Around ninety-five percent of enterprise AI initiatives fail to deliver measurable ROI. That’s the headline finding from MIT NANDA’s 2025 GenAI Divide: State of AI in Business report, and it has been making the rounds for months now – usually wrapped in concerned hand-wringing about wasted budgets, stalled pilots, and disillusioned boards.

I read it differently. I’ve been writing about remote and distributed work since the early 2010s, and watching how organisations of different sizes adopt new tools is one of those things you can’t unsee once you’ve noticed the pattern. Ninety-five percent of enterprise AI failing isn’t a problem for the rest of us. It’s the structural arbitrage of our era.

If you run a one-person business, that statistic is the loudest signal you’ll get this decade that the field is wide open. Let me explain why.

Why enterprise AI keeps failing

The MIT NANDA report goes into the specifics, but the broad strokes are familiar to anyone who has ever sat through a procurement meeting, and further studies in 2026 bear this out ongoingly. Enterprise AI fails for reasons that have very little to do with the technology itself.

Procurement cycles are slow by design. A mid-sized company doesn’t buy a new tool the way you or I do. There’s a vendor evaluation process. A security review. A legal review of the terms of service. A data-protection impact assessment, especially in Europe under GDPR. By the time a single AI workflow has been approved, the underlying model has had two version upgrades and a competitor has launched something better.

Risk and compliance committees say no, because saying no is safe. A CISO who blocks a tool gets praised when nothing breaks. A CISO who approves a tool gets blamed when something does. The incentive structure is asymmetric, and it has been that way for decades. AI tools that touch customer data, financial data, or anything regulated face a wall of caution that’s rational for the individuals involved and disastrous for the organisation as a whole. “No one ever got fired for buying IBM” - but buying IBM is a strategy from a different era, when the world moved at the pace of corporate procurement cycles, see above.

Middle management resists, quietly. This is the one nobody puts on a slide. If your job is managing twelve people who do work that AI could now do in a fraction of the time, you have a strong personal interest in the AI rollout being slow, partial, and surrounded by enough caveats and cautions that nobody seriously evaluates whether your team should still exist at its current size. Change-management theatre is often the formal expression of this resistance. We’ve seen this in the remote work transition, in moving to cloud over on-prem, and in so many other scenarios – it’s just the same fear-as-friction expressed against a new threat.

Legacy systems don’t bend. Enterprise software stacks are decades old in places. Integration with the existing CRM, ERP, ticketing system, document management system, and bespoke internal tools that nobody fully understands anymore is the bit that breaks every transformation programme. AI is no different.

None of this is snark. These are rational behaviours from rational people inside organisations that were not designed for the speed at which AI tooling now moves. The result is that enterprise AI initiatives stall, get watered down, or quietly die.

Why solopreneurs win this round

Now invert every one of those failure modes.

A solopreneur has no procurement cycle. I decide on Tuesday morning that a new tool looks promising, I sign up that afternoon, and I’m using it in a real workflow by Wednesday. The decision-to-execution time is measured in hours, sometimes minutes.

There’s no risk and compliance committee, because there’s me. I am the CISO. I am also the implementer, the legal department, and the person who decides what risk is acceptable. That’s not a recipe for recklessness – I take data protection seriously, and I read the terms of service – but it collapses a six-month process into an afternoon of due diligence.

There’s no middle manager defending their headcount, because there is no headcount. The person whose work might be transformed by AI is the same person deciding whether to adopt it. The political resistance simply has nowhere to live.

And the legacy systems problem doesn’t apply, because solopreneurs don’t have to retrofit AI into existing workflows. We can build new workflows from scratch with AI as a foundational layer. My content production system, my client intake, my project tracking – none of these were designed before AI existed and then bolted onto it. They were designed with AI in mind, which is a completely different kind of system. And thanks to AI I can redesign them in half an hour if I need to

This is the asymmetry that the MIT report quantifies without naming it. The same structural conditions that make enterprises slow make one-person businesses fast. The friction that kills corporate AI rollouts simply isn’t there for us.

What this looks like in 2026, in practice

Abstract claims are cheap, so let me get specific. These are composite examples drawn from people I’ve spoken to over the past year, with details changed.

A freelance content writer rebuilt her entire production pipeline around Claude in a single working week. Research, outline, draft, edit, repurpose for social. What used to be a three-day process per article now takes a day, and the quality is higher because she spends the saved time on the bits AI genuinely can’t do – the interviews, the original opinions, the decisions about what’s worth saying in the first place.

A freelance bookkeeper in the Netherlands set up an AI-assisted client intake process. New client uploads their messy receipts and bank statements; an AI workflow extracts, categorises, and queues the awkward bits for human review. She’s cut admin time by around sixty percent, which means she can take on more clients without working more hours, or work fewer hours at the same revenue. Either way, she chooses.

A single-handed marketing agency owner in Lisbon orchestrates five specialised AI workflows that together do the work she used to coordinate across a team of four contractors. She’s not pretending the AI does it all – the strategy and client relationships are still hers – but the execution layer that used to require constant coordination now runs on prompts and templates she refines weekly.

None of these people are doing anything that an enterprise team couldn’t, in theory, do. The difference is that they actually did it, in weeks rather than quarters, and they kept the productivity gains for themselves rather than watching them disappear into reorganisation overhead.

The window matters

Here’s the part I want to be honest about, because it’s where solopreneur cheerleading often goes wrong.

This advantage is not permanent. Enterprise AI tooling will mature. Vendors will build products specifically designed to slip through procurement and compliance reviews, and they have big teams dedicated to managing their entire sales pipeline. Some of the friction that’s currently slowing big organisations down will reduce, even if it never disappears, so the gap will close.

I’d guess we have eighteen to twenty-four months where the asymmetry is at its widest. After that, well-resourced enterprises start catching up, not because they get faster but because the tools meet them halfway. The solopreneurs who used this window to compound their advantages – building digital assets, learning to use the tools, refining workflows, pricing on the value they deliver rather than the hours they bill – will be in a much stronger position than those who waited for the perfect tool or the definitive guide.

I’d also be cautious about overclaiming. There are things AI still doesn’t do well, and some of them are exactly the things solopreneurs need most. Genuine relationship-building. Original thinking that goes beyond pattern recognition, your experiential learning that will never be in any training data. The judgement that only comes from having been wrong about something interesting. AI is a force multiplier on the work you bring to it, not a substitute for the work itself. If you don’t have your own thinking to multiply, AI just helps you produce more confident-sounding nonsense at scale.

What to do this week

If you’re running a one-person business and you’ve been watching enterprise AI conferences from the sidelines waiting for the dust to settle, my honest advice is to stop waiting.

Stop waiting for the perfect tool. The tool you have access to today, used in a real workflow this week, will teach you more than six months of reading about AI strategy.

Stop benchmarking yourself against enterprise rollouts. They’re solving a different problem with different constraints. The frameworks that apply to a thousand-person company actively mislead you.

Pick one workflow this week – a real one you do regularly – and rebuild it with AI as a foundational layer rather than a polish-it-at-the-end add-on. Notice what changes. Notice what doesn’t. Iterate the following week.

The MIT NANDA report is going to be quoted by enterprise consultants for the next year as evidence that AI is overhyped. They’re reading it wrong. It’s evidence that the current advantage belongs to the small, the fast, and the unencumbered. That’s us.

The window is open. It won’t stay open forever. Build now!