Most AI strategy work should be less theatrical.

You do not need a manifesto to start. You need a sober inventory of where AI changes your economics, customer behavior, and competitive exposure. The point is not to declare that you have a moat. The point is to find where power might compound and where old protection is decaying.

This audit is meant for founders, operators, and executives. It is deliberately practical. If a question cannot change a decision, it probably does not belong here.

Step 1: map the workflow

Start with the actual work, not the org chart.

Where does the customer problem begin? Who touches it? What systems are involved? Where do decisions happen? Where are mistakes caught? What takes too long? What requires expert judgment? What creates customer frustration?

AI opportunities usually hide inside workflows, not inside strategy slides.

Mark the steps that are routine, judgment-heavy, compliance-sensitive, relationship-sensitive, or data-rich. Those labels matter because AI affects each one differently.

Step 2: test each power

Run the Seven Powers as questions.

For scale economies: does AI lower unit cost as volume grows, or does usage create new variable cost?

For network economies: does more participation make the product better for each user, or do we merely have more users?

For counter-positioning: what would incumbents damage if they copied our model?

For switching costs: what performance would customers lose by leaving?

For branding: do customers trust us under uncertainty, and does that trust change sales economics?

For cornered resource: what privileged asset do we have that improves product performance or economics?

For process power: do we learn to use AI better in ways competitors cannot easily observe or copy?

The answer can be "none." That is useful. It saves money.

Step 3: separate adoption from power

AI adoption is not the same as defensibility.

A feature can drive usage and still be copied. A chatbot can improve support and still create no advantage. A model integration can excite customers and still become table stakes six months later.

Score each initiative twice: impact and defensibility.

High impact, low defensibility may still be worth doing. It can improve the business. Just do not call it a moat. High defensibility, low impact is usually a science project. The best opportunities improve today's business while strengthening tomorrow's position.

Step 4: look for decay

Ask where AI weakens existing protection.

Can customers switch more easily? Can competitors produce similar work with fewer people? Can buyers compare pricing and quality faster? Can a platform bundle your feature? Can open-source tools erode your differentiation?

This is not pessimism. It is maintenance.

Every company has advantages with half-lives. AI changes those half-lives. Some stretch. Some collapse.

Step 5: choose the operating bets

The audit should end in decisions.

Pick a small number of bets where AI can improve performance and strengthen power. Assign an owner. Define the metric. Set a review date. Decide what evidence would kill the bet.

Good metrics are boring: gross margin, retention, expansion, cycle time, win rate, support cost, onboarding time, error rate, sales cycle, usage depth.

Avoid vanity metrics unless they connect to behavior that matters.

A simple scoring table

For each AI initiative, score from 1 to 5:

  • Customer impact
  • Economic impact
  • Defensibility
  • Speed to test
  • Risk level
  • Strategic fit

Then write one sentence: "This creates power if..."

That sentence is useful because it exposes assumptions. If nobody can complete it cleanly, the initiative may still be worth testing, but the strategy is not clear yet.

The point

The point of an AI power audit is not to be right forever.

The point is to stop confusing motion with advantage. AI makes motion cheaper. That is helpful. It also makes imitation cheaper. That is the problem.

Durable strategy starts where the imitation gets expensive.

This is part 10 of 10 in Seven Powers in the AI Era.