Building an AI Business Case That Actually Gets Funded: A Practitioner’s Guide

Nancy Wang
12 minutes ago
8 min read

*A fundable AI business case makes four things legible to leadership at once: impact, return, real cost and risk. Most initiatives stall because one of the four stays vague.*

Every AI initiative arrives at the same moment. The exploration is done: the workflow studied, the data evaluated, the vendor landscape reviewed, the fit tested. The initiative seems worth pursuing.

Then someone in the room, often you, says: "Great. Let's build the business case."

That is where too many teams quietly stall. Not because their idea isn't strong, but because the business case isn't actionable. More than most teams realize, building the AI business case is a translation exercise, one that converts conviction into the language of risk and return that leadership can commit resources to fund.

This post is about doing that translation well.

A 2025 MIT study of 300 enterprise AI deployments put a number on the problem: 95% delivered no measurable impact on the profit and loss statement. The failure was rarely in the models themselves. It was in the argument around them: assumptions that didn't hold, value that didn't scale, integration work that wasn't priced in.

The teams that beat those odds tend to share one habit. They treat the business case as seriously as the build.

What a Strong AI Business Case Does

The AI business case has one job: give decision-makers what they need to say yes, and sometimes no, with confidence. In our work with clients, we've seen four things that a strong case, one built to get the nod, makes clear.

The impact is concrete, quantified and rarely comes from just one place. Revenue, cost, risk reduction, strategic alignment: the initiative may touch several of these, and the case has to name which ones and by how much.

The return is defensible. That means honest timing, honest assumptions and a counterfactual that holds up.

The cost is honest. Not pilot economics extended out three years. The real thing, at the scale the initiative is actually being funded to reach.

The risk is legible and bounded. Not 12 unranked bullet points. The one that changes the outcome, and how the organization will know if it's materializing.

Different stakeholders care about different parts of story. Revenue leaders want the top-line impact. Operations leaders want workflow and capacity change. Finance wants the numbers to hold together. Executives want strategic fit and competitive positioning.

A strong AI business case speaks to all of them without hiding anything.

Where the Business Value Comes From

Everything downstream in the case depends on how the impact is defined. In an AI initiative, impact rarely comes from a single efficiency gain or a single new opportunity. It shows up across several areas, and each one is a source of AI business value the case will ultimately quantify in dollars.

In our practice, we've identified seven areas of impact worth considering. The primary one leads. The others reinforce.

Revenue impact. New revenue, faster revenue, retained revenue, revenue per rep, conversion lift.
Cost impact. Reduced cost per transaction, reduced cycle time, capacity freed for higher-value work.
Risk impact. Reduced error rate, reduced compliance exposure, reduced churn.
Capability impact. New products, services or markets enabled that weren't previously accessible.
Strategic alignment. How strongly the initiative advances the organization's stated priorities. This is often why the initiative was flagged in the first place, and it belongs in the case explicitly.
External stakeholder benefit. Value delivered to customers, partners or audiences, not just internal operations.
Competitive advantage. The differentiation this creates, and how defensible it is.

Pick the primary impact, the one that most directly reflects why the initiative matters, and quantify it fully. The others do the reinforcement work. Strategic alignment shows leadership why this deserves capacity.

Competitive advantage shows why now. External stakeholder benefit shows why the organization should invest beyond pure operational return.

Then convert the primary impact into dollars, showing the math. If the impact is "each rep can handle 20% more accounts," walk from that operational metric to revenue with the assumptions visible. If the impact is "50% reduction in error resolution time," walk to cost savings, with the labor rate, the volume and the counterfactual called out.

One discipline separates rigorous cases from optimistic ones: pressure-test the impact claim with the people who own the metric. If the revenue team hasn't signed off on the revenue assumption, it isn't a business case yet. It's a proposal.

Modeling Return: Time Meets Confidence

Return is what the impact is worth minus what it costs, over time, adjusted for confidence. Five components deserve real attention.

The counterfactual is what separates the rigorous cases from the optimistic ones. What would happen without this investment? If the work would still get done another way, the return is the delta, not the total. If competitors are already moving on similar capability, the counterfactual isn't "status quo," it's "falling behind." Name the counterfactual explicitly, so the delta is real and not assumed.
Time to first value matters more than most business cases show. When does the initiative start producing measurable benefit? Not full ramp. The first credible read that the initiative is working. Two to four months is fast. Six to 12 months is normal. Beyond 12 months, the case needs to explain why. Initiatives that promise value only after a long integration phase are harder to fund, and honesty about first-value timing builds trust rather than eroding it.
Time to steady state is when the initiative reaches full ramp. The gap between first value and steady state is often where business cases quietly overstate, showing full-ramp value from month one instead of modeling the actual adoption curve. If the case assumes the workflow, the training and the change management all land on day one, the numbers are already off.
Scalability of impact is the multiplier most cases don't name. Does the impact grow as the initiative expands across teams, workflows or business units? Initiatives that scale, where one build serves many uses, carry a fundamentally different return profile than initiatives that serve a single workflow. If scalability is real, name it and quantify it. If it isn't, be clear that the return is bound to the initial scope.
The confidence range is what shows a leadership team the case is honest. A single-point estimate hides what the team actually knows. Low, base and high scenarios with visible assumptions are more credible than false precision. Show the range. Show what makes the low case low.

Timing is usually where the model and reality part ways. An operations team we advised had built a case for an AI-assisted quality inspection workflow with a first-value estimate of three months and full-ramp value modeled from month four. The build shipped on time. But the inspectors who used it needed six weeks of side-by-side runs before they trusted the system's flags, and the workflow redesign took another quarter. Real first value landed at month five. Full ramp came at month nine.

The initiative eventually cleared its target, but the case had over-promised on year-one return by 40%, and that gap became the thing leadership remembered.

That's why it's critical to frame the return over the horizon that matches the initiative. A quick-return AI use case, like a customer service assistant or a content generation workflow, usually makes sense on a 12 to 18 month view. A capability build with strategic reach, like an agentic workflow platform or a custom model built on proprietary data, often warrants two to three years, long enough to include the second capability cycle.

The horizon should serve the case, not signal false precision.

The Real Cost: Seven Items That Get Missed

The cost side is where many business cases quietly under-deliver, not through dishonesty but through optimism. AI initiatives have more cost drivers than most teams initially model, and naming them explicitly is what makes the case credible.

The organizations capturing real value from AI are the ones redesigning workflows end-to-end, not the ones layering AI on top of existing processes, per BCG's 2025 AI at Work research. Redesign isn't free, and the cost model has to reflect what real deployment actually takes.

Direct technology costs at production scale, not pilot pricing. Model spend, licensing, infrastructure. The pilot subsidy is not the run-rate.
Data readiness costs. Cleanup, integration, joining, structuring, enrichment. If the data isn't ready today, someone has to make it ready, and that work has a real cost.
Integration and maintenance. Ongoing engineering to keep the system connected to the rest of the stack.
Operational readiness. Workflow documentation, role definition, monitoring, feedback loops. AI initiatives that go into production without operational scaffolding become someone's second job, which is where value erodes.
Human oversight and governance. The people who monitor, correct and improve the system, plus the review processes that govern its use.
Change management. Training, workflow redesign, adoption support.
Second-cycle costs. The model upgrade, the vendor renegotiation, the capability refresh that lands 18 to 24 months in. AI moves too fast for the business case to end at month 12.

A B2B SaaS company we worked with built a strong pilot on a customer support AI agent. Pilot economics looked excellent: around $0.04 per resolved ticket, well under human cost. At scale, the picture changed. Model spend held. But four things doubled the per-ticket cost: prompt maintenance, human oversight for edge cases, monitoring infrastructure to catch drift and quarterly re-training as products evolved.

Still profitable, but the year-two budget conversation was much harder than the year-one one, because the initial case had shown a number that didn't include half of what the system actually needed.

Here is the test we use with clients: if a finance team member built this cost model from scratch, would it look the same? If not, the case will get rebuilt in front of you rather than beside you. Better to build it that way from the start.

The other test: are pilot-era assumptions still in the model? Subsidized API rates, sweat equity from the build team, vendor discounts that expire: all of these disappear by year two. Gartner puts a number on where this leads.

By the end of 2025, it found, at least half of generative AI projects were abandoned after the proof-of-concept stage, with escalating cost among the named reasons. The mechanism is the one this section has been describing. A pilot that pencils out on subsidized economics turns uneconomic at production volume, and a case built on the pilot number gets defunded when the real number lands.

This is often where the adoption-is-not-ROI confusion starts, because pilot economics quietly get read as run-rate.

Framing Risk as Governance, Not Disclosure

Risk is where business cases most often overreach ("this is low risk") or under-address ("here are 12 risks, unranked"). A strong AI business case does three things instead.

It names the single biggest risk, the one that would most change the outcome, and explains how it's bounded. Not the full risk register. The one that matters. If a leadership team can't identify what would most likely blow up the case, they haven't finished thinking about it.
It distinguishes reversible from irreversible risk. Reversible errors, where a mistake can be caught and corrected, carry a fundamentally different profile than irreversible ones like data exposure, regulatory violation or customer harm. Naming which category the initiative sits in changes what governance the case commits to. Reversible risks tolerate more experimentation. Irreversible risks demand more scaffolding before deployment.
It names the decisions that would be revisited if key assumptions don't hold. What would trigger a checkpoint? Which numbers, if they drift, would prompt a re-scope or a wind-down? This reframes risk from a defensive disclosure into a governance mechanism. The case isn't an all-or-nothing commitment. It's a plan with built-in review points.

That framing changes the leadership conversation. A case that says "here's what we don't know, here's what would tell us we're wrong, and here's when we'd revisit" is far easier to fund than one that claims certainty it can't have.

Where the AI Business Case Sits

The AI business case is where a team's conviction becomes a decision the organization can commit to. Done well, it isn't a document that sells an idea. It clarifies the idea. It exposes what the team actually believes about impact, timing and risk. It makes the assumptions visible. It separates the initiatives that move from interesting to funded from the ones that stay in pilot purgatory.

The teams whose initiatives get funded are the ones that treat the case-building work as seriously as the technology work. From there, the work is in the build, which is where FORGE, our framework for achieving real business impact with agentic AI, picks up.

But the funding conversation comes first, and this is what wins it.