How to Measure AI ROI: 5 Frameworks from Zapier and Nium CFOs

Measuring AI ROI requires five proven frameworks from Zapier (97% AI adoption) and Nium CFOs: integrate AI investments into existing OKR maps accelerating goal timelines rather than creating separate metrics, implement granular telemetry segmenting spend by team, user, and activity type for baseline establishment, double-click beyond "time saved" to downstream business outcomes like faster upsells or shorter close cycles, compare team output against headcount over time where rising output with flat staffing signals success, and combine AI and headcount into unified budget envelopes letting leaders optimize the most efficient path to outcomes.

Everyone who's used AI knows it's creating value. We feel it. But the feeling is anecdotal, and at some point, finance leaders need something they can actually measure.

I sat down with two CFOs who have built real frameworks for this challenge: Ryan Roccon from Zapier (the AI orchestration platform with 97% AI adoption across their organization) and Andre Mancl from Nium (a global payments infrastructure company moving millions of transactions in real time). The goal was simple: to move beyond the excitement and identify the proven methods for assessing the return on investment for AI initiatives.

While the consensus was there is still a lot of work to be done around ROI, we came up with five frameworks that can help get you started solving some of the biggest ROI challenges: 

1. The OKR Integration Framework: Don't create separate AI metrics—tie AI investments to accelerating your existing goals and timelines.
2. The Telemetry Framework: Segment AI spend by team, user, problem, and activity type. Granularity is the foundation for any meaningful measurement.
3. The Double-Click Framework: When someone says "time saved," keep asking "so what?" until you hit a business outcome you can actually measure.
4. The Team Output Framework: Compare team output against team size over time. If output rises while headcount stays flat or shrinks, AI is working.
5. The Unified Budget Framework: Combine AI and headcount into one envelope for outcomes. Let leaders decide the most efficient path to their goals.

Why AI ROI is So Hard to Measure?

Before diving into measurement, I wanted to understand how these companies approach AI in the first place. The spectrum ranges from highly cautious (worried about over-exposure to risk) to full steam ahead (testing everything).

Zapier's Approach to AI ROI: Top-Down Mandates

Ryan didn't mince words about Zapier's approach: "AI at Zapier is a top-down mandate. It's an expectation for every single person in every role. We measure AI fluency in our interview process for folks joining Zapier."

The result is 97% AI adoption across the organization. Every function, every level.

But that comes with budget implications. "We lead with this concept of access and experimentation, not constraint," Ryan explained. "We look at this as a once-in-a-lifetime shift, and we don't want to be penny wise and pound foolish."

Nium's AI ROI Strategy: Guardrails with Experimentation

Andre described Nium's approach as "let it rip with guardrails." They have approval processes for AI tools and constantly monitor usage, particularly important given they're in the payments business where data privacy is paramount.

"We've heard case studies of dramatic overspending where you've got a thousand seats and fifty people using them," Andre said. "We're keeping a careful eye on that, but we're not going to let caution slow us down."

I asked Ryan about employee reaction to the mandates. Zero resistance, he said. "I've got L2 tax folks writing Python scripts using AI to automate processes. I've got engineers mocking end-to-end product demos in minutes. It's truly transformed almost every function."

Framework #1: Tie AI to Your Existing OKR Map

Here's where both CFOs got down to the brass tax in how they’re implementing things. Andre's framework is elegantly simple: don't create a separate measurement system for AI, plug it into what you're already tracking.

"We run the company off an OKR map where we look at all the different functions and tie them to broader strategic goals," Andre explained. "A lot of that is time-based: reduce friction, increase operational efficiency, eliminate manual processes."

Instead of inventing new AI metrics, Nium expects AI investments to accelerate timelines on existing goals.

"If we're going to spend $100,000 or a million dollars on AI tools, we expect those timeframes to be reduced," Andre said. "Time to delivery, time to close, accuracy. It's about improving operational efficiency, not reduction in spend."

This makes so much sense. Because, you're not arguing about whether AI "works." You're asking whether goals are being hit faster than before.

Framework #2: Granular Telemetry and Cost Segmentation

Ryan's framework starts with visibility. Because Zapier uses its own product internally, they can see exactly where AI-powered workflows are running across the organization.

"We segment by teams, by users, by problems they're solving, all the way through to the activity type being generated," he explained. "With engineers, we look at the code writing process, the QA process, the pushing process. We diagnose each stage to understand what was possible before."

That segmentation creates baselines, and baselines are the foundation for any ROI framework. You can't know what "great" looks like if you don't know where you started.

But Ryan was quick to acknowledge the challenge. Even with that visibility, measurement is hard. "Andre's right that it's a difficult thing to measure," he said. "And ultimately, what you hear a lot of folks talking about is time saved."

Which brings us to the next framework because "time saved" is where most AI ROI conversations go wrong.

Framework #3: Moving Past "Time Saved" to Business Outcomes

Both CFOs agreed that "time saved" is the most common—and most misleading—way people talk about AI value.

The framework that actually works? Keep clicking until you find the business outcome.

Andre shared a memorable stat from a conference: "One of the presenters said the biggest beneficiary of AI tools so far is everyone's dogs. Because the work-from-home crowd has more time to walk their dogs."

This goes directly to the value capture gap I’ve been talking about for a while, which is the gap between employees and employers.

AI has two potential beneficiaries:

• There's the employee who might use AI to do their same work in half the time—and spend the other half on Netflix or walking the dog.
• And there's the organization, which could capture that productivity as more output, faster timelines, or reduced headcount.

Now, we’re not saying that there isn’t innate value in better quality of life for employees. But, we do all agree, that if that is the benefit, it must still be tied to a business outcome. 

Ryan acknowledged the challenge: "It's very easy to see if engineers are writing more code per engineering hour. But the quality could be poor, or you need more review. You really have to get granular and do pre-post and A/B tests."

But he pushed back on the framing that it's zero-sum.

"When employees get to automate the work they don't want to do, the mundane, repetitive, low-value-add stuff, they naturally focus on the opposite. And the company benefits from them focusing on the most valuable work."

Ryan then shared a constructive reframe that's become central to Zapier's measurement approach: when you hear "time saved," keep asking "so what?" until you hit a business outcome.

For instance, their accounting team supports a high-velocity sales motion where customers mid-renewal want to upgrade, requiring credit calculations. Historically, that was manual, going back and forth, looking at subscription dates, running numbers. The process took hours.

"We automated the whole thing with Zapier AI steps," Ryan explained. "Now you come into Slack, ask 'What's the credit calculation for this customer?' and get an instantaneous answer."

The surface metric is time saved. But Ryan doesn't care about that:

"Now we're giving customers an answer immediately, which means our probability of upselling them is incredibly higher. The outcome is that we're upselling folks faster and more efficiently. Time saved, yes, but there's a downstream metric that's even more impactful."

That's what we all need to focus on. Find the business outcome that "time saved" enables, and measure that instead.

Framework #4: Team Output vs. Team Size

Ryan shared a macro-level framework that cuts through the complexity: compare team output against team size over time.

"One of the more macro signals I pay attention to is team output versus team size," he said.

Zapier's accounting team is the perfect case study. They've automated almost every process around month-end close. Over the last three to four years, complexity has exploded—multiple entities, new products, international expansion, Big Four audits. Yet the team has actually gotten smaller, and month-end close has gotten shorter.

"You might not pinpoint every exact efficiency across every stage," Ryan said. "But you can zoom out and look at how the team is performing as a whole."

This framework acknowledges that granular measurement is hard while still providing a clear signal: if output is rising and headcount isn't, AI is working.

Framework #5: The Unified Budget Approach

Costs and budgets were also a huge part of the conversation. We talked about whether AI spend has been surprising, and how budgets have held up against the reality of rapidly changing models and pricing.

Ryan's candid answer: "It's changed, and it's become a much larger line item. It certainly took us by surprise in 2025, but was something we had to revisit pretty often. I feel a lot better in 2026."

For example, they switched their AI copilot from one model to another and cut costs by 75%. Great outcome, but not something he'd planned for in the budget.

"Pretty quickly, it gets eaten by something else we find as we're finding a new model to power something even more impactful."

Both CFOs are moving toward a framework that's reshaping how they think about AI investment entirely: combining AI and headcount into one budget envelope.

Andre's evolution has been to stop separating "AI tools" from "software tools.” It's all just tools. Leaders get a budget for their goals and can mix and match headcount and technology however they want.

"You want to hire 25 people to solve a problem, or try to accelerate through AI? It's your call. All we care about is the ultimate results. Speed and delivery."

Ryan confirmed Zapier is moving the same direction. "We do build, buy, borrow, or automate when we evaluate headcount. We've clearly asked folks looking for headcount budgets to answer the question: could you be automating this or using an AI tool instead?"

When they do annual planning, those budgets are starting to merge. "Here's your big bucket. If there's a need to transfer from one to the other, do it."

This framework changes the question from "how much should we spend on AI?" to "what's the most efficient path to our outcomes?"

AI ROI in Practice: Build vs. Buy Decisions

I asked both CFOs about how AI investments are evolving, whether companies are building on foundational models, buying purpose-built solutions, or dealing with legacy SaaS vendors suddenly claiming to be AI companies.

Andre shared a story that captured the "build" mentality perfectly. Their CEO locked the executive team in a room, made everyone subscribe to Replit, gave them dummy transactional data with fraud hidden in it, and said: "Come up with an app that detects the fraud. We're not leaving until each of you builds one."

"Some of us did better than others, I struggled a bit," Andre admitted. "But it set the posture: you can do this. It's possible to build apps very quickly that solve fairly straightforward problems in real time."

That posture led to their AI hackathon, which generated 138 submissions from 900 employees. "We're not going to come up with all the use cases from the top," Andre said. "You have to find the friction in your individual day-to-day and figure out where AI can fix it."

One executive built a negative news tracker for their partner network—hundreds of banks globally—that sends alerts within minutes when something disparaging appears. It took fifteen minutes to build. That's the kind of thing you'd normally submit a headcount request for.

Ryan shared Zapier's approach: their procurement process literally asks "Can you build this on Zapier?" at the top. If the answer is no, they want specifics on what gaps are preventing it so they can add those to their roadmap.

"We're trying to be really clear: let's buy the things that solve a unique problem we can't currently solve with Zapier. But it's not a long list."

Proven AI ROI Success Stories from Zapier and Nium

I asked both CFOs for concrete examples of these frameworks in action.

Ryan's standout success: a customer brief generator that pulls from dozens of sources—Gong calls, HubSpot, support interactions, public data, usage telemetry—and synthesizes everything you'd want to know before a customer conversation.

"If you ask that question manually, you've got six data scientists pulling stuff from all over the place," he said. "Now account executives and CSMs come to conversations incredibly armed with literally every piece of information they could want."

Using the double-click framework: Time saved? Yes. But the actual outcome? "You close a lot more deals."

Andre's finance team success: AI-powered reconciliation for their payments platform. They move millions of transactions and need to reconcile bank accounts with their API and NetSuite for financial close. The systems handle most of it, but the last mile—maybe 1-2% of transactions—required manual matching because of slight differences in how entities are named.

"It was a very manual process, lots of Excel work. They incorporated AI the other day and it was boom, done, instantly."

The full session recording is available on demand. For more insights on AI-powered procurement and spend management, signup for our newsletter The Spend Table or reach out to learn more about Tropic. 

Frequently Asked Questions About Measuring AI ROI

Why is AI ROI so hard to measure?

Most AI value shows up as time saved, which doesn't automatically translate to business outcomes. Time saved can be captured by employees (more leisure) or by organizations (more output), and without granular telemetry, it's impossible to know which is happening.

What's wrong with "time saved" as an AI metric?

It's one step removed from what actually matters. When someone saves time, double-click to find the downstream business outcome: faster customer responses leading to higher upsell rates, shorter close cycles, fewer errors. Measure that instead.

How should companies approach AI budget planning?

Segment your AI spend between internal use cases (be cost-conscious) and customer-facing use cases (be willing to invest ahead). Expect frequent budget adjustments as models change rapidly—what costs X today may cost 75% less tomorrow with a different model.

Should AI and headcount budgets be combined?

Forward-thinking companies are moving in this direction. Give leaders one envelope for outcomes and let them allocate between people and technology. The question becomes: could you automate this instead of hiring for it?

What separates companies that capture AI value from those that don't?

Having explicit frameworks rather than hoping value materializes. The five frameworks from Zapier and Nium—OKR integration, granular telemetry, the double-click method, team output measurement, and unified budgeting—provide practical starting points any finance team can adopt.