Choose work you can examine.
“We should use more AI” does not give you much to work with. “Every proposal requires me to reconstruct the customer history” does. You can find a recent proposal, see where the information came from and count how often the same preparation occurs.
The first workflow should matter to the business and be small enough to inspect. Follow a real example through the awkward parts. Include the email somebody forwards, the spreadsheet nobody updates and the point where a colleague needs your memory to continue.
For a one-person company, follow your own working day. Work can move between roles even when the same person performs all of them.
Look for the AI Gap.
The AI Gap is the distance between how your company operates today and how it could operate using today’s capabilities. Your company was designed when intelligence was expensive. Reading an unusual request, connecting it to earlier work and preparing a decision generally required somebody’s attention.
Some of that work can now change. The valuable part is identifying where, in your company, the new capability is reliable enough and economically worthwhile. The company still needs trustworthy records, clear responsibility and a way to handle exceptions.
A useful starting question is: what preparation has to happen before a capable person can exercise judgment?
Find the context the work depends on.
An AI-generated draft is only as useful as the information it can use and the review it receives. For the workflow you chose, identify the authoritative records, the relevant history and the rules people apply. Note where those rules live only in somebody’s head.
- What starts the work?
- What information is needed to finish it?
- Where does that information live, and can it be trusted?
- What does a person have to interpret or decide?
- How does the business know the outcome actually happened?
These questions connect to the five Business Nervous System functions: Sensing, Signaling, Processing, Deciding and Regulating. A missing connection or unclear decision may need attention before AI can help.
An example: preparing a proposal.
Illustrative example, not a client result. A customer asks for work that resembles an earlier project. Today, the owner reads the request, finds the history, checks what was promised and prepares the scope.
A possible redesign assembles the relevant records and flags missing details before drafting a proposal. The owner reviews the scope, price and commitments. The system records the approved decision and makes the next action visible.
The experiment needs to measure the total preparation and review time, the quality of the result and any rework. A draft that appears quickly but takes longer to correct has not established a useful improvement.
Measure what happens before changing it.
Collect a representative sample of the current work. Include straightforward cases and exceptions. Record active work time separately from waiting, the number of handoffs, errors or rework, and what prevents completion.
Choose a primary outcome that matters: less preparation time, a shorter turnaround, fewer missed details or capacity to handle more work. Define what acceptable quality looks like. The Prove stage uses that baseline to judge the change.
Time released is potential capacity. It becomes a business benefit when you use it for something valuable or avoid a real cost. Record that distinction when estimating the opportunity.
Make one decision about what comes next.
If the opportunity is credible, agree on a bounded first implementation and the evidence needed to continue. If the records are unreliable or the review burden is too high, fix the underlying constraint or choose another workflow.
The $999 read makes that first recommendation. It includes a 90-minute session and a written read within two business days. Building and ongoing work are scoped separately.
Company-Level AI develops through changes that hold up in the actual operation. You can begin without replacing the business you have already built.