The One Metric Acquirers Scrutinize: AI-Adjusted COGS

Introduction

Katrina Wong, Chief Marketing Officer at New Relic, has spent 20+ years leading marketing and go-to-market at enterprise technology companies including Twilio Segment, Hired, Zuora, Salesforce, and SAP. Asked which metric acquirers scrutinize that most marketers completely overlook, her answer wasn't pipeline velocity or CAC payback. It was the cost of AI itself — and how that spend is quietly rewriting the cost structure of every modern marketing organization.

AI Is Now a Line Item, Not an Experiment

Wong's position is that AI has crossed from initiative to infrastructure. "It's the cost of doing business now that AI is just part of doing business for those of us in tech and kind of an early adopter," she said. The metric that matters is understanding how much you're actually spending on AI to stay competitive — not as an i ovation budget, but as a permanent operating cost.

Her framing is blunt: "It's COGS with AI factored in. I really think that's a new puzzle." For marketing leaders, that reframe has teeth. Token consumption, agent infrastructure, and the engineering time to maintain both don't live in a pilot budget anymore. They live in the margin structure that a buyer, board, or CFO will eventually pull apart line by line.

The Real Tradeoff: Vendor Spend vs. Build Time

The problem gets concrete fast. A five-year relationship with a transcription vendor at roughly $200 a month, spread across seven or eight podcasts, is exactly the kind of expense that looks replaceable the moment agents can do the same job. The question is whether to spend half a day rebuilding it to lower cost of goods sold.

But the savings aren't clean. Rebuilding means investing the time to build it, then maintaining it, and then absorbing rising AI expense and token usage on the other side. You can lower COGS on paper while trading a fixed, predictable vendor cost for a variable one you now own operationally. That time-and-value exchange is the actual calculation — and most marketers are only ru ing half of it.

How New Relic Approaches Buy vs. Build

Wong says this is an active internal conversation, not a settled policy. Her team debates which agents to build versus which problems to wait on, expecting the market to solve them. That distinction — build now or wait for someone else to ship it — is the operating framework worth stealing.

Where They're Waiting

"We're not building agents to run campaigns end to end," Wong said. Her team believes someone is going to solve that category, so they're waiting rather than committing engineering resources to a problem the market is actively racing toward. It's a deliberate decision not to build, based on where vendor competition is thickest.

Where They're Building

The build energy goes somewhere more specific: the handoffs between different sub-functions within marketing. Wong's team is streamlining those transitions and doing agent orchestration around them. That's the seam no vendor understands better than you do — the internal workflow friction unique to how your organization is structured.

What This Means for Marketing Leaders

The pattern is clear enough to apply immediately. Wait on horizontal capabilities where a well-funded vendor category is forming. Build on the co ective tissue between your own functions, where the problem is specific to your org and no one else is going to solve it for you.

Conclusion

Three takeaways carry the episode. First, AI spend is now a COGS component, and the marketers who can articulate that number have an answer to a question acquirers are already asking. Second, buy-versus-build isn't a one-time decision — it's an ongoing conversation about which problems the market will solve for you and which ones you own. Third, cost savings from rebuilding a vendor relationship in-house are never free; maintenance and token usage travel with them. As Wong put it, this is a new puzzle — and admitting you don't have the full answer yet is more honest, and more useful, than pretending the math is simple.