Insights
Your AI budget is backwards: spend on people, not just tools
Ask a company what they spent on AI last year and you'll get a list of subscriptions and maybe a consulting invoice. Ask what they spent teaching their people to use any of it, and you'll usually get silence — or a link to a YouTube playlist someone shared in Slack.
That ratio — nearly everything on tools, nearly nothing on people — is the single most reliable predictor of the outcome: software that gets demoed, admired, and abandoned by week three.
The spending wave is here either way
BCG's AI Radar 2026 survey of over two thousand executives found companies planning to roughly double their AI spending in 2026, and 72% of CEOs now name themselves the primary AI decision-maker — double the year before. The money is coming; the only question is its shape.
And the shape is usually wrong. Budgets get built around what's easy to buy (licenses, seats, platforms) rather than what actually determines returns (whether anyone changes how they work). Unused software is the most expensive kind of software — you pay for it twice, once in cash and once in the credibility your next initiative loses.
A rule of thumb that fixes the ratio
The budgeting heuristic we give clients: for every dollar you spend on AI tools, plan two to three dollars on the people who'll use them. Training, workflow redesign, documentation, the awkward supported first month — the unglamorous line items that determine whether the glamorous one pays off.
It sounds inverted until you look at what actually kills AI initiatives. It's almost never the model. It's a team that was handed a tool without being told what it means for their jobs, trained — if at all — on generic examples that look nothing like their actual work, and then left alone the first week reality pushed back.
What the people-money should actually buy
Not inspiration keynotes, and not certification theater. Four things:
An honest answer to the fear question, first. Before any training lands, someone senior has to address what AI means for people's jobs — directly, without corporate euphemism. For most mid-market roles the honest answer is that AI takes over tasks, not jobs, and the people who master it become more valuable. Skip this conversation and you'll get quiet resistance no curriculum can overcome: tools mysteriously "not working," workflows that revert the moment nobody's watching.
Training built on real tasks. A workshop where each person practices on their own actual work — this report, this inbox, this handoff — changes behavior. A workshop of generic prompt tips gets rated 9/10 and forgotten by Friday. The difference isn't the trainer; it's the material.
Reinforcement on a schedule. Whatever you build, check it at 7, 30, and 90 days: is it still being used, where did it break down, who needs a refresher. Training without reinforcement evaporates — this is the most boring and most violated rule in the entire field.
Documentation as a habit. Every workflow someone masters should end up written down, so the skill belongs to the company and not just the individual. Your AI capability shouldn't resign when your most enthusiastic employee does.
The test for your current budget
Pull up next quarter's AI spending and ask one question: what fraction of this is dedicated to changing how people actually work? If it rounds to zero, you're not budgeting for AI capability. You're budgeting for shelf-ware with a modern logo.
Source: BCG, "AI Radar 2026." Team training is the "Embed" step of the Carbon Copy Method — role-specific workshops built on your team's real tasks, with the 7/30/90-day follow-ups included. Details on the Team AI Upskilling page.