Is your dev team's AI spend buried in the wrong part of your P&L — quietly distorting your gross margin and making your AI ROI impossible to measure?

In episode #387, Ben Murray tackles one of the most pressing accounting questions hitting SaaS CFOs and controllers right now: how to properly classify AI development costs across your chart of accounts. With dev teams racking up daily charges from multiple AI providers, getting these expense codes wrong doesn't just create accounting noise — it misrepresents your gross margin, obscures your true AI economics, and leaves you unable to answer board questions about AI ROI. The right GL structure makes all the difference.

• The single most important question to answer first: is the AI cost production-facing (customer-serving inference) or internal engineering use — because the answer determines whether it hits COGS or OpEx, and your gross margin depends on getting this right.
• A full breakdown of the five AI cost categories SaaS companies are facing today — production inference, dev/testing inference, model training and fine-tuning, GPU compute for production serving, and vector database infrastructure — and exactly where each one belongs on the P&L.
• Why model training and fine-tuning costs may need to be capitalized rather than expensed, just like other qualifying software development costs under existing accounting guidance.
• Why internal AI tool spend (Copilot, ChatGPT, Claude, etc.) needs its own dedicated GL account — separate from internal-use software — so you can actually measure efficiency gains and answer the AI ROI question when the board asks.
• Why this isn't optional housekeeping: as AI costs grow, the difference between properly coded and miscoded AI expenses will show up directly in your gross margin and in how investors read your unit economics.

Tune in to get the GL coding framework SaaS CFOs are building now — before AI expenses get too big to untangle.

Resources Mentioned

• Ben's AI COGS blog post (detailed GL coding guidance): https://www.thesaascfo.com/what-should-be-included-in-ai-cogs/