Are you asking yourself "How much should we spend on AI?"
Whether a business is owner-led, venture-backed, or somewhere in between, the same pattern keeps emerging: leadership teams set aside a dedicated AI budget, then struggle to deploy it well. Tools get bought, pilots get launched, a few impressive demos do the rounds at the board meeting and six months later nobody can point to a material change in anything. The problem is not the size of the spend. It is the frame around it. Treating AI as its own budget line encourages leaders to buy capability first and figure out the use case second.
The better question is not "how much should we spend on AI?" but "how should our operating model change, and where does AI capability belong inside that change?"
This insight examines why the targeted-spend approach fails, offers a simple analogy to reframe the decision and proposes a practical route for any leadership team, whether you lead an established business, a scale-up, or a portfolio of companies, to embed AI where it actually creates value: inside the process, not on top of it.
The pattern we keep seeing
Talk to almost any business owner, managing director, scale-up COO or portfolio operating partner right now and the conversation follows a similar path. The board, an investor, or the owner has asked, reasonably enough, "what is our AI strategy?". The honest answer, in most cases, is that there isn't one yet, so a number gets attached to the question instead. A percentage of the technology budget, a headcount allowance for a small AI team, a licence spend for a handful of tools. The spend exists but the strategy does not.
What follows is predictable. Individual teams adopt tools that solve a narrow, visible problem: a chatbot for customer queries, a copilot for the sales team, a summarisation tool for the analysts. Each purchase is defensible on its own terms but none of them touches the underlying process. The result is a layer of AI tooling sitting on top of the same workflow that existed before, doing the same handoffs, the same approvals, the same manual reconciliation, the same waste, just slightly faster in places and with a new subscription cost attached. Leaders end up with an AI portfolio and a straight-line cost, but no structural change in how the business actually operates. When the board asks for the return on that spend, the answer is thin, because the spend was never connected to an outcome. It was connected to a category.
This is not a technology failure. The tools mostly work as advertised. It is a sequencing failure: capability was acquired before the process it was meant to serve had been examined. It is a version of the old law of the instrument: give a team a new hammer and every problem starts to look like a nail. The tool arrives first, and the business then goes looking for something to use it on.
The electric car analogy
Here is a useful way to test whether a business is thinking about this correctly. No sensible transport or logistics leader sets a 'targeted EV spend' as their starting point. They do not begin with 'we should invest £2 million in electric vehicles this year' and then work out afterwards where to put them. They start with the transport problem: what needs to move, how far, how often, at what cost, under what constraints and with what emissions or regulatory pressure attached. Only once that picture is clear does the electric vehicle enter the conversation, as one option among several, evaluated against the job it needs to do. Sometimes an EV is the right answer. Sometimes a hybrid, a lease arrangement, or a change in delivery scheduling is the better one. The vehicle is a means, chosen in service of a transport strategy, not a category of spend that exists for its own sake.
AI deserves exactly the same discipline. Nobody would defend a 'targeted EV spend' that ignored the actual transport problem, yet that is precisely how most organisations are approaching AI today. They are buying the vehicle before they have asked what needs to move, and where, and why the current route is inefficient. The fix is not to spend less on AI. It is to stop treating AI as the category and start treating the process as the category, with AI as one of the tools available to redesign it.
Why this keeps happening
Three pressures push leadership teams toward the targeted-spend approach, even when they know better.
Board and investor pressure to 'have a position.' Boards and investors, understandably, want to know the business is not being left behind. The fastest way to answer that question is with a budget figure and a tool list, because both are easy to report on a slide. A genuine process review takes longer to explain and does not fit neatly into a single line of a quarterly update, so it gets deferred in favour of something that looks like progress.
Vendor-led buying. AI tooling is being sold aggressively and much of it is sold at the point of a visible pain, not at the point of process ownership. A sales team frustrated with manual lead qualification will buy a tool that promises to fix lead qualification. Nobody in that purchase is asking whether lead qualification, as currently designed, should exist in its current form at all.
A misplaced belief that AI is a skill gap rather than a design gap. Many leaders assume the barrier is technical literacy: get the team trained, get the tools in, and adoption will follow. In practice the barrier is usually that the process itself was designed around human constraints, sequential approvals, manual data entry, judgement calls that exist only because nobody previously had a good alternative. Those constraints do not disappear just because a tool is now capable of removing them. Training people to use a new tool inside an unchanged process caps the benefit at whatever that process was already capable of delivering.
What embedding, rather than layering, actually looks like
The alternative is not complicated in concept, though it is more demanding in practice than buying a licence. It means starting every AI conversation with a process, not a tool.
Map the process before naming the technology. Before any AI spend is agreed, the team responsible for a given process, whether that is customer onboarding, underwriting, supply chain planning, or investor reporting, should map the process as it exists today: the steps, the handoffs, the decision points and the reasons each one exists. Often, a step exists because of a constraint that AI now removes. That is the point at which AI becomes relevant, not before.
Redesign the process, then select the capability. Once the constraint is identified, the question becomes: what does this process need to look like now that this constraint no longer applies? That might mean removing an approval step, collapsing three handoffs into one, or restructuring who owns a decision. AI selection follows from that redesign. It is chosen to fit the new process, rather than the process being left in place to accommodate the tool.
Fund the redesign, not the tool. Practically, this means the budget conversation should sit with the leaders who own end-to-end processes, not solely with a technology or innovation function. The spend approval should be tied to a defined change in a workflow and a measurable outcome, cost per transaction, cycle time, error rate, rather than to a category of software.
Treat AI capability as infrastructure, not a project. Once a process has been redesigned around AI, the capability needs to be maintained and evolved as the business scales, in the same way any core system is maintained. This is a shift from viewing AI adoption as a series of discrete pilots to viewing it as part of the operating model itself.
A quick diagnostic: three questions
Leaders do not need a lengthy audit to tell which camp they are in. Three questions, asked honestly, usually surface the answer.
If we took away the AI tools we have bought in the last twelve months, would the process underneath look any different? If the honest answer is no, the tools have been layered on top of an unchanged workflow rather than embedded within a redesigned one. The process still has the same steps, the same handoffs and the same approvals it had before the spend. The AI is a decoration, not structure.
Who owns the AI budget, the technology function or the leaders who own the process it is meant to improve? When AI spend sits with a central technology or innovation team, disconnected from the people accountable for the end-to-end process, tool selection tends to be driven by what is available and well marketed rather than by what a specific workflow actually needs. When the budget sits with the process owner, spend is naturally tied to an outcome, because that leader has to answer for the result either way.
Can we name the metric that is supposed to move, and has it moved? Cost per transaction, cycle time, error rate, headcount avoided, whatever the relevant measure is. If the AI spend cannot be tied to a specific number that has actually shifted, it was never connected to a defined problem in the first place. A "yes, adoption is up" answer is not sufficient. Adoption of a tool is not the same as improvement of a process.
An organisation that struggles to answer any of these three well is very likely buying vehicles before it has solved the transport problem. Two weak answers out of three is worth a proper process review before the next AI budget cycle is agreed. All three answered confidently, with evidence, is a reasonable sign that AI is genuinely embedded rather than layered on top.
What this means
For established, owner-led businesses, which is where this pattern shows up most often, the trap is treating AI as something the IT function or an outside vendor bolts on while the business carries on as before. The processes that matter most were built years ago and have rarely been questioned, so a tool laid over them changes little. The move is to take one process that genuinely hurts, work out why each step exists, and rebuild it before choosing any technology at all.
For scale-up leaders, the challenge is different: existing processes are already in place, often built under time pressure during earlier growth and changing them carries real short-term disruption risk. The discipline here is prioritisation. Not every process needs to be redesigned at once. The right approach is to identify the two or three processes with the highest cost of inefficiency, redesign those properly and resist the temptation to spread a fixed AI budget thinly across many teams.
For founders, the picture has changed. A venture-backed start-up today is usually AI-native by necessity, so the risk is no longer being slow to adopt. It is the opposite: a strong AI product can disguise an operating model that was never deliberately designed. The advantage of being small is the freedom to build good processes from the outset, so use it, rather than assuming the technology has done that work for you.
For investors, venture and private equity alike, the implication is about diligence. A company's AI spend figure tells you very little. The sharper question is which processes have actually been redesigned, and what changed as a result. For venture, a high spend with no process change is a warning sign rather than a strength. For private equity the same holds with harder edges: ahead of an exit, buyers look past the AI budget line and ask what has structurally changed in the cost base or margin profile, and embedded AI is far more defensible in a valuation conversation than a list of licences that could be switched off without consequence.
So what next?
The instinct to set a targeted AI budget is understandable. It gives leadership teams and their boards a number to point to and a sense that action is being taken. But a budget is not a strategy and a tool is not a transformation. The businesses that will get genuine value from AI over the next few years will not be the ones with the largest AI line item. They will be the ones that treated AI the way a good operations leader treats any capability: as something chosen in service of a clearly understood problem, not as a category of spend that exists to be seen to exist.
Nobody sets a targeted EV budget. They solve the transport problem, and sometimes the answer is an electric vehicle. The same logic should govern how every leadership team, whether owner-led, scaling, or under institutional ownership, thinks about AI. Start with the process, and let the technology follow.
If you want to see where and how AI is most likely to pay back in your business take our free 5-minute AI Reality Check Assessment and get your personalised report and recommendations on where to go next with AI.
