Returning As CEO During the AI Rush Taught Me How To Spend Wisely

After stepping back into the top job in early 2024, I faced a dual challenge: harnessing the surge in generative AI while keeping prices and promises steady for the customers who rely on our platform. Our company was nearing $1 billion in sales but carried modest adjusted earnings, so every dollar funneled into new tools had to either protect or grow the cash we planned to return to investors. We set a concrete aim: deliver $100 million in free cash flow inside three years. That goal reshaped how we evaluated AI projects and how we communicated changes to the merchants who use our products.

First lesson: prioritize the balance sheet. Unlike the handful of tech giants that can absorb expensive experiments, most firms must make every AI bet justify itself financially. We introduced a rule: any AI initiative had to demonstrate how it would defend or increase the free cash flow target before winning money. That meant saying no to bloated, long-term platform deals that could lock us into millions in recurring costs. Instead, we favored tools that let us swap the underlying language models as costs and performance changed, and we reserved the heaviest compute for problems that truly needed it. When a major mobility company reported burning through its AI tool budget in a single quarter, it reminded us how fast unvetted spending can erase hard-won gains.

Second lesson: move with purpose, not panic. The market signaled that every software vendor needed an AI story, but rushing to bolt consumption-based pricing or feature sets onto customers rarely helped. We paused before following trends and asked two simple questions: does this advance the direction we already chose for the business? And can we show measurable value to users? Urgency matters, but it must be disciplined. For example, roughly one out of every three managers who cut roles because of AI later rehired similar positions — a cautionary tale about acting on short-term signals without a plan for measurement and reversal.

Third lesson: design AI for the people who pay the bills. Our internal data problems—fragmented sources, messy labels and slow queries—look a lot like what many of our merchants face, only they have fewer engineers to fix them. So we treated internal pilots as customer pilots: solve our own pain points first, then productize the work for users. A small online seller we work with described how a targeted automation reduced time spent reconciling inventory by days each month; that kind of outcome is what convinces customers to adopt AI features. In short, build for practical gains like time saved, errors cut or revenue uplift, not for novelty.

Fourth lesson: use internal projects as a proving ground and set tight funding gates. Every experiment had clear entry and exit criteria — predicted ROI, expected timeline and fallback plans if results lagged. We leaned on open, interchangeable tooling to avoid vendor lock-in and to reduce marginal costs as models evolved. That approach let us try new techniques without mortgaging future margins. And when something worked, we standardized it quickly so customers benefitted; when it didn’t, we shuttered it and moved on.

These four rules — protect cash, align AI with strategy, prioritize customer value and run experiments with strict guardrails — didn’t make us immune to missteps. But they did keep our spending honest and kept the focus on creating useful products rather than chasing buzz. As CEO, I found that restraint, not abandon, is often the right partner to innovation.