What CFOs Need When Everyone Wants AI Money

The problem is not a shortage of ideas

CFOs are getting hammered with requests to fund new AI initiatives.

Marketing wants a personalization engine that can optimize campaigns in real time. HR wants a recruiting copilot that can screen candidates. Customer service wants agents that can resolve routine requests. Operations wants better demand forecasting. Finance wants autonomous workflows for monthly close.

Individually, each request sounds sensible. Together, they create a new kind of challenge.

These initiatives often compete for the same data, engineers, security approvals and end users. Vendors may offer overlapping capabilities, while teams describe benefits in different ways, making proposals difficult to compare. When every request is urgent, funding becomes a race rather than a coordinated decision.

CFOs are right to be cautious. They need to know these investments will produce a return—not disappear into a hype cycle and leave finance holding the bag.

The answer is not another approval committee. It is better operating systems.

Two frameworks bring control and structure at a time like this: PI Planning and Initiative Funding Gates. One coordinates the work. The other controls the money.

Why approving initiatives one at a time fails

Most investment processes evaluate proposals as though they are independent.

A functional leader writes a business case. Finance tests the assumptions. Technology estimates the work. A sponsor approves the budget. The initiative joins the roadmap.

That process can work when demand is limited and the projects barely touch one another. AI initiatives rarely behave that way. They compete for scarce technical people, rely on shared data and platforms, and often require the same legal, security, procurement and change-management support. The return from one initiative may depend on another being delivered first.

Approving each request separately hides those relationships.

Imagine four initiatives are approved in four different rooms: an AI marketing engine, a service agent, an employee copilot and a finance automation tool. Each has a sponsor. Each has a positive business case. Each needs access to customer or employee data, integration support, security review and help changing the way people work.

This is where delays begin. A data dependency appears late. Two teams discover they bought overlapping tools. The security group becomes the critical path for everyone. One initiative launches but cannot demonstrate value because the process around it never changed. Finance can see what has been spent, but not whether the portfolio is moving toward a shared outcome.

The problem is not that the individual business cases were wrong; it is that they were never assessed and planned as a portfolio using a modern portfolio management system.

What PI Planning does

Program Increment ("PI") Planning is a structured way to align multiple teams around a shared set of objectives for the next 'increment' of work.

The full textbook version can be heavy and most organizations do not need the full song and dance to benefit from it. They need the discipline underneath it: bring the relevant teams into the same planning cycle, make the dependencies visible, agree what matters most, and leave with one coordinated delivery plan.

That is 'light-touch' PI Planning.

Start with a small number of measurable business outcomes. “Use AI across the company” is not an objective. “Reduce average service cost by 15% without lowering customer satisfaction” is.

Next, put the workstreams together. Marketing, operations, technology, data, security, finance and the end users should be looking at the same plan. This is where hidden dependencies surface. If three initiatives need the same data pipeline, that becomes a portfolio priority rather than three separate escalations. If the organization lacks the capacity to deliver everything, the constraint becomes visible before teams start spending.

Then agree what each team will deliver during the increment and how those outputs connect. The goal is not to create a perfect plan. It is to create a plan that is coherent enough for teams to move together and specific enough for leaders to see where it is breaking.

Done well, PI Planning answers four questions:

1. What outcomes are we trying to create?
2. Which initiatives matter most to those outcomes?
3. What does each team need from the others?
4. What signal should exist by the end of the increment?

That final question is where Initiative Funding Gates enter the picture.

What funding gates do

Funding gates release investment in stages, with each tranche dependent on progress from the last.

You still model the full investment up front. Leaders should understand what the initiative may cost if it runs to completion and what value it is expected to create. But approval of the whole case does not mean releasing the whole budget.

Fund the first stage. Agree what signal it must produce. Then return to the gate and make an explicit decision: continue as planned, continue with changes to scope or budget, pause to gather more signal, or stop.

For a digital product, the sequence might move from a working prototype, to signal that users return, to a target number of active accounts, to recurring revenue. For a transformation initiative, it might move from an agreed baseline and business case, to a successful pilot, to adoption across a defined group, to benefits realized in the P&L. For an internal AI tool, the first gate may test whether the tool works at all; the next whether employees use it; the next whether it changes cycle time, quality or cost.

The point is not the label on the gate. It is that the success measure is agreed before the work begins.

That matters because teams are very good at finding a positive interpretation of whatever happened. A pilot misses the adoption target but produces “strong learnings.” A tool saves no measurable time but receives encouraging feedback. A project ships every feature and quietly misses the benefit it was funded to create.

Those may all be useful findings. They are not all reasons to release the next tranche.

A gate forces the distinction between activity and real signal. It asks what was delivered against the agreed measures, what it cost against budget, what is now known that was not knowable before, whether the value case still holds, and what the next release of money will buy.

If the only possible outcome is “continue,” it is not a funding gate. It is a status meeting with a budget attached.

Why the two controls belong together

PI Planning without funding gates can create an impressively coordinated way to spend too much money. Funding gates without PI Planning can create disciplined decisions around work that was never properly coordinated.

The two solve different parts of the same problem.

PI Planning creates alignment around objectives, makes dependencies visible and organizes the work required to produce signal. Funding gates judge that signal and decide whether the next stage still deserves investment.

One operates inside the increment (i.e. during delivery). The other operates at its boundary (i.e. at each funding gate).

That distinction matters. A gate does not create signal; it judges it.

During the increment, customers and employees should interact with the real product or service—not just react to the idea.

If it is technology, let them use it. If it is a service, deliver the part that can already be experienced.

Their behavior, feedback and results become the signal assessed at the gate.

This is where agile delivery is useful. Short delivery cycles, demonstrations and customer feedback create regular opportunities to learn. PI Planning connects that learning across teams. Funding gates connect it to capital allocation.

Together, the two frameworks create:

• Accountability. Every increment has named owners, agreed outputs and a decision waiting at the end of it.
• Visibility. Leaders can see the objectives, dependencies, spend, signal and current value case in one place.
• Agility. Scope and funding can change as signal emerges, while there is still time and money left to redirect.
• Alignment. Finance, leadership, delivery teams and end users work from the same objectives and measures rather than separate versions of success.

There has rarely been a more important time to have all four.

What the operating systems looks like

This does not need to become a new bureaucracy. A workable version can run on a simple cycle.

Set the portfolio objectives

Start with a small number of measurable business outcomes. “Use AI across the company” is not an objective. “Reduce average service cost by fifteen per cent without lowering customer satisfaction” is.

Every proposed initiative should connect to one of those outcomes. If it cannot, it may still be a good idea, but it does not belong in this portfolio yet.

Select the increment

Choose the work that can realistically be delivered together over the next one to three months. Make dependencies and shared constraints visible. If six teams need the same engineering group, resolve that in planning rather than through six escalations later.

Agree the signal

Before funding is released, write down what must be true at the next gate. Use measures that can change the decision: adoption, retention, cycle time, error rates, customer behavior, cost removed or revenue created.

Avoid measures that merely prove the team was busy. Licences deployed, prompts written, people trained and features shipped may be useful operating metrics. On their own, they are not a return.

Deliver and learn

Teams work in short cycles, demonstrate real output and involve the people expected to use it. New information should update the plan as it appears. The goal is not to protect the original business case. It is to find out whether the investment still deserves to exist.

Hold the gate

Keep the gate presentation pack short and consistent:

• Delivered: Results against the measures agreed before the stage.
• Cost: Actual spend against the tranche budget.
• New signal: What is known now that was not known before.
• Value case: Updated benefits, costs and assumptions.
• Next tranche: What the next release of funding will buy.

Then decide. Continue. Change. Pause. Stop.

Reallocate visibly

When an initiative stops, show where the money and capacity go next. This matters more than it sounds. If teams learn that stopping a project only means losing their budget, they will defend every initiative to the end. If they see resources moving toward stronger opportunities, a stop becomes signal that the portfolio is working.

Where this goes wrong

• The first failure is putting the ceremony in place without changing the decisions. Teams attend planning sessions, create dependency maps and prepare gate packs, but every initiative still continues. The organization has added administration, not control.
• The second is making the stages too large. By the time the first gate arrives, the platform has been bought, the vendor contract signed, the team hired and the launch announced. Technically, leaders can still stop. Psychologically and politically, the decision has already been made.
• The third is making them too small. If teams spend more time preparing for gates than producing signal, the stages are wrong. The review burden breeds resentment, and leaders end up judging work before there is enough information to learn anything useful.
• The fourth is letting the sponsor mark their own homework. The person who fought hardest for an initiative is rarely the best person to decide, alone, whether it should stop. Finance should be in the room, along with at least one leader who has no stake in the initiative continuing.
• The fifth is treating all AI initiatives alike. Some investments are experiments. Some are infrastructure. Some are mandatory responses to risk. Some are straightforward productivity tools. They should not all face identical signal requirements, but they should all face an explicit decision appropriate to the uncertainty and money involved.
• The final failure is confusing an AI result with a business result. An agent can answer the question. A model can produce the forecast. A copilot can draft the campaign. None of that creates value until a process changes, someone adopts the output, a customer behaves differently, or the result reaches the accounts. That is the standard the portfolio has to hold.

The plan and the money should move together

The AI investment wave is going to produce important businesses, better products and genuinely different operating models. It is also going to produce duplicated tools, abandoned pilots and expensive lessons dressed up as transformation.

A CFO cannot predict perfectly which initiative will become which. That is not the job.

The job is to build a system that learns early enough to do something about it.

Light-touch PI Planning gives teams one set of objectives, one view of the dependencies and one coordinated delivery approach. Funding gates release investment as the signal earns it. Together, they give leaders accountability for delivery, visibility into spend and value, agility to change course, and alignment between finance and the people doing the work.

When everyone wants AI money, the answer is not yes to all of it. It is one plan, staged funding, and measures agreed before the spending begins.