Not Boring All articles
Technology

Exploring Is Not a Strategy: The AI Announcement-to-Impact Gap

Not Boring
Exploring Is Not a Strategy: The AI Announcement-to-Impact Gap

Photo: executive presentation screen showing AI strategy slide in corporate boardroom, via bellyfull.net

Somewhere right now a company of about four thousand people is running nine AI pilots, has an AI steering committee, has published an AI principles page, has a chatbot on the intranet that answers questions about the intranet, and cannot tell you what any of it returned last quarter.

That's not a failure of technology. It's a failure of the thing that comes before technology, which is deciding what you're actually trying to change.

The Press Release Ships First

Follow the sequence in a typical large-company AI program. Board pressure arrives. The board has heard that AI is existential and wants a story for the next meeting. A senior leader — chief digital officer, chief strategy officer, whoever has the mandate — assembles a program. A vendor is selected. A pilot launches in a function that is friendly, low-risk, and has no meaningful cost of failure. Communications drafts the announcement in week six, before the pilot has produced a single number.

The announcement is the deliverable. That's the tell. When the comms plan runs ahead of the measurement plan, you're watching theater with a Kubernetes cluster.

There's a reason this pattern is so sticky. AI is genuinely hard to attribute. If a support copilot cuts handle time by 9 percent, is that the model, or the fact that you also simplified the ticket taxonomy in the same sprint? Most organizations cannot answer that, so they stop trying, and once you stop trying, you're free to claim whatever the deck needs.

Four Questions That Separate Signal From Noise

I don't care how sophisticated the model is. I care whether anyone can answer these.

What changed in a number that your CFO already tracks? Not adoption. Not prompts. Not 'hours saved,' which is a fantasy unit invented to make pilots look good. Cost per claim. Cycle time on onboarding. Revenue per rep. Something that existed before the AI program and will exist after it.

Who owns the outcome? If the answer is a center of excellence, you don't have a strategy. You have a vendor with internal headcount. Real deployments are owned by the person whose bonus is attached to the metric.

What did you stop doing? This is the question nobody wants. If AI made a workflow faster and the company kept the same headcount, the same approvals, and the same nine-step process with a model bolted onto step four, you didn't transform anything. You added a step.

What's the kill criteria? Programs without a defined failure condition cannot fail, which means they cannot succeed either — they just continue, at a fixed quarterly cost, forever.

Pilot Purgatory

The most common failure state in 2026 isn't a model that doesn't work. It's a model that works fine and never ships.

A pilot proves out in a sandboxed team. It then needs to clear security review, procurement, data governance, the integration backlog, a union conversation, and a change-management plan. That takes eleven months. By month eleven, the vendor has changed pricing, the champion has changed jobs, and the pilot's original owner is now running a different program. The thing quietly expires and gets re-listed in next year's AI portfolio as 'phase two.'

Multiply that by nine pilots and you get an AI program that is extremely busy and structurally incapable of producing a result. Meanwhile the actual AI gains in the company are happening off-book — one analyst automating her own reporting in a spreadsheet-adjacent tool, three engineers wiring an internal search, a customer support lead who built something nobody approved. None of that shows up in the program, because the program only counts what it launched.

What a Real AI Strategy Contains

Short list, and it's boring on purpose.

A small number of named workflows, chosen because they are expensive and repetitive — not because they're impressive. A baseline measurement taken before deployment, using the company's existing financial or operational definitions. A single accountable owner per workflow, sitting in the business, with a number in their goals. A governance path designed to take weeks, not quarters, for anything under a defined risk threshold. A published inventory of what's been tried and what's been retired. And a budget line that can be cut.

That last one matters more than people admit. A program that cannot be defunded is not a strategy; it's an entitlement.

Questions Employees Should Ask

If you work at one of these companies, you have leverage you may not realize. Ask what percentage of announced pilots reached production. Ask whether the AI tool you're required to use has a measured effect on your team's cycle time, or whether usage is being tracked as a proxy for value. Ask who decided that.

Then watch what happens. In healthy organizations, someone has the number and is mildly annoyed you had to ask. In theatrical ones, you get a slide about the transformation journey. The slide is the answer.

The Tell

Here's the simplest diagnostic I know, and it works on any company at any size. Ask ten random employees what specifically changed about their job because of the AI program. If you get ten variations of 'we have a chatbot now' and 'we're encouraged to use it,' you're looking at a communications strategy with a compute bill.

Real transformation is unglamorous and slightly annoying to describe. It sounds like: claims now route automatically, so three people moved to a different team, and our error rate fell by a fifth. Nobody puts that in a press release, which is exactly why you should trust it.

Exploring is a fine thing to do in year one. In year four, it's a confession.

All Articles

Related Articles

Pre-Meeting Culture Is Out of Control. Here's How Some Teams Finally Fixed It.

Pre-Meeting Culture Is Out of Control. Here's How Some Teams Finally Fixed It.

Still Breathing, Already Dead: The Truth About Stalled Startups That Refuse to Quit

Still Breathing, Already Dead: The Truth About Stalled Startups That Refuse to Quit

We Installed Software to Watch You Work. Turns Out, You Noticed.

We Installed Software to Watch You Work. Turns Out, You Noticed.