Efficiency is not a transformation strategy

The most dangerous thing about a useful tool is that it looks like strategy.

AI transformation has a standard shape: find workflows, apply tools, lay off part of the workforce, and measure efficiency savings. The harder question remains unanswered: what will remain when competitors have done the same thing?

Commerce gives a useful warning. Shopify made it dramatically easier to launch and run an online store. A merchant no longer has to build the storefront, payments stack, app integrations, analytics and logistics connections from scratch. The infrastructure is there. The cost of becoming a merchant has fallen by an order of magnitude.

That is progress. It is not protection.

When all merchants can use similar storefronts, similar payment rails, similar fulfilment tools and similar advertising channels, the visible business becomes easier to copy. More brands launch. More brands chase the same customers. Customer acquisition gets more expensive. Product pages converge. Margins get bid away in discounts, ads and operational imitation.

The merchant has better tools. So does everyone else.

The stronger position sits one layer below. Shopify, Stripe, advertising platforms and logistics providers serve many merchants at once. They see more transactions, more failures, more categories and more demand patterns than any single merchant. The merchant receives capability. But regardless which merchant wins, the infrastructure layer accumulates power.

It's likely AI will do the same to knowledge work. Drafting, summarising, search, analysis, coding assistance, support replies and sales notes are becoming easier to buy. Firms should use them. Refusing them would be wasteful. But if a competitor can buy the same model, install the same tools and copy the same workflow gains within a quarter, the result is not advantage. At best, it is parity.

This is the distinction many AI initiatives blur. A customer-service copilot may reduce handle time. A sales-call summariser may improve follow-up. An internal search tool may save employees from asking the same question twice. These are good uses of AI. They deserve funding and measurement. But they should not be asked to carry the transformation story.

The practical test is simple: after a fast follower copies your tool, what advantage remains? If the answer is 'not much', the use case is efficiency. Do it. Measure it by payback, adoption, cycle time and quality. But do not confuse a cheaper workflow with a stronger strategic position.

Some AI work creates a different problem. It makes the firm faster by moving context, workflow control or customer knowledge into someone else's layer. The tool becomes where work begins. Decisions are shaped there. Data accumulates there. Users build habits there. Over time, the supplier may understand the workflow better than the buyer does.

That is dependency. It is not automatically bad. Most companies should rent large parts of the AI stack, just as merchants rent large parts of the commerce stack. The upside is reducing time-to-value. The danger is failing to notice when the supplier becomes central to the work that defines the business.

The strategic cases are narrower. They are the uses of AI that strengthen something competitors cannot easily copy: owned demand, proprietary customer context, scarce supply, distinctive operations, a central workflow position or a decision loop tied directly to how the company makes money.

The same capability can sit in any of the three categories. A generic sales-call summariser is efficiency. Every sales team can summarise calls. A sales system that turns thousands of objections into pricing changes, product-roadmap decisions and sharper qualification rules is different. The advantage is not the summary. It is the route from customer signal to operating change.

The same is true in support. A copilot that drafts replies faster reduces cost. A support loop that turns repeated complaints into product fixes, better routing, clearer documentation and fewer future tickets strengthens the business. One produces cheaper output. The other improves the system.

The same is true in retail. A campaign generator can produce more copy, images and variants. Useful, but easy to copy. A retailer that connects generation to its own demand history, inventory constraints, returns data, loyalty behaviour and repeat-purchase signals may learn something competitors cannot simply buy. The value is not in generation. It is in the proprietary loop around generation.

This gives leaders a better way to review AI work. Do not begin with the longest list of use cases. Begin with your source of advantage. Where does the company already win? Where does it have demand others cannot reach, data others cannot observe, operations others cannot reproduce, or decisions others cannot improve? Then ask how AI could make that advantage compound.

Any AI transformation portfolio should be mapped in three ways. Efficiency buys parity and should be judged by payback. Dependency rents capability and should be judged by how much context, control and habit moves outside the firm. Strategic loops build defensibility and should be judged by whether the company becomes harder to copy as the system is used.

These labels also imply different levels of attention. Efficiency can be run close to the work. Dependency needs commercial and operating scrutiny before habits harden. Strategic loops belong with executive leaders responsible for how the company wins, because they change what the company learns and how fast it can act on that learning.

The implication is not to slow down AI adoption. It is to stop treating all adoption as the same kind of progress. Some initiatives make the firm cheaper to run. Some make it more dependent. Some make it harder to compete with.

The useful question is not 'where can we add AI?' It is: after everyone has added AI, what will still make us different?