Most AI projects start in the wrong place.
A tool. A model. A use case that sounds interesting.
"Let's automate support." "Let's add a chatbot." "Let's use AI for outbound."
None of these are wrong. They're just disconnected.
From the business. From the numbers. From what actually moves the company.
We start somewhere else.
The P&L.
Why the P&L comes first
A founder's P&L is not just a financial statement.
It's a map of pressure.
Where money is made. Where it leaks. Where effort is concentrated. Where margins get thin.
Most teams don't look at it this way. They see totals. We look for patterns.
High spend with low leverage. Revenue lines that depend on manual effort. Costs that scale faster than growth.
That's where AI actually matters.
Not where it's interesting. Where it changes the shape of the business.
The diagnostic we run
We do the same thing at the start of every engagement.
A structured pass through the business, anchored in numbers.
It takes under two weeks.
1. Map the flows
We start by breaking the P&L into flows.
How does money come in? What has to happen for it to happen again? Where does it get stuck?
Revenue is not a line item. It's a sequence.
Leads → qualification → conversion → fulfillment → retention
Each step has inputs, outputs, and friction.
We map that end to end.
2. Find the manual load
Next, we look for human effort hidden inside those flows.
Not just headcount. Work.
Where are people repeating decisions? Where are they stitching tools together? Where are they acting as the system?
This is where most AI opportunities sit.
Not in replacing people. In removing the parts of work that should not require them.
3. Identify constraint points
Every business has a few places where things slow down.
A bottleneck in sales follow-ups. A delay in support resolution. An ops step that cannot scale without hiring.
We map these.
Not all constraints are worth solving. Some are strategic.
The ones we care about have two traits: they repeat often, and they directly affect revenue or cost.
That's where intervention compounds.
4. Score for impact
At this point, there are dozens of possible builds.
We cut that down fast.
Each opportunity is scored on three things:
- Impact on revenue or cost
- Frequency of the task
- Complexity of implementation
Most ideas die here.
The ones that remain are not the most exciting. They are the ones that change numbers.
5. Translate into builds
Only now do we talk about systems.
Each high-impact opportunity becomes a concrete build.
Not "an AI agent for support." A defined system with:
- Clear scope
- Inputs and outputs
- Where it fits in the workflow
- What happens when it fails
This is the difference between an idea and something you can ship.
What founders usually notice
The surprise is not where AI can be used.
It's where it shouldn't be.
Large parts of the business don't need automation. They need clarity, better process, or better positioning.
The diagnostic filters that out.
What remains is smaller than expected. And far more useful.
A few focused builds. Tied directly to how the business runs.
What you leave with
At the end of two weeks, there's no ambiguity.
You have:
- A mapped view of your revenue and cost flows
- A clear list of constraint points
- A short list of high-impact builds
- A sequence to execute them
Not a backlog.
A plan.
The TL;DR
- Start with the P&L, not the tool.
- Map how money actually flows through the business.
- Find where human effort is acting as glue.
- Focus on constraints that repeat and impact numbers.
- Cut ideas aggressively.
- Only build what changes the business.
Most teams start with AI and try to find value.
We start with value and decide where AI belongs.