.svg)
.jpg)
AI adoption has tripled in just a few years: the share of organizations using AI in at least one business function has climbed from 33% to 78%. Yet fewer than one in five report a meaningful lift in earnings because of it. Deployment is up and to the right; AI ROI has barely moved. If that gap sounds familiar, it should: every technology wave, from electricity to the cloud, opened with the same lag between excitement and payback. The good news for finance leaders is that the problem usually isn’t the technology. It’s how the business case is built.
Building thatcase well is a CFO problem, and a solvable one. Here’s how to think about thereturn on an AI investment the way the companies actually capturing it do.
The first mistake is framing the question as “should we use AI?” That’s rarely the decision that matters. AI isn’t a strategy on its own; it’s a means to improve a specific process or deliver a specific outcome. The organizations getting this right don’t build one sweeping business case for AI. They build specific business cases for specific processes, each with its own metric, its own economics, and its own definition of success.
Those cases tend to fall into two categories, and the distinction matters more than it looks. One is rolling out an AI assistant, a tool like a coding copilot that helps people do their existing jobs faster. The other is transforming a process, redesigning how the work itself gets done. They carry very different economics, and confusing the two is one of the fastest ways to end up with a business case that never pays back. The mindset that ties them together: treat AI as a value initiative, not a cost-cutting one.
The most commonplace business cases fall down is the very first question: what are we actually trying to achieve? Adoption is not the only answer. A strong success metric has three parts:
• Business value. Why you’re doing this at all: top-line revenue, customer acquisition, customer retention. Start broad, then go a level deeper. “Grow revenue” is too vague to tie to one process; “improve lead-to-contact conversion” or “increase weighted pipeline creation” is a leading indicator of revenue that a specific process can actually move.
• Quality. Value without quality is a trap. In software, you can ship more features faster, but if quality slips, the gains reverse downstream. Whatever the process, pair every value metric with a quality metric that keeps it honest.
• Risk. What does the worst case look like? Scale cuts both ways. An AI agent that drifts in a call center isn’t one employee making an occasional mistake, it’s every call going wrong at once. Understanding and monitoring that downside is part of the case, not an after thought.
It’s also why revenue-oriented cases tend to outperform cost-cutting ones. Cutting cost sounds clean on a slide, but realizing it means identifying exactly who and what comes out of the process and then acting on it, which is hard to do unless the work is highly uniform. Building the case around growth is often both more achievable and easier to prove.
Handing people an AI tool feels like progress, and it can help. But on its own, it’s an efficiency play: the same job, done a little faster. Think of it like poker: buying the tool is the ante that gets you a seat at the table, not a credible bet on the outcome. Too often, the promised savings never materialize, because the business case was never tied back to a real value-creation event.
Real ROI comes from transformation: redesigning the work, not layering AI on top of it. AI is an accelerator, and that cuts both ways: point it at a broken process and you’ll simply hit the same bottlenecks faster. Put it on a process that already works and you’ll get modest efficiencies that may not justify the implementation and ongoing token costs. The companies seeing outsized returns spend most of their early time and budget somewhere else entirely: redesigning the process around its inputs and outputs, and asking how AI can produce a better output, faster, rather than preserving the process as it exists today.
A useful rule of thumb: in the first 3 to 12 months, roughly 80% of the budget should go to process transformation, adoption, and change management, not the tool itself. Consistency is part of that. A tool that some people love and others ignore is nearly impossible to measure; a new routine that becomes second nature for the whole team shows up clearly against historical output. If your AI business case is almost entirely tool and token cost, it’s likely a business case for disappointment.
Predicting exactly how much AI will move a metric is mostly guesswork. No one has done it enough times to know whether you’ll gain five points or fifteen. Instead of asking “how much will this improve customer acquisition?” ask the question you can actually answer: how much would customer acquisition have to improve to make this investment worthwhile?
You alreadyknow the costs: implementation plus ongoing usage. Working backward from themusually reveals that the required improvement is surprisingly small, becausesmall gains in the metrics that drive top-line revenue compound into largeresults. Framing it this way turns a guessing game into a conversation everyonein the room can follow. It also prevents a common self-inflicted wound: settingan arbitrary 10% target, hitting a genuinely strong 7%, and calling asuccessful project a failure.
Two guardrails keep the math credible. First, make sure the metric is actually measurable: vague goals like “better decision-making” are real but nearly impossible to quantify, so anchor to durable, core-capability measures like support tickets resolved per day or call resolution time. If the right metric doesn’t exist yet, derive it with a transparent equation and, wherever possible, an existing benchmark. Second, sanity-check the ambition: if your case depends on more than doubling the best result your organization has ever posted, it’s time to scope it down.
For all itsnuance, building an AI business case comes down to three steps.
• Ask. AI ROI is a CFO game, so start with the CFO. Agree up front on what you’ll measure, what “good” looks like, and the benefit accumulation point: the moment it’s realistic to see measurable results. If your business is seasonal, that moment may be a full cycle away, and the plan has to account for it.
• Solve. Work backward from that point to today. How much work is required, how will you deliver it, and how will you measure, temper, and adjust your benefit expectations as you learn?
• Resize. If the improvement you’d need is more than twice your best year ever, keep scoping down until the target is believable. A 90-day horizon is often the sweet spot: long enough to do something meaningful, short enough to prove value quickly.
AI ROI isn’t amystery to be admired from a distance. It’s an equation — one with a metric, acost, a timeline, and a target — and it’s solvable. The organizations pullingahead simply do the work of building the case before they build the solution.
At Launch, that business case is often where our engagements begin. If you’re weighing an AI investment and want to pressure-test the numbers before you take them to the board, let’s talk.
AI adoption has tripled in just a few years: the share of organizations using AI in at least one business function has climbed from 33% to 78%. Yet fewer than one in five report a meaningful lift in earnings because of it. Deployment is up and to the right; AI ROI has barely moved. If that gap sounds familiar, it should: every technology wave, from electricity to the cloud, opened with the same lag between excitement and payback. The good news for finance leaders is that the problem usually isn’t the technology. It’s how the business case is built.
Building thatcase well is a CFO problem, and a solvable one. Here’s how to think about thereturn on an AI investment the way the companies actually capturing it do.
The first mistake is framing the question as “should we use AI?” That’s rarely the decision that matters. AI isn’t a strategy on its own; it’s a means to improve a specific process or deliver a specific outcome. The organizations getting this right don’t build one sweeping business case for AI. They build specific business cases for specific processes, each with its own metric, its own economics, and its own definition of success.
Those cases tend to fall into two categories, and the distinction matters more than it looks. One is rolling out an AI assistant, a tool like a coding copilot that helps people do their existing jobs faster. The other is transforming a process, redesigning how the work itself gets done. They carry very different economics, and confusing the two is one of the fastest ways to end up with a business case that never pays back. The mindset that ties them together: treat AI as a value initiative, not a cost-cutting one.
The most commonplace business cases fall down is the very first question: what are we actually trying to achieve? Adoption is not the only answer. A strong success metric has three parts:
• Business value. Why you’re doing this at all: top-line revenue, customer acquisition, customer retention. Start broad, then go a level deeper. “Grow revenue” is too vague to tie to one process; “improve lead-to-contact conversion” or “increase weighted pipeline creation” is a leading indicator of revenue that a specific process can actually move.
• Quality. Value without quality is a trap. In software, you can ship more features faster, but if quality slips, the gains reverse downstream. Whatever the process, pair every value metric with a quality metric that keeps it honest.
• Risk. What does the worst case look like? Scale cuts both ways. An AI agent that drifts in a call center isn’t one employee making an occasional mistake, it’s every call going wrong at once. Understanding and monitoring that downside is part of the case, not an after thought.
It’s also why revenue-oriented cases tend to outperform cost-cutting ones. Cutting cost sounds clean on a slide, but realizing it means identifying exactly who and what comes out of the process and then acting on it, which is hard to do unless the work is highly uniform. Building the case around growth is often both more achievable and easier to prove.
Handing people an AI tool feels like progress, and it can help. But on its own, it’s an efficiency play: the same job, done a little faster. Think of it like poker: buying the tool is the ante that gets you a seat at the table, not a credible bet on the outcome. Too often, the promised savings never materialize, because the business case was never tied back to a real value-creation event.
Real ROI comes from transformation: redesigning the work, not layering AI on top of it. AI is an accelerator, and that cuts both ways: point it at a broken process and you’ll simply hit the same bottlenecks faster. Put it on a process that already works and you’ll get modest efficiencies that may not justify the implementation and ongoing token costs. The companies seeing outsized returns spend most of their early time and budget somewhere else entirely: redesigning the process around its inputs and outputs, and asking how AI can produce a better output, faster, rather than preserving the process as it exists today.
A useful rule of thumb: in the first 3 to 12 months, roughly 80% of the budget should go to process transformation, adoption, and change management, not the tool itself. Consistency is part of that. A tool that some people love and others ignore is nearly impossible to measure; a new routine that becomes second nature for the whole team shows up clearly against historical output. If your AI business case is almost entirely tool and token cost, it’s likely a business case for disappointment.
Predicting exactly how much AI will move a metric is mostly guesswork. No one has done it enough times to know whether you’ll gain five points or fifteen. Instead of asking “how much will this improve customer acquisition?” ask the question you can actually answer: how much would customer acquisition have to improve to make this investment worthwhile?
You alreadyknow the costs: implementation plus ongoing usage. Working backward from themusually reveals that the required improvement is surprisingly small, becausesmall gains in the metrics that drive top-line revenue compound into largeresults. Framing it this way turns a guessing game into a conversation everyonein the room can follow. It also prevents a common self-inflicted wound: settingan arbitrary 10% target, hitting a genuinely strong 7%, and calling asuccessful project a failure.
Two guardrails keep the math credible. First, make sure the metric is actually measurable: vague goals like “better decision-making” are real but nearly impossible to quantify, so anchor to durable, core-capability measures like support tickets resolved per day or call resolution time. If the right metric doesn’t exist yet, derive it with a transparent equation and, wherever possible, an existing benchmark. Second, sanity-check the ambition: if your case depends on more than doubling the best result your organization has ever posted, it’s time to scope it down.
For all itsnuance, building an AI business case comes down to three steps.
• Ask. AI ROI is a CFO game, so start with the CFO. Agree up front on what you’ll measure, what “good” looks like, and the benefit accumulation point: the moment it’s realistic to see measurable results. If your business is seasonal, that moment may be a full cycle away, and the plan has to account for it.
• Solve. Work backward from that point to today. How much work is required, how will you deliver it, and how will you measure, temper, and adjust your benefit expectations as you learn?
• Resize. If the improvement you’d need is more than twice your best year ever, keep scoping down until the target is believable. A 90-day horizon is often the sweet spot: long enough to do something meaningful, short enough to prove value quickly.
AI ROI isn’t amystery to be admired from a distance. It’s an equation — one with a metric, acost, a timeline, and a target — and it’s solvable. The organizations pullingahead simply do the work of building the case before they build the solution.
At Launch, that business case is often where our engagements begin. If you’re weighing an AI investment and want to pressure-test the numbers before you take them to the board, let’s talk.
