Earlier this year, I sat in a board meeting where the AI budget for the coming year had already been approved. Six figures, comfortably. Nobody in the room was uncomfortable with the number. What nobody had discussed — until I asked — was how that number would actually be divided.
When we pulled the spend apart line by line, the split was almost exactly 90/10. Ninety percent was going into automation projects: invoice processing, ticket routing, document extraction, the kind of AI work that shows up in next quarter's efficiency report. Ten percent — a single line item, owned by no one in particular — was earmarked for what the CEO kept calling "the transformation piece." A new AI-native product line that could, in three years, change what the company sold.
Nobody had decided this split on purpose. It had simply happened, the way budgets tend to happen: each department requested funding for the automation work it could already justify with a clean ROI case, and the transformative bet — harder to quantify, riskier, slower to prove out — got whatever was left over. In this case, that was ten percent and no accountable owner.
This is not an unusual board. In my work advising boards and executive teams on AI investment, I see this pattern more often than any other single mistake. It is worth naming precisely, because the fix is simple once you see it.
Why an AI Budget Isn't a Strategy
A budget number tells you how much you're willing to spend. It does not tell you what you're buying, or when you expect to see a return, or who is accountable for the bet that matters most. Boards routinely treat "approve the AI budget" as the strategic decision, when the real strategic decision is the allocation underneath it.
The 90/10 board I described did not lack ambition. Their CEO talked about transformation in every meeting. But ambition without an allocation rule defaults to whatever is easiest to fund — and low-risk automation is always easiest to fund, because it comes with a spreadsheet attached. The one bet that could actually change the business is the one with the least evidence behind it, which means it is also the one that gets cut first when budgets tighten mid-year. I have watched this happen twice in the same twelve-month cycle at two different companies.
A budget becomes a strategy the moment a board decides, deliberately, how much risk it is willing to fund across different time horizons — and holds someone accountable for each horizon separately.
The Three Horizons — Automate, Augment, Transform
I ask every board I work with to sort their AI spend into three buckets before they argue about the total number. Each bucket has a different job, a different payback expectation, and a different review rhythm.
- Automate. Definition: AI applied to existing, well-understood processes — document handling, support triage, reporting, scheduling. Payback window: 3–9 months, measured in hard cost or time savings. Review cadence: monthly operational review, folded into normal budget tracking. This is the horizon boards already fund well.
- Augment. Definition: AI embedded into how existing teams do higher-judgment work — sales forecasting, underwriting, research, design iteration — making people faster and better, not replacing the role. Payback window: 6–18 months, measured in output quality and capacity, not just cost. Review cadence: quarterly, alongside a named business-unit sponsor.
- Transform. Definition: AI-native bets that could change what the company sells or how it competes — new products, new business models, new categories of customer. Payback window: 18–36 months, and the metric in year one is often learning velocity, not revenue. Review cadence: quarterly board-level review, with a named executive owner who reports progress even when there is no revenue yet to report.
Most underperforming AI portfolios I review are not underfunded overall. They are misallocated across these three horizons, with Transform treated as a footnote rather than a horizon with its own rules.
A Simple Allocation Rule Boards Can Actually Use
The rule I bring into these conversations is not complicated, and that is the point — a rule a board won't actually use in a busy quarter is not a rule. I suggest starting from roughly 60 percent Automate, 25 percent Augment, 15 percent Transform, and treating that as a starting hypothesis rather than a formula carved in stone. The exact numbers matter less than the discipline around them.
Two things make the rule work. First, every horizon needs a named owner who reports on it separately — Transform cannot be a shared line item with no single accountable executive, or it will quietly lose funding to whichever horizon has a clearer story that quarter. Second, the split has to be revisited every quarter, not set once a year and left alone. Markets move faster than annual budget cycles, and a Transform bet that looked reasonable at 15 percent in January can look badly underfunded by the third quarter once you have real learning to act on — or it can look like it should be cut, which is a legitimate outcome too, as long as the board decided it on purpose.
The board from my opening example adopted a version of this rule at their next meeting. Within two quarters, their split had moved from 90/10 to closer to 65/20/15, with a named executive sponsor for the transformation bet and a monthly one-page update — even when that update was just "here is what we learned this month, and here is what we still don't know."
The Question for Your Next Board Meeting
If your board has an AI budget number already approved for this year, the question worth asking before your next meeting is not "is the number big enough." It is: how is that number actually split across automate, augment, and transform — and who owns each piece?
If you don't know the answer immediately, that is itself the answer. It usually means the split happened by accident, the same way it did in the meeting I described — and the one bet that could matter most in three years is quietly running on whatever was left over.
Written by Brian, Dr. Jonah Tebaa's AI partner, on his behalf.
