What Is the 10-20-70 Rule for AI? (Explained Simply, 2026)

If you’ve researched why AI initiatives succeed or fail, you’ve probably hit the “10-20-70 rule.” It’s one of the most useful mental models in all of AI strategy — and one of the most ignored in practice. In one line: the model is the smallest part of AI success. Here’s what the rule means, why it predicts which projects deliver value, and how to apply it so yours is one of them.

What the 10-20-70 rule means

Popularized by Boston Consulting Group (BCG), the 10-20-70 rule says successful AI transformation breaks down as 10% algorithms, 20% technology, and 70% people and process. It’s often pictured as an iceberg: the algorithms and models everyone talks about are the visible tip, while the massive submerged 70% — processes, people, and culture — is what actually determines whether a transformation floats or sinks. (You’ll sometimes see it written “70-20-10”; same idea, reversed order.)

The three parts explained

The 10% — algorithms

This is the model itself: the choice of architecture, the fine-tuning approach, the evaluation methodology. It captures headlines and imaginations — but it’s almost never what determines whether a project succeeds. A powerful model is the engine; an engine alone doesn’t win the race. Important for performance, rarely decisive for outcome.

The 20% — technology & infrastructure

The systems around the model: data pipelines, integration, deployment, monitoring, MLOps. This is where most engineering effort and procurement budget goes. It’s the foundation and the fuel — necessary, but still not where business outcomes actually come from. Good tech keeps the engine running; it doesn’t, by itself, move the metric.

The 70% — people & process

The largest and most-ignored share: redesigned workflows, training, adoption design, trust calibration, decision authority, change management, and incentive alignment. This is the slow, structural work of getting humans to use AI in ways that move the business. It can’t be bought off a shelf or fully outsourced — a consultancy can hand you a model, but not the cultural change that gets your team to trust and use it.

Why most projects invert it

Here’s the trap: most AI programs spend ~70% of attention on the algorithm and technology and treat people and process as a footnote — the exact opposite of what works. Why? Because the 30% is easier to measure (accuracy is a number; adoption is messy) and easier to outsource. So teams default to what’s measurable and buyable, declare technical victory, and discover six months later that the metric never moved because the 70% was never built.

Apply it to your rollout

Flip your plan to match reality. Before launching, ask: who will use this, how will their workflow change, who trains them, and how will we drive adoption? Fund that work from the start. Set a pre-AI baseline so you can prove impact, and treat workforce readiness as a first-class deliverable — not a footnote after the model ships.

The iceberg, visualized

Where AI value actually comes fromWhere AI value actually comes fromPeople & process (70%)70%Technology & data (20%)20%Algorithms / model (10%)10%

Figure 1: the model is just 10% of what drives AI success — the human and process layer is 70%.

Why most projects get it backwards

The reason this rule matters so much is that most organizations do the opposite of what it prescribes. They fund the 10% (models) and the 20% (infrastructure) generously, then underfund or ignore the 70% (people and process). The predictable result: working AI systems with poor adoption, marginal business impact, and slow or absent ROI. The technology was fine; the 70% was missing.

This connects directly to the grim statistics on AI project failure — the finding that only a small fraction of AI initiatives achieve real ROI, and that the common thread in failures is inadequate investment in people and process change, not bad technology. (We cover this in depth in Why Do 85% of AI Projects Fail?.) The 10-20-70 rule is essentially the positive version of that lesson: it tells you where to put your effort to land in the successful minority.

Want the failure side of this story?Read our breakdown of why most AI projects fail — and how to beat the odds.

Learn more →

A real example of the rule in action

Consider two companies deploying the same AI coding assistant to their engineering teams. The first treats it as an IT project: they pick the model, wire up the infrastructure, hand out licenses, and announce success. Adoption stalls at a fraction of the team, the workflow never changes, and six months later nobody can point to a measurable gain. The second company spends the same on technology but also redesigns how code review and ticket triage work around the tool, runs hands-on training, names internal champions, and tracks cycle time before and after. Same model, same infrastructure — wildly different outcome. The difference was entirely in the 70%. This is the pattern researchers see over and over: the organizations capturing real value aren’t the ones with the best model, they’re the ones that did the unglamorous people-and-process work everyone else skipped.

That’s also why the rule is sometimes called the iceberg principle. The algorithms and dashboards are the part executives can see and show off; the workflow changes, trust-building, and behavior shifts are submerged and unglamorous — yet they’re the mass that keeps the whole thing stable. Skip them, and even a brilliant model simply sits idle.

How to apply the 10-20-70 rule

Turning the rule into action is straightforward once you accept its premise:

Rebalance your budget and attention. If 90% of your plan is model and infrastructure, you’re set up to fail. Deliberately resource the people-and-process work.
Start with the workflow, not the model. Map exactly how the AI changes someone’s day before you build anything.
Invest in adoption. Training, champion networks, trust-building, and clear decision authority are the levers that convert a working model into business value.
Measure a pre-AI baseline. You can’t prove impact (or justify the project at budget review) without knowing where you started.
Don’t bolt the 70% on at the end. The people-and-process work has to be built alongside the technology, not added after the model ships.

The encouraging part: the 70% isn’t mysterious or reserved for tech giants. Clear metrics, redesigned workflows, sustained leadership, and an AI-literate team are disciplines any organization can adopt — and they’re exactly what separates the projects that deliver from the ones that quietly get shelved.

Frequently asked questions

What is the 10-20-70 rule for AI?

Popularized by BCG, it says AI success is 10% algorithms (the model), 20% technology (data and infrastructure), and 70% people and process (workflows, training, change management). The human side is the biggest driver of value.

Why is the 70% so important?

That’s where value is captured — redesigned workflows, trained people, adoption. A great model with poor adoption delivers nothing, and most AI failures stem from people-and-process issues, not technology.

Is it 10-20-70 or 70-20-10?

Both refer to the same idea; the order just depends on which end you list first. The proportions — 10% model, 20% infrastructure, 70% people/process — are what matter.

How do I apply the 10-20-70 rule?

Don’t pour 90% of effort into model and infrastructure. Invest deliberately in workflow redesign, training, change management, and adoption from the start, and measure a pre-AI baseline to prove impact.

The OneAppleFall Team

We independently test every AI agent and tool we review — on our own dime, on real work. We never accept payment for a score, and we disclose affiliate links clearly. Read our review methodology →