What Is the 10-20-70 Rule for AI?

If your AI rollout isn't generating the ROI you expected, there's a good chance the problem isn't your technology. It's how you allocated the investment around it.

The 10-20-70 principle — documented by BCG in their 2025 AI Radar report and validated across hundreds of enterprise AI deployments — describes the resource split that consistently separates AI leaders from organizations still waiting for their investment to pay off.

The breakdown

Most organizations do this backwards. They spend 80–90% of their AI budget on technology — vendor contracts, implementation, infrastructure — and treat people and process as an afterthought. A training session. A FAQ page. An email from IT.

Then they wonder why adoption is at 20%.

Where it comes from

The 10-20-70 framework was surfaced prominently in BCG's research on high-performing AI organizations. The finding wasn't theoretical — it came from analyzing what top quartile companies actually did differently when deploying AI at scale. The split held consistently: organizations that allocated the majority of resources to people and process transformation generated measurably higher ROI than those that didn't.

"AI doesn't fail at the algorithm level. It fails at the adoption level. The 70% is where failure happens — and where the fix lives."

This tracks with what every change management practitioner has known for decades: technology adoption is a human problem. PROSCI's research across 50,000+ change initiatives found that projects with excellent change management are six times more likely to meet objectives than those with poor change management. AI is not the exception.

What the 70% actually covers

When organizations hear "70% on people," they assume it means more training sessions. It doesn't. The 70% encompasses the full scope of what it takes to change how an organization works:

• Awareness building — making sure employees understand what AI is being deployed, why, and what it means for their role before it lands on their desk
• Desire creation — addressing fear, skepticism, and resistance with a compelling case for why adoption benefits them personally, not just the organization
• Knowledge transfer — role-based training that goes beyond tool mechanics into actual workflow integration and judgment calls
• Ability development — giving employees practice, feedback, and support until the new behavior is actually habitual
• Reinforcement systems — measurement, accountability structures, and ongoing coaching that prevent reversion to old habits

These five elements map directly to PROSCI's ADKAR model — and they're not optional. Skip any one of them and you get a gap. Gaps compound. A workforce that never understood why the tool matters (Awareness gap) won't develop the desire to use it. A workforce that wants to use it but was never shown how to integrate it into their actual work (Ability gap) won't sustain adoption past week three.

For the full ADKAR breakdown applied to AI, read ADKAR for AI: A Change Framework Built for What's Coming.

Why most organizations get the ratio wrong

The 10-20-70 ratio feels counterintuitive because AI is sold as a technology product. Vendors pitch the algorithm. Procurement buys the platform. IT owns the implementation. By the time anyone thinks about change management, the budget is spent and the rollout date is fixed.

This is the structural problem. Technology investment is visible — it shows up in contracts, headcount, and infrastructure spend. People investment is diffuse — it looks like training time, manager bandwidth, and communication planning. It's easy to cut. And it's the first thing cut when timelines slip.

The result: AI pilots that succeed and rollouts that fail. Pilots have high people investment by accident — small teams, direct access to support, visible accountability. Scale removes all of it. The 70% is what replaces it at scale.

How to apply it

You don't need to retrofit a perfect 10-20-70 split onto a deployment that's already underway. What you need is an honest audit of where your investment is actually going — and a plan to close the gap.

In practice, that means answering three questions before your next AI rollout:

• Do your employees know this is coming and why? Not from a company-wide email — from their manager, with a clear message about what changes for them specifically.
• Have you mapped resistance? Who is most likely to push back, and what are their actual objections? Fear of job loss, distrust of the output, lack of confidence in the tool — each one has a different intervention.
• What does reinforcement look like 90 days post-launch? If the answer is nothing, adoption will regress. It always does without a system to sustain it.

If those questions don't have solid answers, the 70% is where to start — not after the technology is deployed, but before it is.