Switching costs are often confused with inconvenience.

A customer may dislike migrating data, retraining staff, or changing vendors. That does not mean you have power. It means the customer has chores. If the replacement is meaningfully better, cheaper, or bundled into a system they already use, chores get done.

AI raises the standard for switching costs because it can reduce migration pain. It can map schemas, rewrite workflows, train users, generate documentation, and build connectors. A lot of old friction becomes less intimidating when software helps customers leave.

So the question changes: what would the customer actually lose by switching?

Shallow lock-in

Shallow lock-in is built on mess.

Bad exports. Custom fields nobody understands. Integrations held together by one consultant. Training debt. Internal habits. Procurement fatigue.

These things can slow churn, but they are not attractive power. They create resentment. They invite competitors to make leaving easier. AI is good at attacking this kind of mess because much of it is translation and cleanup.

If your retention depends on customer confusion, assume the clock is ticking.

Deep workflow lock-in

Deep lock-in is different.

It happens when the product becomes part of how the organization makes decisions, coordinates work, and remembers what happened. The system contains context that is hard to reconstruct: approvals, exceptions, preferences, performance history, risk judgments, customer nuances, and institutional knowledge.

AI can strengthen this because it makes context more useful. A database of old tickets is mildly useful. An assistant that knows how your team resolved similar issues, which customer promises matter, and when to escalate is much more useful.

The lock-in comes from accumulated operating memory, not from trapping data.

Personalization that matters

Many AI products will claim personalization as switching cost. Most will exaggerate.

A model that knows a user's writing style is nice. A dashboard that remembers preferences is nice. But nice is not power.

Personalization becomes strategic when losing it damages performance. A sales agent that has learned account-specific objections, legal constraints, approval patterns, and follow-up timing may create real switching cost. A finance tool that understands company-specific budget logic, vendor history, and exception rules may be sticky.

The test is simple: after switching, does the customer operate worse for a meaningful period?

If the answer is no, personalization is a feature.

AI as a switching-cost destroyer

AI also helps customers switch away from you.

This part gets less airtime because it is inconvenient. Agents can compare vendors, extract data, rebuild processes, generate training material, and monitor contract renewal windows. A determined competitor can package migration as part of the sale and use AI to make it cheap.

That means defensive strategy cannot rely on friction alone. The product has to keep earning its place.

The best switching costs will be paired with ongoing value creation. Customers stay because leaving would sacrifice a live advantage, not because the basement is full of wiring nobody wants to touch.

Operator test

To test switching costs in an AI business, ask:

  • What does the customer lose if they leave tomorrow?
  • How long would it take them to recover the lost performance?
  • Can AI reduce the migration effort enough to change the decision?
  • Is the lock-in based on customer value or vendor-created friction?
  • Does more usage create better operating memory inside the product?

Strong switching costs feel less like a cage and more like a nervous system.

If the customer removes you, the business can still function. It just functions worse. That is the line worth building toward.

This is part 5 of 10 in Seven Powers in the AI Era.