Ask most operations leaders why their AI proposal stalled, and you’ll hear some version of “finance just didn’t get it.” Ask the CFO, and you’ll usually hear something more specific: the numbers weren’t there, or they weren’t the right numbers. Those are two very different problems, and only one of them is fixable with a better slide deck.
CFOs aren’t skeptical of AI in the abstract. A recent survey of 600 finance chiefs, commissioned by Coupa and conducted by Wakefield Research, found that increasing AI investment ranked among the top three strategic priorities for 2026, just behind strengthening supplier relationships. The appetite is there. What’s missing, in a lot of proposals, is a business case built the way finance actually evaluates spending, rather than the way a technology team is used to pitching a project.
Why “It Will Save Time” Doesn’t Clear Finance Review
“This will make the team more efficient” isn’t a financial claim. It’s a hope. CFOs need to translate efficiency into cost avoidance, headcount reallocation, or risk reduction before they can weigh it against every other request competing for the same budget.
This shift in scrutiny is already visible in how individual finance leaders are operating. Match Group CFO Steve Bailey told CFO Dive in early 2026 that he now requires “a business case with clear impacts either in the form of cost savings or efficiency gains” before approving any material AI spending, explaining that a blank check for AI makes it difficult to balance where to tighten spending against where to invest for growth. That’s not an unusually strict bar. It’s becoming the standard one.
The gap between enthusiasm and proof is measurable. A December 2025 RGP survey of 200 US CFOs found that 66% expect significant AI ROI within two years, but only 14% report meaningful value from AI today. Every proposal that lands on a CFO’s desk right now is entering that environment, whether the person presenting it realizes it or not.
The Metrics That Actually Move a Budget Conversation
Three categories of numbers carry weight in a finance review, and they’re rarely the ones featured most prominently in a typical AI pitch.
Cycle time is the first. How long does the specific process take today, start to finish, and what would the automated version realistically take? This needs to be measured against the actual current process, not an idealized version of it, because finance teams will ask where the baseline came from.
Rework and error cost is the second, and it’s often larger than people expect once someone actually tracks it. Manual data entry, approval routing, and document review all carry a hidden cost in corrections, follow-up, and delay. If that cost hasn’t been measured before, a 60 to 90 day tracking period before the proposal is finalized will produce a far more credible number than an estimate.
Headcount reallocation is the third, and it’s the one that requires the most care in how it’s presented. This isn’t about promising layoffs, and framing it that way tends to backfire both with finance and with the team that will actually use the new system. It’s about showing where existing staff time gets freed up for higher-value work, which is a real and measurable outcome finance teams can plan around.
Which Departments and Use Cases Show the Clearest ROI First
Where you start matters as much as how you build the case, and the data on this is more specific than most internal debates give it credit for. A widely cited 2025 study from MIT’s Project NANDA, “The GenAI Divide: State of AI in Business 2025,” found that roughly 95% of enterprise generative AI pilots showed no measurable effect on profit and loss, while only about 5% extracted significant value. Worth flagging directly: this was a preliminary report based on interviews with 52 organizations and survey responses from 153 leaders, not a peer-reviewed study, so the exact figure should be treated as directional rather than precise. But the pattern it points to is worth taking seriously regardless of the exact percentage: the report found that roughly half of generative AI budgets went toward sales and marketing use cases, while some of the clearest, most measurable returns actually showed up in back-office and administrative automation instead. Companies were spending where the excitement was, not necessarily where the payoff was.
That pattern lines up with separate, independently sourced research. McKinsey’s 2025 State of AI survey, based on nearly 2,000 respondents across 105 countries, found that reported revenue increases from AI use cluster most consistently in marketing and sales, strategy and corporate finance, and product and service development. Note that this measures revenue impact specifically, which is a different question than “where does a first pilot most reliably clear a measurement threshold.” The two data points together suggest a practical sequencing: administrative and back-office processes tend to be the safer, more measurable starting point for a first pilot, while customer-facing and revenue-generating use cases often show real upside but carry more variability and take longer to prove out cleanly.
With that in mind, a few departments and use cases consistently make sense as a first pilot, specifically because they combine high transaction volume, repetitive and well-defined tasks, and data that’s already structured enough to measure against:
Accounts payable and finance operations. Invoice processing, expense report review, and reconciliation tasks are high-volume, rules-based, and already tracked closely enough that a baseline (processing time, error rate, days to close) usually already exists somewhere in your finance systems.
IT service desk. Ticket triage, password resets, and access requests are repetitive, already ticketed and timestamped, and typically have existing SLA data that doubles as your pre-pilot baseline without any extra measurement work.
Customer support tier-1 resolution. Routine, high-volume inquiries that don’t require judgment calls, such as order status or account questions, are well suited to automation and usually already have call or ticket volume and resolution-time data tracked.
HR case management. Benefits questions, onboarding paperwork, and policy lookups follow a consistent pattern and are typically already logged in an HR case system, which again means a baseline is often sitting there unused.
Procurement and vendor intake. Purchase requisition routing and vendor onboarding documentation are structured, recurring processes where automation reduces cycle time in a way that’s easy to isolate and measure.
The common thread across all five is that a baseline already exists or is easy to establish quickly, the volume is high enough that improvements show up in weeks rather than months, and the risk of a wrong decision by the system is low and easily caught by a reviewer, which keeps the pilot’s governance requirements manageable.
How to Structure Your First AI Pilot for Measurable ROI
A pilot only proves what it’s designed to measure, so the structure matters more than the technology choice underneath it. Four practices consistently separate pilots that produce a usable answer from ones that produce an ambiguous shrug six months later.
1. Scope it to one process, not a department or a function. “Improve customer service with AI” isn’t a pilot, it’s a mission statement, and it will be nearly impossible to attribute a specific dollar figure to it later. “Automate tier-1 order-status inquiries currently handled by the support team” is a pilot, because it has a clear boundary around what’s being measured.
2. Set the success threshold before the pilot starts, not after you see the results. Decide, in writing, what “worked” means: a specific reduction in handling time, a specific error-rate improvement, a specific cost-per-transaction target. If the pilot doesn’t clear that threshold, the honest answer is to not scale it yet, not to quietly redefine success after the fact to justify the investment already made.
3. Timebox the measurement window to 60 to 90 days for most high-volume processes. This is long enough to see a real pattern in the data and short enough to keep the pilot from becoming a permanent, unmeasured fixture that nobody revisits.
4. Report the result either way. A pilot that doesn’t clear its threshold isn’t a failure if it produces a clear, honest answer about why, whether that’s a data quality issue, a scoping problem, or a process that genuinely wasn’t a good fit for automation. That information is what makes the next pilot more likely to succeed, and it’s also exactly the kind of disciplined, evidence-based approach that makes a CFO more comfortable approving the next request.
Building a Phased Business Case Instead of a Big-Bang Ask
A single large request asking for full-scale investment up front is harder to approve than it needs to be, because it asks a CFO to accept all the risk before any of the value has been demonstrated. A phased structure changes that dynamic. Propose a pilot scoped to one process or one department, with a defined measurement window and a specific dollar or hour figure attached to success. Only after that pilot produces real numbers does the proposal ask for the larger investment to scale it.
This isn’t just a negotiating tactic. It reflects how the money actually gets spent well. Research on CFO AI investment consistently points to a wide gap between what a pilot costs and what scaling it to full production requires, since scaling brings in infrastructure, integration, change management, and ongoing operational costs that a small pilot doesn’t need. Budgeting for the pilot without planning for that gap is one of the more common reasons a promising AI project stalls right after its first success.
Presenting a conservative scenario alongside the expected one also builds credibility. If the AI initiative still produces a positive return assuming implementation takes longer than planned and benefits land below the optimistic case, that’s a much easier “yes” than a single rosy projection.
What to Include in the Proposal Itself
A finance-ready AI business case should be able to answer these questions directly, in this order:
- What specific process or decision does this affect?
- What does it cost today, in hours, dollars, or both, based on measured data rather than estimate?
- What’s the pilot scope, timeline, and success threshold?
- What does scaling cost if the pilot succeeds?
- And who owns the initiative, both during the build and after it’s live?
If your team can’t answer all five with real numbers yet, that’s a sign the project needs a short measurement and scoping phase before it goes in front of finance, not a sign the project isn’t worth pursuing. Getting that scoping right the first time is usually what separates a proposal that gets funded from one that gets sent back for more detail.
Ready to Build an AI Business Case That Holds Up?
Optimum’s AI consulting engagements typically start with exactly this kind of scoping and readiness work, building the baseline data and financial case alongside the technical plan so the proposal that reaches your CFO is one built to hold up under real scrutiny. If you’re preparing an AI investment case for your next budget cycle, reach out for a complimentary discovery session and we’ll help you build the numbers before you build the pitch.
About Optimum
Optimum is a nationally recognized IT consulting firm and official partner of Microsoft, Make.ai, ServiceNow, and Outsystems, dedicated to helping organizations connect workflow automation, AI, and business intelligence into one cohesive, high-performing operation.
We focus on driving efficiency, reducing operational costs, and supporting digital transformation through an assessment-led, partnership-driven approach. Our expertise spans AI agent design, workflow automation, data and analytics, and enterprise platform implementation. We help organizations automate work and ensure that work is grounded in clean data and surfaces in the reporting environments leadership actually uses to make decisions.
Reach out today for a complimentary discovery session to explore how Optimum can help you build a connected automation, AI, and analytics strategy that delivers measurable results across your organization.
Contact us: info@optimumcs.com | 713.505.0300 | www.optimumcs.com
Frequently Asked Questions About Building an AI Business Case
What does a CFO actually look for in an AI investment proposal? CFOs generally want a business case with clear impacts stated in terms of cost savings or efficiency gains, backed by measured baseline data rather than estimates. This means naming the specific process affected, its current cost in hours or dollars, a defined pilot scope with a success threshold, and a clear owner for the initiative both during and after the build.
Why do AI proposals get deferred instead of rejected outright? Most deferrals happen because the financial case wasn’t built the way finance evaluates spending, not because the underlying idea lacks merit. A December 2025 survey of 200 CFOs found that 66% expect significant AI ROI within two years, while only 14% report meaningful value today, which means proposals now enter a more skeptical review environment than they would have a year or two earlier.
Should an AI business case include a full rollout or start with a pilot? A phased approach, starting with a pilot scoped to one process or department with a defined measurement window, is generally easier to get approved than a single large request for full-scale investment. It also produces real data that strengthens the case for scaling, rather than asking a CFO to accept all the risk before any value has been demonstrated.
How should headcount impact be framed in an AI business case? Framing headcount impact as freeing up existing staff time for higher-value work, rather than as a reduction in headcount, tends to land better with both finance and the team that will use the new system. CFOs are typically evaluating financial impact broadly, and reallocated time is a legitimate, measurable component of that even when it isn’t a direct cost reduction.
Which department should run the first AI pilot? Departments with high-volume, repetitive, well-defined processes and an existing baseline of tracked data tend to make the safest first pilot, which usually points to finance operations, IT service desk, customer support tier-1 resolution, HR case management, or procurement. These areas typically already track cycle time, volume, and error rates in an existing system, which removes the extra step of building a measurement baseline from scratch.
Should the first AI pilot focus on cost savings or revenue growth? For a first pilot specifically, cost savings and cycle-time reduction in back-office processes tend to be easier to measure cleanly and prove out faster than revenue-generating use cases. A widely cited 2025 MIT study found that while roughly half of enterprise generative AI budgets went toward sales and marketing, some of the clearest, most measurable returns actually showed up in back-office and administrative automation instead, suggesting that revenue-focused use cases, while often valuable, are better pursued after a pilot has established credibility with a clean initial win.
