AI Didn't Invent the 70% Failure Rate — BCG Found It First

Walk into almost any executive briefing on AI right now and you'll hear some version of the same alarmed question: "Why is this so hard? The technology is incredible — why can't we get it adopted?" It's asked like the difficulty is new. Like generative AI broke something that used to work fine.

It didn't. In October 2020 — before ChatGPT existed, before "AI adoption" was a line item in anyone's budget — BCG researchers Patrick Forth, Tom Reichert, Romain de Laubier, and Saibal Chakraborty published a finding that should be sitting on every AI steering committee's desk today: 70% of digital transformations fail to meet their objectives. Not AI transformations. Digital transformations, broadly — ERP rollouts, cloud migrations, CRM overhauls, automation programs. The whole category.

The number hasn't moved much since. What's changed is the technology everyone's pointing at when they ask why.

The diagnosis was already written down

Here's the part of the BCG research that gets skipped when people cite the 70% figure as a scare stat: the report doesn't blame the technology. It's explicit about where the failure actually lives.

"The technology is important, but the people dimension (organization, operating model, processes, and culture) is usually the determining factor. Organizational inertia from deeply rooted behaviors is a big impediment." — Forth et al., BCG (2020)

Read that quote again, and notice what's missing: any mention of the tool. No CRM, no cloud platform, no algorithm gets named as the culprit. The determining factor was always the people dimension — organization design, operating model, process, culture. That's not an AI-era insight. That's a five-year-old insight that AI happens to be the newest test case for.

Why this history matters more than the newest stat

I write about a lot of individual AI failure statistics on this site — why 42% of AI projects never make it to production, why pilots stall at rollout, why data-readiness predictions are really governance predictions in disguise. Each one is true and each one is useful. But taken individually, they can accidentally reinforce the wrong narrative: that AI is uniquely, unprecedentedly difficult to adopt, and that the failure rate proves something is different this time.

The BCG 2020 data says otherwise. It says the 70% failure rate predates AI. It's the same number you'd have found for cloud migrations in 2018, ERP rollouts in 2012, and CRM implementations in the 2000s. The common thread across every one of those failure waves was never the specific technology stack. It was the same missing ingredient every time: an organization that treated transformation as a technical delivery project instead of a change management program with a technical component.

That reframe matters practically, not just historically. If AI adoption were a genuinely novel problem, you'd need genuinely novel tools to solve it. It isn't, so you don't. The same discipline that has driven successful transformations for two decades — sponsorship, structured communication, resistance management, adoption measurement — is exactly what closes the AI gap too. You don't need to invent a new playbook. You need to actually run the one that already works, instead of skipping it because "this time the technology is different."

What "the people dimension" concretely means for an AI rollout

BCG's language — organization, operating model, processes, culture — maps directly onto decisions I watch AI project sponsors skip every quarter:

• Organization. Who owns AI adoption as a job function, not a side project? Most AI rollouts have a technical owner and no adoption owner. That gap alone predicts failure.
• Operating model. Does the workflow actually change, or does the AI tool get bolted onto the old process? Bolted-on tools get abandoned within two quarters — they add a step instead of removing one.
• Processes. Are new AI-assisted workflows documented and trained, or is "explore the tool yourself" the entire enablement plan?
• Culture. Is using the AI tool something the organization visibly rewards and models from leadership down — or something quietly optional that nobody wants to be the first to fully adopt?

None of these four questions require a data scientist to answer. They require a change management plan — the same one BCG was describing when the technology in question was still legacy ERP systems, not large language models.

The takeaway for leaders staring at their own AI rollout numbers

If your AI initiative is stalling, resist the instinct to treat it as evidence that AI is uniquely unmanageable. It isn't. You're looking at the same 70% failure pattern organizations have been generating for over a decade — a pattern with a known cause and a known fix. The technology changed. The determining factor didn't.

Build the organization, operating model, process, and culture case for adoption with the same rigor you'd apply to any transformation program — because that's exactly what this is. AI is not exempt from change management. It's the latest, highest-stakes proof that change management was never optional in the first place.