Value before velocity: Budgeting for AI does not make it your AI strategy

Every department wants an AI budget. Who owns the strategy?

Many organizations fund AI the way they fund everything else: one department at a time. Marketing asks for budget to cover its tools. Engineering asks for budget for its licenses. Operations asks for a pilot. Each request is sensible. Added together, they make up a budget, not a coherent strategy.

The trouble is that AI doesn’t follow the org chart. The value sits in the seams between functions, where a decision in product changes the work in engineering, the data in finance, and the experience of a customer. A budget built department by department can overlook those seams, funding the parts while missing the whole.

We’ve been exploring why this keeps happening and what leaders can do about it. Our view is that the way you fund AI is part of the way you work, and it deserves the same redesign as everything else. People, process, and technology deliver more effectively when they are treated as one system. The budget is where the system either holds together or quietly breaks apart.

Why the budget process rewards the wrong bets

Start with how the budget gets built. It is usually a bottom-up exercise. Each team works out what it needs, adds some extra to show ambition and a little more because it expects cuts anyway, all based on last year’s numbers. It’s painful enough before anyone tries to account for AI. Expect several rounds of negotiation and expect them to run into the new year.

That design has a side effect. Asking for more can become a way to protect an allocation, and spending it can feel safer than leaving it unused. So teams reach for requests that are credible, easy to explain, and very likely to come off. A cross-functional idea means depending on someone else’s priorities and someone else’s delivery. A high-risk idea is harder to defend. Aggregate from the bottom and the least attractive items on the page are, by design, the cross-functional and the uncertain.

AI is exactly that kind of idea. Some of its biggest opportunities need several functions to change how they work at the same time, and the outcome can’t be promised in advance. The process screens out the opportunities with the most upside.

Zero-based budgeting, which asks every team to justify its spend from scratch, can push the habit further. When every line has to survive a round of questions focused on certainty, high-risk initiatives can lose out. In too many organizations the budget has stopped being an investment discipline that weighs risk against reward. It has become a writing exercise: how do I describe what I want to do so that it attracts the fewest objections?

Efficient departments don’t add up to an efficient organization

Lean thinking has an old answer to this: global optimization beats local optimization. Pull every request for a shared resource into one central team, and it looks efficient on paper. Apply that logic across a whole organization and the central team becomes the bottleneck that slows every project down.

Budgets usually follow the same pattern. Each department’s allocation is carefully tuned, but nobody is looking at the end-to-end value stream, the full path from an idea to something customers value. Speed up one step and the constraint simply moves to the next.

We see the effect in how AI is deployed today. For an individual, it can make you remarkably productive. For a team, it often becomes one more tool, because it only improves the flow of work when it is applied at the bottleneck. Similarly, a budget carved up by department doesn’t ask who will bridge the gaps.

There is a second blind spot. Technology is often the only thing that gets a line item in the budget. The people development and process change making the tech useful cannot be overlooked. A financial services firm that wants to behave more like a fintech, perhaps by turning its capabilities into software services, may struggle to get there if it plans and funds only by function. That is the people, process, and technology system we described at the start, funded as three separate things.

Start with value, not headcount

Ask how an AI program is being justified and you’ll often hear a headcount story. How many roles can we make redundant or move to a cheaper location to pay for the transformation? It’s the wrong way around. The first questions are where AI can create value and how much it can create. Savings may follow, but they can’t be the starting point.

Why is headcount still the currency of budgets? Because it’s the easiest number to believe. A budget request gets approved when it is credible, easy to follow, and easy to imagine becoming a reality. A headcount is all three. Customer uptake isn’t. Neither is anything that rewires your processes or restructures your cost base. So the safer number wins, even when it’s wrong. There’s a better lens: revenue or profit per head, paired with an honest view of how likely that outcome really is. Making the trade-off explicit takes practice and evidence you can stand behind.

We’d rather be candid about the harder part: measuring the return on AI across an organization remains difficult, as it does for any technology initiative. Studies have found productivity gains in specific settings, but those measures rarely add up to substantial returns for the organization as a whole.

Uncertainty about returns is an argument for discipline, not guesswork. Distinguish outcomes from output. Put an assumption and a probability against every expected benefit and be ready to revisit both as evidence arrives. Value before velocity, especially when AI makes velocity so cheap.

Fund high-risk bets separately from the business as usual

One of the most practical actions to take is to stop judging different kinds of spend by the same rules. Running the business means protecting current customers and current revenue. This makes low-risk and cost-saving initiatives more attractive than high-risk ones.

Innovating or radically growing the business means pursuing upside at considerable risk. This calls for a zero-based mindset and a tolerance for risk. Judging a high-risk venture by the standards of business as usual is like judging a rocket by its speed on the launch pad.

This is part of why AI-native companies and fintech startups move the way they do. They have fewer legacy commitments and established revenue streams to protect. That can give them more room to take risks. An established organization can borrow the idea by isolating the high-risk work. Call it a skunkworks, a new venture, or a separate experiment. What matters is that the budget, the rules, and the expectations are different.

There is a catch. Separating the two halves of the business solves a budgeting problem but creates a lifecycle problem: a bet that succeeds has to become part of the core business eventually, so the halves have to be brought back together when you think about products and development. And separate budgets don’t mean separate money. A business will only put so much cash into AI, so the bets and the day-to-day operations are always drawing on the same finite pot.

The point holds for organizations of every size. Getting AI to work isn’t mainly about the technology. It’s about what you are investing, what you are risking, and where.

Let the product lifecycle steer the money

If you already work in a product model, you have a better organizing principle than the cost center. A product’s lifecycle position should inform how freely it can spend and how much risk it can take. Products in growth, or in the middle of a pivot, deserve room to invest and experiment. Products in decline or being phased out should face a much tighter budget.

Larger enterprises are already reshaping their delivery models around AI while raising their AI investment, which makes the question of where the money goes more pressing, not less. The budget process works far better when it starts from value streams, products and product lines, and the experiences customers have, and far worse when it starts from the organizational unit, the cost center, or the list of activities.

Seeing the portfolio as products serving value streams, deciding where AI deserves investment, and agreeing what should stop is not a side project. AI will be a core part of the strategy. The result isn’t a better plan for next year. It’s a better ability to make the next investment decision, and the one after that.

Make experimentation a budget line, and keep the evidence

This is where experimentation comes into its own. In every product there are things you already know to be true and needn’t test. Software handling private customer data needs appropriate access controls. With AI, the ground moves far faster. Model providers release new models within weeks of each other, some cheaper, some more capable and more costly. Tools come and go. Nobody can promise exactly which processes they will automate before they have tried.

So the budget has to assume learning. A good policy or a good business case today won’t necessarily hold next quarter, because regulations, cost structures and competitors keep changing. That makes experimentation a continuing practice, not a phase you complete before the real build begins. It means returning to your hypotheses and checking whether they still hold. If your budget cycle runs for a year, allow for conditions to shift well before the next annual review.

AI has made it cheap to generate more ideas, more code, and more content. It hasn’t made it cheap to know which of them deserve to survive. That judgment comes from fast, inexpensive experiments and a willingness to say no early, before an idea has consumed real investment.

An experiment’s value is easier to sustain if its result outlives the team that ran it. That needs a single place where teams record the opportunity, the assumptions behind it, the evidence the experiment produced, and the decision that followed: fund, park, or kill. When the next budget conversation arrives, it starts from evidence rather than memory.

Finance wants what you want

The usual objection is that finance won’t support it. Finance wants certainty up front, a fixed forecast and a yearly cycle, while the business wants many small pots of money to test ideas. We saw the same tension when agile first arrived, and we’ve found it is more solvable than it looks.

At a large oil and gas company, we ran an experiment to answer a simple question: did the finance policy and process need to change to support agile delivery? It turned out they didn’t. What finance cared about wasn’t only the forecast spend. It was the return on investment, and no mechanism they had used before had given them a clear view of it. We committed to a multi-year spend and gave monthly updates on what was being spent against the value being generated, or the potential value where nothing had yet gone live. That was enough.

Much of the ceremony in traditional finance and procurement exists to protect against the unknown. Agile shortened the feedback loop, so the unknown became known sooner. The same logic applies to AI.

The same thinking works for experimentation budgets. Ring-fence a sum for the experimentation phase, sized so that it covers a good deal of experimenting across the workforce, and track it weekly and monthly. If spending exceeds the allocation, review the evidence and agree any additional funding with finance. Where policy allows, carry underspend into the next wave of experimentation. Respect the calendar too. For organizations planning on a calendar-year cycle, January can offer room to experiment before delivery commitments build. The timing should follow your planning cycle, with space for learning throughout the year.

By the way, none of these recommendations ask finance to abandon its principles. The principle was always to track what you spend and spend it responsibly. Few finance teams love the traditional budget process. The real obstacle is habit: the comfort and structure of familiar processes and the hesitation to accept uncertainty and risk, even when both are natural.

Questions to ask before your next budget cycle

AI and automation represent a significant shift in how work gets done. Making the most of it depends less on the technology you buy than on the organizational foundations beneath it. Before the next round of department-by-department requests lands, it’s worth asking:

  • Can anyone see our AI spend and its expected value end to end, across functions?
  • Is each investment justified by value created, or by headcount saved?
  • Have we separated the bets from the business as usual, with different rules for each?
  • Does the lifecycle of our products shape where the money goes?
  • Do we have a ring-fenced experimentation budget, with a monthly view of spend against value?

Beneath those questions sit three foundations. Do you have data you can trust and the governance to back it? Have you moved from a project model to a product operating model? Do you have a transparent way of working that balances value, flow, and quality? If the honest answer to any of them is no, the foundations aren’t there yet, and automation risks amplifying those weaknesses.

The organizations that get the most from AI won’t be the ones that spend the most on it. They’ll be the ones that get better, cycle after cycle, at deciding what matters, building only what has earned the investment, and developing people who can keep making those decisions themselves. That capability, not the size of the AI line in the budget, should be the strategy.