Most companies bolt AI onto the business they already have. They buy the tool first, then ask what it should do. The order is wrong — and it's why so many AI initiatives stall six months in with an expensive subscription and nothing to show for it.
The right sequence starts earlier. Before you choose a tool, you decide what the business needs. Before you decide what the business needs, you understand where the business actually is. AI is the last decision, not the first.
The decisions you should make before you choose a tool
The companies that get AI right share a habit: they treat AI like any other strategic capability. They define the work the AI is supposed to change before they evaluate vendors. Six decisions, in order, before a single tool gets demoed.
- Decide what's broken. Not "we want to use AI" — specifically which workflow, which team, which output is below where it should be.
- Decide what "better" looks like. What measurable improvement justifies the effort? Hours saved, errors prevented, revenue unlocked. If you can't name the number, the project doesn't exist yet.
- Decide who owns the outcome. Not the vendor, not the consultant — a named person inside the company. AI without an internal owner is software, not strategy.
- Decide the data reality. What data does the model need, where does it live, who has access. The honest answer is usually "scattered" — and that's the work before the work.
- Decide the change management cost. Every AI deployment changes how people do their jobs. If the team isn't bought in, the tool will be unused inside ninety days.
- Decide the kill criteria. What signal tells you this isn't working, and at what point do you stop? Most AI initiatives die slowly because nobody set a deadline.
Each of these is a decision you can make without a vendor in the room. Each one is harder than picking a tool. That's why most companies skip them.
What the vendor conversation should look like
Once the six decisions are in writing, the vendor conversation becomes short and uncomfortable for the vendor. You walk in with a specific problem, a specific outcome, a specific owner, and a specific kill criterion. You're not asking what the tool does — you're asking whether it can do this.
Most demos collapse under that level of specificity. That's the point. The demos that survive are the ones worth paying for.
The leadership checklist
If you're a CEO, COO, or operating partner trying to make AI real inside a mid-market business, run through this list before the next budget cycle. The ones you can't check are the work for the next quarter.
- We have a written list of the three workflows AI should change in the next 12 months.
- For each workflow, we have a measurable target — not "improve productivity," but a number.
- Each AI initiative has an internal owner whose performance review references it.
- We have an honest audit of where the data lives and what shape it's in.
- We have a change-management plan, not just a software-rollout plan.
- Every active AI initiative has a written kill criterion and a date.
- We can name the next three decisions before we name the next three tools.
What this looks like when it works
The companies doing AI well don't have more AI. They have fewer, better-chosen pieces of it — embedded in workflows where the outcome is defined, the owner is named, and the data is in shape. The CFO can point at a number. The COO can point at a workflow. The CEO can point at a strategy.
That's leverage. The opposite — buying tools, hoping they help, defending the spend at the next board meeting — is clutter. AI is the same as any other strategic capability: the discipline is in the order of operations. Decide first. Tool last.
Drawn from real engagements with operators building what's next.