A lot of the companies we talk to right now can't quite say what AI is supposed to get them, but they feel like they have to be doing it anyway. The pressure comes from every direction. Vendors swarm them at conferences and fill their inboxes with "we can do AI for you." Board members feel like they are expected to push for it. Peers in the industry keep talking about what they have rolled out. We have clients who walked out of an event with a stack of business cards and still couldn't tell you what any of those vendors would actually do for their business, because the pitches were mostly jargon.
So companies buy something. And six months later, when someone asks whether it worked, nobody has a good answer, because nobody wrote down what "worked" was supposed to mean.
Here's what I think fixes most of this. Before you buy or build anything, write down which business lever the project is supposed to move and the number you'll watch to know if it moved. If you can't do that, you're probably not ready to spend the money yet, and that's fine. It just means the next step is a conversation instead of a purchase.
A business only grows in a few ways
At a basic level, a business grows by reducing costs, getting more transactions, raising the value of each transaction, or keeping customers longer. You can call them something else or split them up more finely, and plenty of finance people will, but almost everything a company does to grow lands on one of those.
AI doesn't get a special category. It's another way to pull one of those levers, the same as hiring a person, changing a process, or buying any other piece of software. If an AI project isn't moving one of them, you're mostly patting yourself on the back for implementing AI. That might feel good in a board update, but it doesn't show up anywhere that matters.
The nice part about thinking this way is that it cuts through a lot of the noise. When a vendor pitches you an agent, a copilot, or a company brain, you can ask one plain question: which of these four does it move for us, and by how much? Good partners will have a real answer, or at least they'll want to work it out with you. The ones who keep talking about features are telling you something too.
Write down the before number, even if it's rough
The first thing we map out with a company at Spej is the before number. How long does this workflow take today? What does it cost? How many quotes go out a week, how many deals close, how many customers renew? Whatever the lever is, there's a number attached to it right now, before AI touches anything.
People sometimes freeze here because they don't have perfect data, and I get it. It's honestly fine if the before number starts as a rough estimate. Ask the people who do the work. They usually know roughly how many hours a week a task eats, even if nobody has ever measured it formally. Write the estimate down, note that it's an estimate, and fine-tune it as you go. A rough number you keep improving is far more useful than no number at all, because without one you'll never be able to say what changed.
This also protects you later. When the budget conversation comes around next year, you'll be able to point to a before and an after instead of a feeling that things seem a little better.
The soft benefits still count, and you can still measure them
Some of the value from AI shows up in places that are harder to put on a spreadsheet: employee satisfaction, morale, how much easier a job gets, time saved on the tedious parts of someone's week. I hear people say those things can't be measured, so they either skip them or use them as a vague justification for whatever they already bought.
I think you can still find a way to measure them. You can survey the team before and after. You can track how long a task takes now versus three months from now. You can watch turnover or how quickly new hires get up to speed in a role. And most of those soft benefits eventually connect back to one of the levers anyway. Time saved either reduces cost or frees people to do work that brings in more business. Better morale tends to help you keep good people, and good people keep customers. Naming that connection out loud makes the soft benefit a lot easier to defend.
FOMO makes you skip this step
When you're buying out of pressure, the lever conversation feels slow. Someone on the board wants to see movement, a competitor just announced something, and a vendor is offering a discount if you sign this quarter. Taking a few weeks to figure out the lever and the before number can feel like falling behind.
In my experience it goes the other way. The companies that skip this step end up with licenses nobody uses, pilots that never turn into anything, or a tool that technically works but doesn't change any number leadership cares about. Then they have to go back and do the thinking anyway, except now they've already spent the budget and burned some trust with their team. Spending a little time up front is usually the fastest way to get somewhere real.
Pick one lever and one workflow to start
You don't need an AI strategy for the whole company before you do anything. Pick the lever that matters most to your business right now. Then find one workflow that feeds it, where the people doing the work are already frustrated and would welcome some help. That gives you a clear number to watch, a team that wants it to succeed, and a result you can actually point to when you decide what to do next.
What to do this week
If you've got AI projects running or a few pitches sitting in your inbox, try this:
- List every AI tool or project you're paying for or seriously considering.
- Next to each one, write which lever it's supposed to move: lower costs, more transactions, bigger transactions, or keeping customers longer.
- Write the before number for that lever, even if it's a rough guess from the people who do the work.
- For anything where you can't fill in the lever or the before number, pause before spending more and have that conversation first.
- Bring this list to your next vendor or internal AI meeting and ask them to fill in the after number they expect.
You'll probably find a couple of projects that don't map to anything, and that's useful to know. You'll also likely find one or two that clearly do, and those are the ones worth putting your energy behind.
If you want help mapping your levers and before numbers so AI spending ties back to real results, that's where we start with every company at Spej. Talk with the Spej team.
