The Driver Tree Method: Finding AI Use Cases That Actually Move EBIT

Stop asking "where can we use AI?" Start asking "where is our EBIT leaking?"

This single reframe separates AI initiatives that generate real returns from the majority that consume resources without impact. Most organizations approach AI use case selection backwards—they start with the technology and go looking for problems. High performers start with expensive problems and evaluate whether AI can solve them.

The method I'm about to share comes from private equity, where the discipline of value creation is ruthlessly quantified. PE firms don't invest in "innovation" abstractly. They identify specific operational levers, calculate the financial impact of improving them, and deploy resources accordingly. The same discipline applied to AI use case selection transforms it from an exercise in possibility to an exercise in priority.

The Problem with "Where Can We Use AI?"

When organizations ask where they can use AI, they generate long lists. Customer service chatbots. Document summarization. Email drafting. Meeting transcription. Code generation. Content creation. The list grows easily because AI capabilities are genuinely broad.

But breadth isn't the problem. Priority is.

Every item on that list requires investment: implementation time, integration work, change management, ongoing maintenance. Resources are finite. Deploying AI to solve a low-value problem means not deploying it to solve a high-value one. And most organizations, lacking a framework for prioritization, end up pursuing use cases that are easy or interesting rather than use cases that matter.

The result is familiar: a portfolio of AI pilots that work technically but don't move business metrics. Leadership asks what return the AI investment is generating. The answer is uncomfortable silence.

The Driver Tree Method prevents this by starting with financial impact and working backward to use cases.

Building the Tree

A driver tree is a hierarchical decomposition of a financial metric into its component parts. You start with the outcome you care about—typically EBIT, revenue, or a key cost category—and break it down into the factors that determine it.

Take EBIT as the starting point. EBIT equals Revenue minus Costs. Revenue equals Volume multiplied by Price. Volume is driven by leads, conversion rate, and retention. Price is influenced by product mix, discounting, and competitive positioning. Each of these can be decomposed further.

On the cost side, you have labor costs, materials costs, and overhead. Labor costs break down by function: sales, operations, customer service, finance, and so on. Each function has its own cost drivers: headcount, productivity, error rates, rework.

The tree keeps branching until you reach operational metrics that are specific enough to act on. The number of customer service inquiries handled per hour. The error rate in invoice processing. The time required to generate a proposal. These are the leaves of the tree—the granular activities that roll up into financial outcomes.

The power of this structure is that it connects day-to-day operations to financial results. Every leaf has a quantifiable impact on EBIT. Improve that leaf, and the improvement flows up through the tree to the bottom line.

Finding the Expensive Problems

With the driver tree built, the next step is identifying which leaves represent expensive problems—operational pain points where improvement would generate meaningful financial impact.

Not all problems are equally expensive. A process that costs your organization $50,000 annually in labor is a different priority than one costing $5 million. This sounds obvious, but most AI use case discussions happen without this quantification. Teams pursue the problems they find interesting or the problems that are easiest to solve, regardless of financial magnitude.

To find expensive problems, work through the tree asking three questions at each leaf.

First, what is the current cost or revenue impact of this activity? Quantify it in dollars. If it's a labor cost, calculate fully-loaded compensation multiplied by time spent. If it's revenue, estimate the value at stake from conversion rates, deal sizes, or customer retention.

Second, what is the realistic improvement potential? Not the theoretical maximum, but what a successful implementation might actually achieve. A 30% reduction in processing time? A 50% decrease in error rates? Be conservative.

Third, what is the annualized financial value of that improvement? Multiply the current impact by the improvement potential. This gives you a rough order of magnitude for the prize.

Rank your leaves by this annualized value. The top of the list shows you where expensive problems live.

The AI Overlay

Only now—after you've identified expensive problems—do you ask whether AI can help solve them.

This is crucial. The question isn't "what can AI do?" It's "can AI address this specific expensive problem better than alternatives?"

For each expensive problem on your list, evaluate AI fit using four criteria.

Is the task high-volume and repeatable? AI excels at processing volume. If the expensive problem involves ten instances per year, AI probably isn't the answer. If it involves ten thousand, AI becomes compelling.

Does the task involve pattern recognition or generation? AI's core capabilities are recognizing patterns in data and generating content based on learned patterns. Problems that fit these capabilities are strong candidates. Problems requiring physical manipulation, real-time judgment in novel situations, or deep relationship management are weaker fits.

Is the data available and accessible? AI requires training data or context to perform well. If the expensive problem involves information trapped in legacy systems, scattered across email threads, or locked in people's heads, the AI implementation becomes primarily a data problem.

Is the error tolerance appropriate? AI systems make mistakes. For some applications, a 5% error rate is acceptable—the efficiency gains more than compensate. For others, a 0.1% error rate is unacceptable. Match the AI's realistic accuracy to the problem's requirements.

Problems that score well on all four criteria—expensive, high-volume, pattern-based, data-rich, and error-tolerant—are your highest-priority AI opportunities.

A Worked Example

Let me make this concrete. Consider a mid-sized professional services firm—a consultancy, law firm, or accounting practice—with $200 million in annual revenue and EBIT margins under pressure.

Building the driver tree for labor costs in client delivery reveals several branches: partner time, senior associate time, junior associate time, and support staff time. Each branch has activities: client communication, research, analysis, document creation, review, and administration.

Quantifying these activities surfaces an expensive problem: proposal and pitch creation consumes roughly 15,000 hours annually across the firm. At blended rates, that's approximately $3 million in labor cost. More importantly, slow proposal turnaround is costing deals—the business development team estimates $5 million in annual revenue lost to competitors who respond faster.

The total prize for solving this problem: $8 million annually between cost reduction and revenue capture.

Now apply the AI overlay. Proposal creation is high-volume—the firm produces hundreds annually. It involves pattern recognition (matching client needs to firm capabilities) and generation (creating customized narrative and pricing). Historical proposals provide training data. And error tolerance is moderate—proposals go through human review before delivery, so AI mistakes get caught.

This is a high-priority AI opportunity. The expensive problem is quantified. The AI fit is strong. The business case writes itself.

Compare this to another potential use case surfaced by the same firm: using AI to summarize internal meeting notes. It's technically feasible. It would save some time. But quantifying the impact reveals maybe $100,000 in annual value—two orders of magnitude smaller than the proposal opportunity.

Without the Driver Tree Method, a team might pursue both opportunities with equal enthusiasm, or even prefer the meeting summarization because it's simpler to implement. With the method, the priority is clear.

Connecting to the Business Case

The Driver Tree Method does more than identify opportunities. It constructs the business case for pursuing them.

When you've worked through this process, you can walk into a meeting with your CFO and say: "We've identified that proposal creation costs us $3 million in labor and loses us $5 million in deals annually. We believe AI can reduce labor costs by 40% and improve win rates by accelerating turnaround. The conservative annual value is $2.5 million against an implementation cost of $400,000. Payback period is under three months."

That's a conversation about investment and returns, not a conversation about interesting technology. It's the conversation that gets funding approved.

The organizations stuck in pilot purgatory often struggle to make this case because they never did this analysis. They built AI capabilities first and tried to justify them afterward. The Driver Tree Method inverts this sequence, ensuring every AI initiative is grounded in quantified business impact from the start.

Getting Started

If you want to apply this method in your organization, start small. Pick one branch of your cost structure or revenue model—a single function or process area. Build the driver tree for that branch. Quantify the leaves. Identify the two or three most expensive problems.

Then evaluate AI fit for those specific problems. You'll likely find that some expensive problems aren't good AI candidates, and that's valuable information. You'll also likely find at least one opportunity where the combination of financial impact and AI fit makes a compelling case.

That's your starting point—not the use case that's easiest or most interesting, but the one that actually moves EBIT.

Measured AI helps business leaders identify and prioritize AI opportunities that generate measurable returns. Subscribe for weekly frameworks on crossing the gap from experimentation to enterprise value.