The 70/20/10 Rule: Why AI Success Is Mostly Organisational, Not Technical

The 70/20/10 Rule: Why AI Success Is Mostly Organisational, Not Technical

Here is a number that should reframe how you think about every AI decision your business makes. The resource split that consistently predicts whether an AI programme succeeds is roughly 10 per cent algorithms, 20 per cent data and infrastructure, and 70 per cent people and process.

Sit with that for a moment. The part everyone obsesses over, the model, the tool, the shiny capability, is the smallest slice. The part almost nobody wants to fund, the organisational change, is by far the largest. And organisations that invert this ratio, pouring their money and attention into technology while neglecting people and process, fail with remarkable consistency.

This is the single most important and least understood idea in enterprise AI today.

Why everyone gets the ratio backwards

The reason is entirely human. AI tools are tangible, purchasable, and demonstrable. You can see a demo, sign a contract, and point to something concrete. Organisational change, by contrast, is slow, messy, and unglamorous. It does not demo well in a board meeting.

So teams do the easy, visible thing. They buy the technology, announce the initiative, and quietly assume the hard part will sort itself out. It does not. The model gets deployed onto unchanged workflows, the people around it carry on exactly as before, and six months later the programme is quietly shelved as a "learning."

As McKinsey put it, AI success is "20 per cent algorithms and 80 per cent organisational rewiring." The framing is slightly different from the 70/20/10 split, but the message is identical: the technology is the easy part. The transformation is the work.

The evidence is hard to argue with

This is not a motivational slogan. It shows up clearly in the data on who actually captures value.

Deloitte's research separates organisations into those that merely automate, bolting AI onto existing operations, and those that genuinely transform, upgrading the underlying ways of working in parallel. The transformers reported the strongest AI returns, around 72 per cent seeing strong ROI, outpacing the automators by several percentage points. The lesson is direct: AI value emerges when the surrounding system is rebuilt alongside it, not when AI is constrained by legacy processes and old habits.

The failure research points the same way. MIT's analysis of enterprise AI found that the primary reason pilots stall is poor integration with existing workflows, not any shortcoming of the models themselves. The technology was rarely the problem. The organisation around it almost always was.

What the 70 per cent actually involves

If most of the work is people and process, it helps to be concrete about what that means in practice. The 70 per cent includes:

Workflow redesign. Rebuilding the actual sequence of how work gets done so that AI is woven into it, rather than awkwardly attached to the side of an unchanged process.

Change management. Helping the people whose jobs are affected understand what is changing, why, and what is in it for them. This is where adoption is won or lost.

Training and enablement. Building genuine fluency, not just granting access. The gap between "everyone has the tool" and "everyone uses it well" is almost entirely training, champions, and clear use cases.

Governance and accountability. Deciding what AI is allowed to do, who is responsible for outcomes, and how decisions are checked and audited. This is what turns a risky experiment into a dependable system.

Role evolution. Redefining jobs as AI absorbs the routine parts, so people move up toward judgement, exceptions, and relationships rather than being left with a hollowed-out role.

None of this is technical. All of it is decisive.

How to rebalance

If your AI budget and attention are skewed heavily toward technology, here is how to correct the tilt.

Start by being honest about your current split. Most organisations discover they are spending the overwhelming majority of their effort on tools and almost nothing on change. Naming that is the first step.

Then resource the 70 per cent deliberately. Assign real ownership to workflow redesign and change management, not as a side task for someone already stretched, but as a core part of the programme with its own budget and accountability.

Fix the data layer, your 20 per cent, before you scale. Siloed, messy, or inaccessible data quietly undermines even the best model.

And keep the technology decision in proportion. Choosing the model matters, but it is a 10 per cent decision. Spend your energy where the returns actually live.

The takeaway

The businesses winning with AI are not the ones with the best technology. In a world where capable models are available to everyone, the model is no longer the differentiator. The differentiator is the organisational work: rewiring how the business actually operates so that the technology has something solid to plug into.

That is genuinely hard, which is exactly why it is valuable. Anyone can buy the tool. Very few do the rewiring. The ones that do are pulling away.

This is the heart of what Argonix does for mid-market businesses across APAC and the US. We treat AI as an operating-model decision, not a software purchase, focusing on the 90 per cent that actually decides the outcome: the data, the workflows, the people, and the process. The technology is the easy part. We make sure the rest is in place.

If your AI effort is heavy on tools and light on transformation, that imbalance is worth addressing before you spend another dollar on technology.

Sources: McKinsey "The State of AI," 2025; Deloitte "AI maturity and digital value" and "State of AI in the Enterprise," 2025 to 2026; MIT NANDA Initiative research on enterprise AI, 2025.