Where to actually start with AI when everything sounds urgent

Everywhere you look, someone is telling you AI is going to change your business. Almost no one is telling you how — or where to even start. So the pressure builds, and it pushes a lot of good teams into the same wrong first move: a flashy demo that wows the room on Tuesday and never touches real work again by Friday.

We’ve watched it happen enough times to recognize the pattern. A chatbot gets bolted onto the website. Someone pipes the whole company wiki into a tool nobody asked for. There’s a burst of excitement, a screenshot in the group chat, and then — nothing. The work that was slow last month is still slow this month. The only thing that changed is another login and a quiet new subscription.

The better starting point is dull on purpose. Not the most impressive thing AI can do. The most boring thing it can do reliably, on a task you already wish you didn’t have to do. Here’s how we find it.

Don’t start with AI. Start with the work.

The mistake hiding inside “where do we start with AI” is that it puts the technology first. The question that actually leads somewhere is the opposite: where does our work get stuck?

Spend a week paying attention to the small frustrations. The report someone rebuilds by hand every Monday. The inbox that fills with the same five questions. The quote that takes two days to go out because three people have to touch it. None of that sounds like “AI.” That’s exactly why it’s the right place to look — it’s real, it’s yours, and it’s costing you something today.

When we sit with a team during a discovery, we’re not hunting for the most futuristic use case. We’re watching how the work actually happens and listening for the sighs. The sighs are the signal.

The three tests for a good first project

Once you have a short list of annoying tasks, run each one through three quick tests. A good first AI project passes all three. If a task fails even one, set it aside — not forever, just not first.

1. Your team already hates it. Pick something nobody enjoys and nobody will miss. If the work is tedious, repetitive, and a little soul-deadening, you’ve found a candidate. Bonus: people are far more forgiving of a new system when it’s taking a chore off their plate rather than changing work they’re proud of.

2. It happens often. A task you do fifty times a week is worth automating. A task you do twice a year is not — even if it’s painful each time. Frequency is what turns “saved a few minutes” into “saved a few hours every week.” Rare problems are interesting; frequent problems are profitable.

3. It follows the same shape every time. This is the one people skip, and it’s the most important. AI is reliable when the task has a predictable structure — same kind of input, same kind of output, clear rules for what “good” looks like. Sorting incoming emails into categories has a shape. “Use good judgment on anything that comes up” does not. The more a task looks the same each time, the more dependable the result.

A task that’s hated, frequent, and same-shaped is the sweet spot. It’s not glamorous. It will not impress anyone at a conference. It will quietly hand your team back hours they didn’t think they could get back — and that’s the whole point.

Size it before you build it

Once you’ve picked the task, resist the urge to make it bigger. The instinct is almost gravitational: while we’re in here, we may as well also… That “may as well” is how a one-week win turns into a three-month project that never ships.

Keep the first build deliberately narrow. One workflow. One clear before-and-after. Something small enough that a person can still check the output by hand at first — because they should. We tune the line between what the system handles on its own and what it routes to a human for review, and early on that line sits well toward the human side. The system never approves what it isn’t sure about; it flags it, explains why, and waits for a person. Trust gets earned, then the line moves.

A good way to size it: if you can’t describe the project in one sentence, it’s too big. “Draft a first reply to every support email so a person can edit and send it” is the right size. “Transform our customer experience” is not a project — it’s a slogan.

How to know in a week whether it worked

Here’s the part most “AI strategies” leave out: a way to tell, quickly, whether the thing was worth doing. You don’t need a quarter to find out. You need about a week, and one number you already track.

Before you start, write down the before. How long does the task take now? How many times a week does it happen? Who does it? Don’t estimate from memory — memory inflates. Time it honestly for a few days if you have to.

Then run the new way alongside the old way and measure the after against that same number. Not “does it feel faster” — does the number move? Hours back per week. Replies out the door before lunch instead of after. Errors caught before they reached a customer. Pick the one that matters to you and watch it.

And decide your exit criteria up front — the conditions under which you’d stop. A pilot you’re allowed to stop is a pilot you can run honestly. If the number doesn’t move in a week or two, you stop, you’ve lost very little, and you’ve learned something real about where AI does and doesn’t fit your business. That’s not failure. That’s the cheapest research you’ll ever buy.

Start small, earn the next step

The reason we keep the first project small isn’t caution for its own sake. It’s that each step should justify the one after it. Ship one workflow that pays for itself, prove it with a number, and you’ve earned the right — and the credibility — to do the next one. Skip that, and you’re back to demos and shelfware and a vague sense of falling behind.

So if everything about AI sounds urgent right now, here’s the calm version of the plan. Don’t chase the most impressive thing. Find one task your team already hates, that happens all the time, that looks the same every time. Build something small enough to check by hand. Measure one honest number. Keep it if it works, drop it if it doesn’t.

It’s not the exciting answer. It’s the one that’s still paying off a year later.

---

PineyWoods helps small and medium businesses in the U.S. find the few places AI genuinely earns its keep — then build them and hand them off. If you want a second set of eyes on where to start, book a free call. Thirty minutes, a clear next step, and it’s useful either way.