Contents7
You have probably read it in a slide deck, an article or a consulting offer: in an AI project, 70% of the difficulties come from people and processes, 20% from technology and only 10% from algorithms. The sentence looks like a study finding. It is repeated as one, including by us: our article on AI return on investment presents it as a recommendation, which is what it is.
It is not a measurement. It is an in-house rule of Boston Consulting Group, a firm that sells AI consulting, based on its experience of engagements with large companies, far from our audience. Its story shows how a conviction becomes a statistic. And why, even so, the idea is worth keeping.
In short
The “10-20-70” rule comes from BCG, which bases it on “our experience”, not on a survey.
A 2024 press release presented it as what a survey “reveals”; the report itself only measured how often difficulties were cited.
BCG has worded it six ways since, and its own institute even reversed it in July 2026.
The only recent measured figure: the most advanced companies put nearly three quarters of their priorities into people, organisation and processes.
The idea holds: other studies place the obstacles on the side of organisation and data, not models.
Where the rule comes from
BCG’s October 2024 report, “Where’s the Value in AI?”, itself calls it “BCG’s 10-20-70 model”. It writes that “our experience, corroborated by our new research” indicates that about 70% of the challenges relate to people and process. The key word is “experience”: the split comes from consulting work, and the survey of 1,000 executives is meant to confirm it.
Yet what the survey measures, on page 14, is something else: the share of respondents citing each difficulty (establishing a return on investment, integrating with existing systems, getting reliable models). Those difficulties are then sorted into the model’s three boxes. Sorting answers into an existing frame does not measure the weight of each box.
The press release issued the same day takes the last step: “the survey reveals” that about 70% of the challenges stem from people and processes. That shorter, more quotable sentence is the one that travelled.
Six wordings in three years
A rule that measured something would keep the same subject. This one changes subject from one medium to the next.
- October 2024
In the report, 10% of resources to algorithms, 20% to technology and data, 70% to people and processes; in the press release, 70% of the challenges.
- January 2025
A BCG executive speaks of 70% of efforts devoted to people, processes and culture.
- September 2025
The annual report speaks of 70% of “strategic focus”.
- July 2026
BCG’s research institute gives 10% to technology and 20% to algorithms and data, the reverse of every other version.
- September 2026
The annual report returns to effort, with 10% for algorithms.
- BCG offer page
70% of the transformation’s value, drawn from “thousands” of engagements.
Resources, challenges, efforts, focus, value: five different subjects for the same three numbers, plus one version where two of them swap places. The sources can all be read online: the 2025 report, the January 2025 press release, the 2026 page, the institute’s article and the offer page. None describes a sample for the rule itself.
How a rule becomes a statistic
The mechanism is neither specific to BCG nor dishonest. A few retellings are enough.
The journey of an in-house rule
- Experience
Years of engagements produce a conviction.
- The model
The conviction takes round, memorable numbers.
- The press release
“the survey reveals” replaces “our experience”.
- The retelling
The sentence circulates without its source, and becomes a fact.
At each step the sentence loses a caveat and gains authority. A conference quotes the press release, an article quotes the conference, an offer quotes the article. Three years later, people read that “70% of AI failures come from people” the way they would read an unemployment rate.
Watch out
The reflex to keep
A round number (10, 20, 70, 80/20) that recurs from one medium to another with no sample and no fieldwork date is almost always a working rule. It can be useful; it is not quoted as a measurement.
What is actually measured
BCG’s September 2026 report contains one measured figure that supports the rule, from 1,330 executives, with more than half of the companies above 5 billion dollars in revenue. It concerns the priorities set by the companies BCG considers most advanced.
of their AI priorities concern people, organisation and processes
concern technology
concern algorithms
It is self-reported, and it is a correlation: nothing proves that this split explains their lead. But it is a figure, with a sample, a page and a date. It can be quoted as such.
Why the idea still holds
Other studies, by other sponsors, describe the obstacles the same way: rarely the model, almost always organisation, data and integration.
- In France, KPMG, which sells audit, consulting and AI services, lists as barriers to scaling the complexity within business functions (39%), technical skills (38%) and integration with existing systems (36%), among companies with at least 100 million dollars in revenue, far from our audience.
- In the US and Canadian “middle market”, RSM, which sells audit and consulting, puts data quality first (34%); when pilots failed, it rises to 53%, and integration to 47%. That market starts at 30 million dollars in revenue.
Our own experience points the same way, on a small scale. On the translation pipeline we run for Palmora Property, 95% of segments come out of the cache after the first pass, without calling the AI again (project CNT.7). That figure owes nothing to the model: it comes from an organisational decision, giving each segment a fingerprint and retranslating only what has changed.
How to use it
Keep the idea, not the decimals. For a team of 20 to 500 people, the rule translates into budget questions more than percentages.
- Count the team’s time.
Training, rewriting procedures, reviewing outputs: it is the line people forget, and often the largest.
- Prepare the data before the tool.
A poorly kept CRM gives AI false raw material.
- Plan the integration.
A tool that does not talk to the rest of the system ends in copy and paste.
- Choose the model last.
It is the easiest choice to change, and often the cheapest.
Key takeaway
Our position, signed Jim
The 70/20/10 rule is good advice and a bad figure. We now quote it as “BCG’s rule, drawn from its experience”, never as a survey finding. We would change our minds if a study published the sample and method that measure this split.
To read the other figures in circulation, our article on studies that contradict each other about AI profitability applies the same grid. And to split the effort of a first project before buying anything, that is the purpose of our AI Audit Sprint and roadmap.
Frequently asked questions
What is BCG’s 10-20-70 rule?
A rule for splitting effort in an AI project: 10% for algorithms, 20% for technology and data, 70% for people, organisation and processes. BCG draws it from its consulting experience, not from a measurement.
Is it true that 70% of AI failures come from people?
No published study measures it. The figure comes from a 2024 press release that presented BCG’s in-house rule as a survey finding. Other studies do confirm, however, that the most cited obstacles are organisation, data and integration.
Why do people say 70/20/10 or 10-20-70?
It is the same rule, read one way or the other. BCG writes “10-20-70”; the most common retelling puts 70 first. What the 10% and the 20% cover also varies between sources.
Should you still follow this rule?
The idea, yes: plan for the team’s time, the data and the integration before choosing a model. The exact percentages, no: they are orders of magnitude, not a target to hit.
Sources
- “Where’s the Value in AI?” (new tab)web-assets.bcg.com
- Press release issued the same day (new tab)bcg.com
- The 2025 report (new tab)media-publications.bcg.com
- The January 2025 press release (new tab)bcg.com
- The 2026 page (new tab)bcg.com
- The institute’s article (new tab)bcg.com
- The offer page (new tab)bcg.com
- KPMG (new tab)kpmg.com
- RSM (new tab)rsmus.com
Jim · founder of Brain Plus AI
Also in business ai
Stepping back11 min
AI return on investment: why it is taking so long
The figure of 95% of AI projects with no return went round the world. What it really measures, why the other studies say roughly the same thing, and the practical lessons to draw from them for an SME.
Stepping back8 min
Half of large companies profit from AI. What about the rest?
One study says nearly half of large companies finally get value from AI, another finds only 6% of leaders in the accounts, a third that 56% of CEOs see no significant benefit. Who should you believe, and what is a survey worth when everyone describes their own results?
Stepping back10 min
The human role alongside AI: knowing how to judge
AI excels at some tasks and is confidently wrong on neighbouring ones. The human's value shifts: knowing where the line runs, keeping your judgement, answering for the result.