As AI adoption spreads across organizations, new questions are emerging: how do we monitor and manage spend, how do we align that spend with our strategies and tactics, and how do we measure and attribute ROI to spend?
In this session, Doug Guess, Finance Operating Partner at RTP, focused on these questions and provided current guidance for leadership and finance teams.
Key Takeaways
1. AI spend needs to move from a cost line to an investment category.
For much of the past year, most companies have tracked AI spend the way they’d track a utility bill: a number to watch, without much context for why it’s moving. The more useful shift is reporting the why behind that spend, and starting to think about it as an investment rather than an expense to control.
2. Visibility starts with three dashboards: cost efficiency, ROI, and workforce leverage.
You can’t get to ROI, or plan capital allocation, until you can see and manage the underlying spend. The framework Doug shared builds from a cost efficiency view to an ROI statement that ties spend to value created on the P&L. From there it moves to ROSE+ (recurring revenue divided by the cost of employees, contractors, and AI spend), a modern read on productivity and labor leverage.
3. Small, stackable levers can meaningfully cut inference costs.
Prompt caching and model routing are the most impactful levers, and both are conversations to have directly with your technical team. Layered together with other efficiency measures, these levers can meaningfully reduce inference costs, even as usage keeps climbing. Gateways like AWS Bedrock also let you tag spend by department, so power users show up correctly on the P&L.
4. Token spend alone doesn’t tell you whether AI is creating value, ROI does.
Doug walked through a simple example: an agent drafting responses on 1,900 support tickets cost $8,400 to run against roughly $13,000 in labor value it replaced, a 58% return. Getting to that level of specificity takes agents scoped narrowly enough that you can isolate their cost and their output.
5. Tier AI investments by payback period heading into budget season.
Not every initiative should be judged on the same timeline. Some pay back in months; full workflow redesigns built around AI can take much longer but carry a larger return. Setting an inference efficiency ratio floor for new products and marking a ROSE+ baseline today gives you something concrete to measure progress against a year from now.
Quotes
“Capturing token spend isn’t really that helpful in understanding how AI is creating value in your business. You need to look at what actually hits the P&L, and what the return is on that spend.”
— Doug Guess, Finance Operating Partner
To watch our previous episodes, click below.
Episode #1 – Our AI Operating Framework
Episode #2 – Security, Governance, and Risk Management