AI for Executives: The Five Decisions That Are Yours, and the Twenty That Aren't (2026)

TL;DR: What an Executive Needs to Know About AI in 2026

You don't need to understand transformers. You need to make five calls and then get out of the way. I sit in these rooms as a fractional CTO, and the executives who get results from AI are not the ones who took the most courses. They're the ones who funded three named workflows instead of one platform, wrote down what they were refusing to do, put one person's name on each initiative, and could describe the outcome to a board in hours or margin instead of in tools. This post is that briefing, minus the slide deck: the five decisions that are yours, the twenty that aren't, the budget mistake that costs the most, and what executive AI training should cover if you're going to pay for it.

The five decisions that are actually yours

Every executive-level AI conversation collapses into one of these. If a question isn't one of the five, it's a delegation.

1. What to fund. Not "should we do AI" — your staff already are, approved or not. The decision is which two or three workflows get real money and a real owner this quarter.
2. What to refuse. The no-list: the use cases you are deliberately not pursuing this year, a sentence each on why. Almost no executive makes this decision explicitly, which is why companies end up with eleven pilots and zero results.
3. Who owns it. One accountable executive for the portfolio, one named manager per workflow. Not a committee. Not the vendor. Not "the innovation team."
4. What the board is told. Results in the board's units — hours returned, margin on a process, a cost line that went down — for named workflows. Never a list of tools purchased.
5. What "we use AI" means in the P&L. Where the cost sits, where the return is expected to show up, and by when. If you can't point to the line, it's not a business initiative yet. It's a hobby with a budget.

The twenty decisions that aren't

Which model. Which vendor. Claude or ChatGPT or Copilot for a given team. How to write the prompt. Which data to connect first. Whether the output needs a human review step. Which integration to build. Chat window or automation. I could keep going past twenty.

These are real decisions. They are also decisions the person who runs the workflow every day will make better and faster than you. The pattern that works is the one I lay out one level down in AI for managers: the manager owns the workflow, picks the tool, runs the pilot, reports in numbers. Your job is to have funded the right manager and to refuse to be the bottleneck on the twenty.

The most expensive mistake: buying the platform before naming the workflow

I've watched this at companies from twelve people to several hundred, and it always looks responsible while it's happening. An executive gets a demo and approves an enterprise agreement — seats for everyone, a rollout plan, a training calendar. Six months later adoption is a fraction of the seat count, nobody can name a process that changed, and the renewal conversation is awkward.

The cause is always the same: the platform was chosen before anyone wrote down which workflow it was for, what that workflow costs today, and what "better" looks like in a number. Reverse the order and the whole thing gets cheap. In the engagements I run, once a workflow is named — quoting, intake, support triage, weekly reporting — the software is the smallest line in the budget; seats and usage for a real business workflow are typically tens to low hundreds of dollars a month. The money goes to redesigning the process, connecting the tool to the systems that hold the data, and getting people to change how they work. If the proposal in front of you is mostly licenses, send it back with one question: which workflow, and what does it cost us today?

Budget: fund workflows, not tools

A practical framing I use with executives, as of 2026: think of AI spend in three buckets, and be suspicious of any proposal where the first bucket is the biggest.

• Software. Seats, API usage, the odd add-on. Small, and roughly predictable once the workflow is running.
• Build and integration. The work to connect the model to your CRM, your inbox, your documents, your ticketing system. This is where the real cost sits, and it's where a good AI consultant or fractional CTO earns their fee — by scoping it honestly instead of letting it sprawl. I've written up what that scoping typically costs in how much AI consulting costs.
• Change. The manager's time to redesign the process, the team's time to learn it, the two months where it's slower before it's faster. Unbudgeted in almost every plan I see, and the reason most pilots die in month three.

Fund one workflow end to end, all three buckets, before a second. Slower for a quarter, dramatically faster after.

Risk: the three that are real and the one that's theater

The real risks, in the order they bite: data leaving the building (customer records pasted into a consumer chat tool with no business agreement); wrong answers acted on (a confident summary of a contract clause that isn't what the contract says); and vendor dependence (a critical process on a tool whose pricing or terms change under you). All three are handled by boring governance — an approved-tools list, a human-review step on anything that reaches a customer or regulator, and no critical workflow on a single vendor without an exit.

The risk with the most airtime and the least exposure is the abstract one — "AI replacing our people." In every company I've worked with, the near-term reality is that AI removes the parts of jobs people hated and the constraint becomes finding enough good people to run the faster process. Plan for the real three. Have a sentence ready for the fourth.

Ownership: who owns AI in a company

The honest answer for most companies under a few hundred people: not a new executive. The portfolio belongs to whoever already owns operations — usually the COO, sometimes the CEO — and each workflow belongs to the manager who runs it. Department heads own their department's use cases and report up in the numbers they already report. The Chief AI Officer versus fractional CTO question comes up here, and my view hasn't changed: hire a full-time AI executive when the portfolio is big enough to need one, not to create one. Before that, a fractional technical seat reporting to the operations owner covers it at a fraction of the cost.

The rule I'd hand any executive: don't fund anything without a name on it. Not a team, not a vendor, not a committee. A person, whose quarterly numbers now include this workflow.

What the board will ask, and what to have ready

Four questions, nearly always in this order: what are we doing, what's it costing, what's the risk, what are competitors doing. The one that goes badly is the second, because "we rolled out Copilot to two hundred people" is a cost, not a result. Have two or three named workflows with a before-and-after in the board's units — hours returned, margin on a process, a cost line that dropped — and the no-list. A board that hears "three things we're doing, the results, and five things we decided not to do" stops asking whether you have an AI strategy. You just showed them one.

AI executive training: what's worth paying for and what isn't

Most executive AI training, as of 2026, is a tour: what a large language model is, a demo, a slide on ethics, a Q&A. You leave able to talk about AI and no closer to a decision.

The version worth paying for is a working session with your own workflows on the table. It ends with three workflows priced in all three buckets, a named owner for each, a written no-list, and a one-page answer to the four board questions. If a training offer can't tell you in advance which of your workflows it covers, it's a lecture. The free version of that structure is the AI readiness checklist; the paid version is the $2,500 AI Readiness Assessment — two weeks, a written roadmap, a walkthrough, an owner's name next to every line.

Use the tools yourself, for one reason only

You don't need to become the expert. You need thirty days of contact with Claude or ChatGPT on real work — a board memo, a contract summary, "argue the other side of this decision" — so that when someone puts a demo in front of you, you can tell what the tool does from what the salesperson says it does. Same reason I tell people running teams to go first in AI for leaders: credibility on the twenty decisions you're delegating comes from having touched the thing.

The micro-action for today

Write the no-list. Five AI use cases your company is not pursuing this year, a sentence each on why. It takes twenty minutes, it's the decision almost nobody makes explicitly, and it will clarify the five you are funding faster than any briefing. If you want to pressure-test the list with someone who sits in these rooms for a living, that's a free 30-minute strategy call, no pitch — bring the list, leave with it shorter or sharper.