Boards have moved quickly from asking what AI can do to asking what it can return. This installment of the Agentic Leadership Playbook examines how CFOs should fund AI, where they should look for returns, and how the technology will reshape finance itself.
How have your conversations with CFOs about agentic AI changed over the past year?
A year ago, the ask was often, “Show me what AI can do.” Now I get called into meetings where the question is, “How do I run this program for a return and transform my business?”
That is a very different conversation. I think that there’s a pivot for CEOs and CFOs and, quite frankly, the general marketplace towards requiring that these investments provide a return. You are not giving money away. And leaders are starting to viscerally experience the potential of the technology. This is moving from an interesting experiment that helps people write better white papers to something that can fundamentally replace human toil. Rote, routinized work and even judgment-based work can go away, letting people move up the value curve.
When you start to think about it that way, this is the industrial revolution for white-collar work. You are creating virtual robotic assembly lines to manage workflows straight through in a way that was not possible before. RPA promised something similar ten years ago and was largely abandoned because it was a dumb robot that could only automate micro steps with no ability to apply judgment. These are smart robots. They are better at multi-step tasks, contextual interpretation, and working with each other as a virtual team.
So the question has moved from experimentation to industrialization. And when you industrialize, the expense goes up materially. Therefore, the returns have to go up commensurately.
How does a CFO fund a transformation when the board wants a payback period and the value does not follow a simple curve?
Start with the business, not the AI budget. The first question is, “What creates advantage and value in this business, and where can AI change the way we create it?”
That answer will differ by company. A hypergrowth technology business is valued heavily on future revenue and growth. A utility needs stable, predictable returns. Those are different starting points, and the funding logic should reflect that.
Then connect the AI program to the full P&L. Force a review of operations, assets, and expenses, and create a program to fund the journey. It doesn’t have to be dilutive. You don’t sacrifice profitability just because there’s a new business reality that you’re facing.
Inside the AI program, you also have to operate at two speeds. Invest toward the long-term vision, but harvest value now through rapid deployment, and use that performance delta to fund the larger ambition. Reinvent how you build, too. AI-native engineering is already producing 30% to 40% productivity gains in development, and the aspiration is higher.
Focus matters as much as funding. The market is moving away from a thousand flowers blooming to a well-curated garden. Investment dollars, engineering capacity, and managerial capacity are all scarce. So concentrate these resources on the major workflows where reinvention can create real value, rather than scattering them across dozens or hundreds of disconnected pilots.
What separates the finance functions making progress from those that are stuck?
You cannot benchmark your way to innovation.
Finance organizations, especially in regulated industries like banking, have a muscle I run into all the time: “Show me the twenty companies in my industry that have done exactly this and banked the value.” My answer, consistently, is no, or at least not yet, because that list does not exist. You’re either going to be first to market or you’ll be a fast follower.
An insurance client in Asia wanted to zero-base its finance function to eliminate the sub-ledgers, build an immutable data store that automatically reconciles every transaction, and close the books on a rolling basis. We mapped the architecture, tested the feasibility, and confirmed it was workable. They said, “Great. Name the ten insurers who have done this.” There was not one.
So you either get stuck in the “doubting Thomas” moment, or you run the proof of concept, move to an MVP, and prove it to yourself as you scale. My one Canadian comment is that you have to skate to where the puck is going or you will get left behind.
Where should finance functions begin, and where is the value still underappreciated?
Point a massive probability engine at probabilistic workflows, not deterministic ones. Using it to produce basic numbers is killing a fly with a howitzer.
Planning and forecasting is far and away the most developed use case because forecasting is a prediction problem. Connect the forecast to the lead input metrics—such as feet on the street, clients covered, average sale, and price point—and those driver trees do the work, so finance can say on day one: You are not going to hit your numbers, and here’s why.
That changes the role of finance. Finance starts talking to the business about the inputs that need to change rather than scolding it for missing a number. Ideally, the CFO becomes the chief performance officer.
Management reporting is next. We have deployed a finance insight agent, essentially an automated FP&A analyst, that handles the ad hoc requests that once forced humans to unwind the spaghetti. It returns facts, commentary, and suggestions with the FP&A leader still in the loop. In one deployment, it automated more than 90% of the business intelligence work in nine months. That used to take three years.
And the value exceeds the savings. In many FP&A teams, 80% of the work is data extraction, cleansing, form preparation, distribution, and simple calculations. That is not what you want FP&A doing. Use the app; do not be the app. Right now, too often, the team is the app and Excel is the codebase.
The underinvested area is treasury and balance sheet management, where value transcends the finance P&L. Think fully integrated hedging, cash modeling, sources and uses of cash, and asset-liability matching. One white goods manufacturer was losing about 3% a year to inefficient hedging. For banks, better insight into deposit stickiness can change how much liquidity buffer they need to hold. That is real enterprise value.
What do CFOs still tend to misunderstand about agentic AI deployments?
Many think it is about the application layer: what to buy or build, which systems integrator drops it in, whether it is like installing a new ledger or a business intelligence tool.
It is not. It is about the workflows you are reimagining. You are not dropping in a calculation kernel. You are creating a virtual workforce configured to do what your physical workforce does today. That means understanding what the work is, in what sequence it happens, using which systems, and where judgment enters the process.
Agents will often use your applications, not replace them. In one deployment, the reporting package was built in Anaplan. We did not rip out Anaplan. We built an agent to sit on top of it, produce the reports a person produces now, and interpret them the way an analyst would.
The other missing ingredient is the trilingual team. A finance person over here, technology people over there, and the work collapses because the third language, the business, is missing. Finance people are literate in finance, sometimes in the business, and almost never enough in technology. Technology teams often do not understand the finance workflow or the business context. Getting those languages, plus change management, inherently connected is the heart of it. The rest is program management and agile discipline.
If agents take on more analytical and transactional work, what do the CFO and finance team need to become?
The CFO needs to experience the technology and what it can do in as hands-on a way as possible. When the technology started to gain traction, I decided to get upskilled on everything and took all available trainings and started to experiment with the technology. Over a few weekends, I built a charity website and donation platform from scratch, incorporated three companies, and built a macroeconomic research engine and a trading algorithm just to see what could be done.
Instead of a PowerPoint deck, I run live demos with clients now. I will build the quarter-end competitive benchmarking that an analyst might spend days on, in front of the finance team, in three minutes. This makes it real, and shows them what they—and the technology—can do together. You learn by doing, or you do not learn.
For the team, the shift is from producers of things to interpreters, advisors, and deliverers of impact. The harder question is how to create the next generation of those people when the training ground was ten to fifteen years of doing the work that agents now do. We are outsourcing some of the work that trained critical reasoning itself. I see three answers: more classroom learning, more apprenticeship models, and assigning tasks you may not strictly need done, simply because people still need to learn how the work works.
What should CFOs do in the next 12 months?
Do something big and fast.
That applies most to white-collar businesses, which are the most disruptible. A year from now, others will be in the market with what they have built. The moats and walls are coming down. Scale is not the defensible position it once was. You used to have to hire an army to get it; now everybody can have scale.
For finance specifically, if you have not started reinventing business intelligence end to end, that has to move to the top of the agenda. If you cannot sit at your computer and get a heads-up display of the business, with the ability to drill down to the root-cause drivers of performance, you need to move quickly. That is where finance can become faster, better, and cheaper all at once. You lower the run-rate cost of finance, and you unlock better business decisions.
Built right, that layer also hardwires the connection between business activity and financial outcome. Knowledge that used to sit in a few people’s heads becomes institutional—preservable, referenceable, and teachable to the next generation. Done right, finance will know the business better, and be able to unlock transformative value.
