You don't need an AI strategy. You need your invoices done.

The strategy deck exists because it's safer than shipping anything. That's the uncomfortable part, so let me sit on it for a minute.

Planning feels like progress. You get meetings on the calendar, a consultant's logo on the title slide, a roadmap with quarters on it. Nobody's job changes. Nobody's workflow gets touched. And nothing can go wrong, because nothing is running.

A strategy can't fail until it ships. That's exactly why it never ships.

I understand the instinct. The AI market is loud, the vendors contradict each other, and every executive has heard a horror story. Writing a document feels like the responsible move. But a plan built before you've run a single system in production is a plan built entirely on guesses: about your data, your team, your edge cases, your appetite for reviewing machine output. You don't know any of those things yet. You can't. The only way to learn them is to put something real inside a real workflow and watch what happens.

The deck defers that moment. Sometimes for years. And the whole time it sits there, the invoices keep arriving.

So skip the deck and find the workflow that bleeds.

In most businesses it's one of the same few suspects: invoices, lead follow-up, the support inbox, weekly reporting. Invoices are my favorite first target because the pain is measurable and the work is mechanical. A document arrives. Fields get extracted. Amounts get checked against a purchase order or a contract. An entry gets posted. Exceptions get flagged for a human to look at.

None of that requires anyone's judgment on a good day. All of it eats someone's week anyway.

Run your own math here, because your numbers are the only ones that matter. If one person on your team spends six hours a week keying and checking invoices, that's roughly 300 hours a year. Multiply by whatever a loaded hour costs you. Then add whatever the quieter failures have cost: the duplicate payment, the early-payment discount that expired somewhere in the pile, the month-end close that slipped because AP was behind. You know those numbers better than any strategy consultant ever will. Write them down.

Now hold up the two paths side by side. Path one is a quarter of workshops, a committee, and a pilot scheduled for "when things settle down." Path two is picking this one workflow, shipping a first version, and putting an agent in front of real invoices with a human reviewing every exception it flags. One path produces slides. The other gives you your Tuesdays back.

Here's what I noticed inside my own companies once the first system was genuinely running: the next projects stopped being a debate.

Not because we got smarter about AI in the abstract. Because a working system teaches you things no workshop can. You learn where your data actually lives, which is rarely where the org chart implies. You learn who on your team reviews an agent's output carefully and who rubber-stamps it. You learn what breaks, how often, and what a sane checking cadence looks like. You learn how to write the runbook for the second system in half the time it took for the first.

That accumulated, earned knowledge is the strategy: a short list of systems that run, what each one taught you, and what that makes possible next. Invoices handled well point straight at collections reminders. A triaged inbox points at drafted replies. Every working system exposes the next bleeding workflow, because you can finally see your operations at the resolution of actual tasks instead of boxes on a diagram.

I have never regretted starting too small. I have regretted starting too abstract every single time.