The BCG Math: Why 70% of AI ROI Comes From People, Not Algorithms

Ten percent on algorithms. Seventy percent on people and process. That split, not the model you picked, is what BCG's research says actually predicts AI ROI.

This is BCG's 2026 research on companies that got real return from AI, not pilots that looked good in a slide deck. If you're deciding where the next AI dollar goes, the honest answer might be: not on a better model.

What is BCG's 10-20-70 rule for AI ROI?

It's a finding about where AI budget and effort actually produce return, not a technology recommendation. Top-performing AI adopters allocate roughly 10% of effort to algorithms, 20% to data and technology, and 70% to people, process, and cultural change (BCG, 2025). Companies that hold that ratio see about 2.1 times greater ROI than peers who don't.

Leaders also go narrower before they go wide. An average of 3.5 use cases in production, against 6.1 for everyone else (BCG, 2025). Depth beats breadth.

Why do most companies get the ratio backwards?

Most organizations spend close to 70% of their AI budget on algorithms and technology, and something near 10% on the process and change-management work that actually determines whether the tool sticks. It's the 10-20-70 rule, flipped.

That inversion has a direct cost. About 70% of AI projects fail to deliver the ROI they were pitched on (BCG, 2025). The primary cause is organizational resistance and process mismatch, not model quality. The algorithm usually works fine. The organization around it never changed to use it.

What does "70% people and process" actually mean in practice?

Not a training budget line. It's the work of redesigning how a role or workflow operates once AI does part of it: who owns the decision point, what the human still reviews, which step gets cut outright.

Three things tend to sit inside that 70% for the companies that get it right:

• Reworking the process itself, not bolting a tool onto the process that already existed
• Upskilling the people whose job actually changes shape, not just the people who'll click the new button
• Governance and measurement built in from day one, so a pilot has a baseline to prove itself against

The case against this framing: a 10-20-70 split sounds like it argues against investing in better models at all, and that's not quite right. Algorithms being only 10% of the effort doesn't mean they're 10% of the risk. Get the model badly wrong and no amount of change management saves the pilot. The rule isn't "ignore the tech." It's "don't let the tech be the only thing you fund."

How should a company decide where to spend first?

Pick fewer use cases. Fund the 70% properly on each one instead of spreading the inverted ratio across more pilots. BCG's own numbers back this directly: 3.5 focused use cases beat 6.1 scattered ones (BCG, 2025).

A short test before funding any AI use case:

1. Is there a named owner for the redesigned process, not just for the tool?
2. Is there a baseline metric this pilot has to beat, agreed before it starts?
3. Have you decided, on purpose, which parts of the workflow stay human?
4. Is there an actual change-management plan, or just a training deck?

If more than one answer is no, the 70% isn't funded yet. Whatever the tech line item says.

What separates the top performers getting 2.1x ROI?

Not better models. Better sequencing. The companies BCG classifies as leaders treat algorithms as the easy 10% (the part that was rarely the real bottleneck) and put scarce budget and attention where transformation actually happens. And they measure fewer things more rigorously instead of running more pilots loosely.

The 2.1x isn't a technology gap. It's a discipline gap. I'd bet on the company with 3 well-governed use cases over the one with 8 ungoverned ones every time, and I'd want to see the baseline metrics before I believed either number.

If your experience runs the other way, tell me what broke the pattern for you. I'd genuinely like to know where 10-20-70 stops holding.

Frequently asked questions

What is BCG's 10-20-70 rule for AI?

A budget and effort allocation framework: 10% on algorithms, 20% on data and technology, 70% on people, process, and cultural change, based on what top AI performers actually do.

Why do most companies get AI budget allocation wrong?

They invert the ratio, over-funding algorithms and technology while under-funding the process redesign and change management that determines whether AI actually sticks.

How much more ROI do 10-20-70 companies see?

Roughly 2.1 times greater ROI than companies that over-index on the technology layer, per BCG's research.

Key takeaways

• BCG's 10-20-70 rule: 10% effort on algorithms, 20% on data/technology, 70% on people and process. Companies that hold this ratio see 2.1x greater ROI.
• Most companies invert it, funding technology heavily and process work as an afterthought. That's a major reason 70% of AI projects miss their expected ROI.
• The 70% isn't a training budget. It's redesigning who owns the decision, what the human reviews, and what gets cut.
• Leaders go narrower before wider: 3.5 use cases in production versus 6.1 for everyone else.
• Before funding an AI pilot: name an owner, set a baseline, decide what stays human, and build a real change-management plan, not just a training deck.