle those responsibilities by doing everything from delivering data-backed insights to recommending new brands to quickly building a professional website.

Because our success depends on merchants’ success, our starting question was how AI could make their lives easier, not how much they might use it.

Before deploying AI for customers, figure out where they struggle. Technology will keep changing, but the fundamentals of delivering real value won’t.

4. Use AI to redesign work

I came back to a tangled web. We had two flagship products, a host of different systems from acquisitions and data spread across dozens of pockets in the company. A lot of that work still ran on Google Sheets, and I doubt we were alone there.

AI is a golden opportunity to streamline how a business works, but only if you don’t confuse adoption with outcomes. Simply having access to the technology won’t make a company more productive on its own.

That distinction is critical. Only about 10% of S&P 500 companies had deeply integrated AI into their business processes as of 2025, while roughly the same share were using it directly to produce goods or deliver services.

The real gains come from combining proprietary data with redesigned workflows. We used AI to connect data sources that had been siloed for years and build a new approach to revenue operations. Instead of asking team members to hunt across systems and scour spreadsheets, we could bring the right information together in one place.

That work reinforced something important: the internal sprawl we were fixing probably resembled what our customers face, often with far fewer resources to solve it.

So the AI systems we build for ourselves can become a proving ground for the products we create for merchants. When a solution helps our teams make sense of scattered data, there’s a good chance it can help our customers do the same.

Ultimately, putting the customer first is what it’s all about for any business in the AI era. Almost two years into our rebuild, I know rapid technological change will bring many more surprises. By staying disciplined and focused on customer needs, we’ll be ready for whatever’s next.