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

There isn't one 70/20/10 rule — there are three different frameworks people confuse. (1) The 70/20/10 learning and development model (70% experiential, 20% social, 10% formal training), attributed to McCall, Lombardo and Eichinger at the Center for Creative Leadership in 1996. (2) Google's '70/20/10' innovation/resource lore (70% core, 20% adjacent, 10% experimental) — business folklore, not a documented methodology. (3) The AI-specific 10/20/70 rule from BCG: 10% algorithm, 20% technology and data, 70% people, process and change management. The BCG version is the one that matters most for AI projects.

Is the 70/20/10 learning model scientifically proven?

It has limited empirical support. The 70/20/10 learning and development model originated with McCall, Lombardo and Eichinger at the Center for Creative Leadership in 1996, based largely on self-reported data from roughly 200 executives in their book The Career Architect Development Planner. It is widely used as a practical rule of thumb in corporate L&D, but the specific 70/20/10 ratio should be treated as a heuristic, not a precise scientific finding.

What is the '30% rule for AI'?

There is no canonical, authoritative source for a '30% rule for AI.' People usually mean one of two things: either Gartner's prediction that around 30% of generative-AI projects will be abandoned after proof of concept by end of 2025 (often mislabelled as a 'rule'), or a loose 30% technology / 70% people heuristic — essentially an inversion of BCG's 10/20/70. If someone cites a '30% rule,' ask which they mean. We define it cleanly below rather than pretend it's an established standard.

Why does the BCG 10/20/70 rule matter for UAE businesses?

Because most AI value in a UAE service business is unlocked by changing how people work, not by buying a better model. A Dubai clinic, salon or restaurant that deploys an AI tool but leaves staff workflows, incentives and training untouched typically sees no measurable return. The 10/20/70 lens forces budget and attention onto the 70% — adoption, process redesign and change management — which is where AI projects most often succeed or fail.

70/20/10 for AI: The Three Rules People Confuse (2026)

Q: Where does the 10/20/70 AI rule come from?

The 10/20/70 split — 10% algorithm, 20% technology and data, 70% people, process and change management — is attributed to BCG, specifically Ernesto Pagano, a Managing Director and Senior Partner at BCG. The core argument is that AI return on investment is bottlenecked by adoption, workflow redesign and change management, not by the model or algorithm itself. Most failed AI projects spend their budget the wrong way round: heavy on the model, light on the people and process work that actually drives results.

There are three different '70/20/10' rules and people mix them up. The L&D learning model, Google's innovation lore, and BCG's 10/20/70 AI rule — clearly disambiguated, with sources.

There is no single "70/20/10 rule for AI." There are three different frameworks that share the same numbers, and people mix them up constantly. The one that actually governs whether your AI project succeeds is BCG's 10/20/70 rule: 10% algorithm, 20% technology and data, 70% people, process and change management.

This post separates the three cleanly, names the source for each, and explains which one to use when. If you have ever been handed a "70/20/10" slide and quietly wondered which version it was — this is the answer.

The Three Rules That Share the Same Numbers

Rule	Split	Field	Source
L&D learning model	70 experiential / 20 social / 10 formal	Employee development	McCall, Lombardo, Eichinger (CCL, 1996)
Google innovation lore	70 core / 20 adjacent / 10 experimental	Resource/innovation budgeting	Business folklore (Google downplayed it)
BCG AI rule	10 algorithm / 20 tech and data / 70 people and process	AI delivery and ROI	BCG / Ernesto Pagano

Notice the third one is inverted. That inversion is the whole point — and the reason confusing it with the others is expensive.

Rule 1: The 70/20/10 Learning and Development Model

This is the oldest and most familiar. It says people develop roughly 70% through on-the-job experience, 20% through others (coaching, mentoring, social learning), and 10% through formal training (courses, classrooms).

The model is attributed to Morgan McCall, Michael Lombardo and Robert Eichinger at the Center for Creative Leadership, articulated around 1996 in The Career Architect Development Planner. It was based largely on self-reported survey data from roughly 200 executives.

Honest caveat: it has limited empirical support beyond that original survey. It is a useful rule of thumb for designing training programmes — don't over-invest in classroom learning — but the precise 70/20/10 ratio is a heuristic, not a proven law. Treat it as a planning prompt, not science.

When this one is relevant to AI: designing how your team learns AI tools. It argues most of that learning happens by using the tools in real work (70%), supported by peers (20%), with formal training as the smallest slice (10%).

Rule 2: Google's 70/20/10 Innovation Lore

The second version — 70% core business, 20% adjacent projects, 10% experimental bets — is associated with Google's early resource-allocation and innovation culture, alongside the famous "20% time."

Treat this one as business lore, not documented methodology. Google itself has downplayed the "20% time" narrative over the years, and the 70/20/10 innovation split was never published as a rigorous framework. It is a useful metaphor for portfolio thinking — keep most resources on the core, reserve some for bets — but don't cite it as if it were an established standard.

When this one is relevant to AI: deciding how much of your AI budget goes to safe, proven automations versus experimental pilots. Useful as a sanity check on portfolio balance, nothing more.

Rule 3: BCG's 10/20/70 — The One That Governs AI ROI

This is the rule that matters most for anyone actually deploying AI. The split is 10% algorithm, 20% technology and data, 70% people, process and change management. It is attributed to BCG, specifically Ernesto Pagano, a Managing Director and Senior Partner at the firm.

The argument is blunt: AI return on investment is bottlenecked by adoption, workflow redesign and change management — not by the model. The algorithm is roughly 10% of the work. The data and technology plumbing is about 20%. The remaining 70% is people and process — and that 70% is where AI projects succeed or fail.

This explains a pattern we see constantly. A business buys an impressive AI tool, integrates it technically, and then... nothing changes. Revenue and cost are unmoved. Why? Because the budget went to the 10% and the 20% — the parts that rarely fail — and skipped the 70% that actually drives results.

For a UAE service business, the 70% looks like: which staff workflows change, who is accountable for using the tool, how they are trained, how incentives shift, and how you measure whether it worked. A Dubai clinic that deploys AI scheduling but never retrains reception staff or changes their process is spending on the 10% and ignoring the 70%.

Own the "30% Rule for AI"

People increasingly ask about a "30% rule for AI." Here is the honest answer: there is no canonical source for it. When someone uses the phrase, they almost always mean one of two things:

A mislabelled Gartner statistic — Gartner predicts around 30% of generative-AI projects will be abandoned after proof of concept by end of 2025. That is a forecast of abandonment, not a budgeting rule.
A loose 30/70 heuristic — roughly 30% technology, 70% people. This is essentially an inversion of BCG's 10/20/70, collapsing "algorithm + tech and data" into a single ~30% technology bucket.

So if you want a clean, defensible definition: the "30% rule for AI" is best understood as a simplification of the 10/20/70 principle — roughly 30% of effort on technology and data, 70% on people and process. Use it as shorthand, but cite BCG's 10/20/70 as the real source, and never present "30%" as if it came from a standards body. It doesn't.

How to Apply This to a Real AI Project

The practical move is simple: stop investing like the algorithm is the hard part. It isn't.

Decide which "70/20/10" you mean before you quote it.
For delivery, use BCG's 10/20/70 — put 70% of effort into people and process.
Audit your current AI spend. If over 30% is going to models and tooling, you are over-invested in the part that rarely fails.
Write the 70% plan explicitly: workflows, owners, training, incentives, success metric.
Set a kill-switch — the condition under which you stop and fix adoption before spending more.

That last point matters. Most AI projects don't die from bad models. They die from low adoption nobody planned for.

Where This Fits in AI Consulting

A good AI consultant spends most of their time on the 70%, not the 10%. The model selection is often the easy, fast decision. The hard work — and the work that determines ROI — is redesigning how people actually do the job around the tool. If a consultant is selling you mostly model and tooling, they're selling you the 30% and leaving the part that fails to you.

We go deeper on this in the AI Consulting in Dubai pillar guide and in Why Do AI Projects Fail?, which puts hard numbers on the failure rates. To discuss how the 70% applies to your specific operation, book an AI consultation via evolvxai.com, or read more about how we work.

Sources

McCall, Lombardo & Eichinger / Center for Creative Leadership — 70/20/10 learning and development model (1996): en.wikipedia.org/wiki/70/20/10_model_(learning_and_development)
BCG / Ernesto Pagano — 10/20/70 AI rule (people-centric AI transformation): pmi.org/blog/ai-transformation-people-insights-bcg and artefact.com
Gartner — 30% of generative-AI projects abandoned after POC by end of 2025: gartner.com