Roughly nine out of every ten pounds being spent on AI right now is going into infrastructure. Chips, data centres, tooling, platforms, integration work. The remaining fraction is going into the people who are actually supposed to use any of it.

That ratio comes from the AI Daily Brief's new Maturity Maps framework, a benchmarking exercise that pulls together 480-plus studies and roughly 150,000 professional survey responses from a single quarter. Their conclusion, after all that: 93% of enterprise AI spend lands on infrastructure, and 7% lands on people. The people bucket includes training, change management, redesigned roles, the unglamorous plumbing work of actually teaching an organisation to use what's been built.

Reasonable people can argue with the exact number. The framework is new, and I'm sure the methodology will get picked apart. But the direction is consistent with everything else that's being published right now. KPMG's Q1 2026 Global AI Pulse Survey — based on responses from senior leaders across eleven countries — found that only 11% of companies have actually reached the point of deploying and scaling AI agents to produce enterprise-wide outcomes. Everyone else is stuck somewhere in the pilot-to-production canyon.

The most persuasive pushback I've heard on this is that infrastructure spend front-loads naturally — you can't train people on an AI system that hasn't been deployed yet, so of course the ratio looks lopsided early. Fair point, as far as it goes. The trouble is that I can't find many people actively budgeting the people-side catch-up for when the infrastructure lands. If it were genuinely a sequencing issue, you'd expect to see three-year training plans kicking in at year two. Mostly what I see is year-one infrastructure and year-two... more infrastructure.

The familiar shape of the problem

None of this should be surprising to anyone who's lived through a previous technology transition. ERP rollouts in the 1990s. CRM in the 2000s. Cloud migrations in the 2010s. The pattern is always the same: the technology is the easy part, relatively speaking. The hard part is rebuilding the workflows, retraining the people, and absorbing the fact that your organisational structure was shaped around constraints that no longer exist.

What's different this time is the speed. The gap between "this tool exists" and "this tool is central to how work gets done" used to be measured in years. It now gets measured in months, and in some cases in weeks. Organisations that budgeted a three-year transformation programme are finding that the three-year version of the technology ships in six months, and the people side of the programme hasn't caught up.

I watched a version of this unfold when I was running teams at Vertical Leap. We'd roll out a new analytics platform, the tool would work perfectly, and three months later I'd discover that most of the team was still running their old spreadsheet workflow alongside it. Not out of obstinance. Because nobody had walked them through what the new workflow actually looked like, why it was different, or what they were supposed to stop doing. The tool was in place. The capability wasn't.

That's the shape of the current moment, but with billions on the line instead of thousands.

Why the imbalance is rational, and still wrong

It's worth being generous about why companies are pouring money into infrastructure rather than people. Infrastructure is legible. You can put it on a slide for an earnings call. "We signed a $4 billion commitment with Anthropic and Microsoft, and we're building three new data centres" is a story investors understand. It has a shape. It has a dollar figure. It photographs well.

Training your workforce to use AI effectively, redesigning operational processes, helping middle managers figure out which of their responsibilities just got automated — none of this photographs well. There's no press release for "we spent £400,000 on an internal AI upskilling programme and 60% of our sales ops team are now using agentic workflows in their pipeline reviews." That's an internal memo at best, buried somewhere in an HR system nobody reads.

So the money flows to what's legible, and the work nobody photographs gets under-resourced. Nobody's being dishonest about it. The system is just biased towards the bit that shows up on the balance sheet.

The problem is that the quiet, unphotogenic work is where the value actually sits. The AI Daily Brief's Maturity Maps framework scores organisations on six dimensions — deployment depth, systems integration, people, governance, and two others. Seven out of ten functions inside the surveyed organisations are scoring "significantly behind" on the people dimension. Not because those teams are stupid, and not because they don't want to use AI. Because nobody's invested the time to help them figure out how.

The sales team example

The clearest illustration I've seen of what this looks like in practice comes from the most recent sales research. Around 88% of organisations now report regular AI use in at least one business function. Among sales teams specifically, 87% are using AI for prospecting, forecasting or email drafting. These are big, impressive numbers. A CMO showing them to a board would be pleased.

But dig one level deeper. Only about 24% of sales teams have actually got agentic AI — the autonomous, workflow-driving kind — embedded in their revenue pipelines. The rest have it sitting next to their existing workflow, functioning more like a smart chatbot than a process owner. Ask it to draft an email, copy the email out, paste it into the CRM, send it from the CRM.

I wrote last year about why asking AI is already obsolete — the pattern of using AI as a conversation partner while doing all the downstream busywork yourself. That's the pattern these numbers describe. The AI is everywhere. The work hasn't changed.

And this is where the people investment gap starts to bite. Teaching someone to ask ChatGPT better questions is a half-day workshop. Redesigning the revenue workflow so an agent is actually owning a set of sub-tasks, coordinating with Salesforce, handing off to a human at the right moments — that's months of patient, cross-functional work. Process mapping, integration, permissions, monitoring, handling the inevitable edge cases. It's expensive. It's slow. It's also where the actual productivity shift lives.

Companies that have stopped at the workshop version think they've adopted AI. They haven't. They've adopted a better version of the search bar.

The canary: customer service

The signal I'd watch hardest right now is customer service. It's the function that's been pushed to deploy AI most aggressively, often without corresponding investment in people — and it's starting to break in visible ways.

Look at the pattern. Legacy contact centre cuts headcount by 30% based on the theory that AI will handle the volume. AI handles the easy tickets fine. It struggles with the messy ones, which is exactly where the more experienced humans used to step in, except now most of them have left. Customer satisfaction scores sag. Escalation rates climb. The remaining agents burn out. The company either rehires quietly or loses customers. Rinse and repeat across the sector.

None of this is inevitable. Companies that paired the AI deployment with genuine investment in their human agents — new skills, new tooling, new escalation pathways, a proper redefinition of the role from "first-line responder" to "complex case resolver" — are getting different outcomes. Better CSAT, better retention, lower attrition. The tech and the people moving together.

The contact centres that deployed AI without rebuilding around the humans are the canary in the coal mine. They went first, mostly because their work was the most scripted and the easiest to slot an AI into. They're also where the spend-versus-people imbalance hits hardest. If you want to see what "93/7" looks like after eighteen months in production, look there.

What the honest version requires

I don't have a tidy framework to offer here. This piece isn't "seven steps to AI maturity." I'm increasingly suspicious of anyone offering one.

What I do think is roughly true is that the next eighteen months will separate the companies treating AI as a procurement exercise from the ones treating it as an operating-model change. The procurement version is what most organisations are running. Sign the contract. Stand up the infrastructure. Announce the adoption. Move on.

The operating-model version looks very different. It means actually sitting with a team and asking: which parts of your current role should we try to hand to an agent, which should stay with you, and which should we redesign so neither of you is doing them in their current form? It means budgeting for genuine retraining, not a one-hour webinar. It means managers who've actually used the tools enough to help their teams use them. It means HR and Ops and IT in the same room, which most organisations find harder than it sounds.

The Celonis work I wrote about in the agentic readiness gap piece a few weeks back was making roughly the same point from a process angle. 85% of enterprises want to be agentic, 76% admit their operations can't support it, 19% are actually running multi-agent systems. Different data set, same underlying reality. The ambition is miles ahead of the operational capacity to deliver it.

Where I'm not sure this lands

I'll be honest: I'm not sure how much of this is fixable inside the current shape of most large organisations. Some of it almost certainly isn't. The companies best positioned for this transition look different already — flatter, fewer approval layers, cross-functional teams with real authority to rebuild their own workflows. The companies worst positioned look like most large organisations do: siloed, vertical, with transformation work parked in a team that doesn't own any of the ground-level reality it's trying to change.

Nobody really knows where this lands, which is part of the problem. I don't think the 93/7 ratio stays at 93/7 — it would be a weird market where it did. But I don't know whether it closes through organisations waking up and spending more on their people, through the painful route of companies that didn't adjust being overtaken by ones that did, or through some third thing none of us has quite named yet.

What I do know is that if you're running a business right now and you can look at what you've spent on AI over the past twelve months, the ratio is probably closer to 93/7 than you'd like it to be. It's worth doing the maths. And then asking the harder question underneath: which is, if the people side of this is the bottleneck, why haven't we treated it like one?

The technology isn't waiting.