The AI Success Formula McKinsey Found Before ChatGPT Existed

Here's the detail that should stop you: this finding is from 2019. ChatGPT didn't exist. "Generative AI" wasn't a phrase anyone used in a budget meeting. The AI boom that's driven three years of frantic enterprise spending hadn't started. And McKinsey's survey data already showed, with total clarity, exactly where the money needed to go for AI to actually work.

Nobody listened. Or rather — everyone read it, nodded, and then went back to buying more licenses.

What the research actually found

Writing in Harvard Business Review, McKinsey partners Tim Fountaine, Brian McCarthy, and Tamim Saleh reported on a pattern from the firm's global AI survey data: the organizations that had successfully scaled AI weren't the ones with the most sophisticated models or the biggest data science teams. They were the ones that treated AI as an organizational change project first and a technology deployment second.

Nearly 90% of the successful scalers had spent more than half of their analytics budget on activities that drove adoption — not on the models, not on the infrastructure, not on the tooling. On the human side: redesigning how work actually got done, communicating why it was changing, and training people to work differently.

That's not a footnote. That's the majority of the budget, in the majority of successful organizations, going somewhere other than the technology everyone assumes is the point.

Why this number still holds, six years later

If this were an isolated finding, you could dismiss it as one survey's artifact. It isn't. BCG's 2025 research landed on almost the identical conclusion from a completely different angle: the 10-20-70 rule, where top-performing organizations allocate roughly 10% of their AI effort to algorithms, 20% to data and technology, and 70% to people, process, and cultural change. I broke down the mechanics of that split in The 70% Rule.

Two different consultancies. Two different survey methodologies. Six years apart — one before the generative AI boom, one squarely inside it. Same conclusion: the technology is the smaller line item. The organizational change work is where the ROI actually gets made or lost.

When a finding survives that big a shift in the underlying technology without changing, it's not a trend. It's a structural truth about how organizations absorb change — and AI hasn't altered that structure. It's only raised the stakes.

Where "adoption spend" actually goes

This isn't an abstract budget category. In practice, it breaks into three concrete buckets:

• Workflow redesign. The processes people actually use to do their jobs have to be rebuilt around the new capability — not bolted onto the old process with an extra login step.
• Communication. Not a launch email. A sustained narrative that answers the question every employee is silently asking: what does this mean for my job, my judgment, my value here?
• Training. Not "here's the tool, here's a 20-minute demo." Structured skill-building that gets people to proficiency, with reinforcement built in past the first week.

None of that shows up on a vendor invoice. All of it shows up in whether the deployment actually changes what people do.

The governance gap nobody names

Here's where most organizations quietly fail before they even start: there's no line item for this. Technology spend gets its own budget code, its own approval chain, its own executive champion. "Adoption" — the activity that McKinsey's data says determines whether any of it works — usually gets folded into "training" as an afterthought, if it gets a budget line at all.

That's a governance failure, not a resourcing accident. If the research is this consistent, this durable, and this well-documented, the absence of a formal adoption budget isn't an oversight — it's an organization that hasn't yet decided AI adoption is a discipline worth funding like one.

PROSCI's ADKAR model exists precisely to give that discipline structure: Awareness, Desire, Knowledge, Ability, Reinforcement. It's the operational version of what McKinsey's survey found empirically. The data says spend on adoption. ADKAR tells you exactly what "adoption" has to include to actually stick.

What this means for the budget conversation you're having right now

If you're building an AI business case in 2026, the framing that gets approved shouldn't be "$X for the platform." It should be a split — and based on the research, adoption should be the larger half, not the rounding error. That's a harder conversation to have with a CFO who's used to approving software line items. It's also the conversation that separates the 90% who scale successfully from everyone else who's still wondering why the pilot never became the rollout.