01
Everyone rents the same models
Models get better and cheaper every few months, and every company gets the upgrade on the same day. A system that is only a model behind a screen is a feature a competitor can match next quarter.
So the question for a 2027 budget isn’t which model. It’s what each system leaves behind once it runs.
02
Two starting points
If you already own data nobody else has, the moat exists. The risk is that someone else builds the intelligence on top of it and sells that to your customers. We wrote about that case separately: stop selling access to data.
If you don’t, you build the moat yourself, by running the loop.
03
The loop
A first system does a job and records what happens:the questions people ask, the corrections experts make, the decisions taken with it and what came of them.
The record gets analyzed:what people need that nobody planned for, where the answers are weak, which decisions pay.
The analysis points to the next system,which starts from the first one’s data, definitions and users.
The next system records more,and the loop turns again.
Each turn costs less than the last, which is the argument of our pillar on the compounding value of AI systems. This post is about the other half: each turn is also harder to copy, because it runs on what only your company has seen.
04
Why it’s hard to copy
The data is yours.Nobody else saw your customers’ questions or your experts’ corrections.
The definitions are yours.What a term means in your business, agreed and written down.
The habit is yours.People who know what to record, and owners who act on what it shows.
A competitor can hire the same vendor and rent the same model. It can’t rent three years of your loop.
05
What the loop needs
Owners.Someone decides what each system must record. Someone acts on what it shows.
Capacity.Once systems produce data, the work becomes many small analyses and quick feasibility studies, run side by side. Most organizations staff for one build at a time. The loop needs a stream.
The right level.The data from use has to stay with you. On a subscription, much of it sits with the vendor; at Level 4 it is all yours. Our Practice Note on which level keeps your data goes through it.
06
The trade-offs
Cost.The levels that keep your data, 3 and 4, cost more to build and to run.
Time.The first turn of the loop takes months, not weeks.
People.Analysis and feasibility work in parallel needs more hands than a single build.
Independence.Owning the system means owning its upkeep.
Rules.Every new use of data brings a privacy review, and in some industries a question of where the data may go.
The loop that never starts.Without owners, systems record data nobody looks at.
None of these is a reason not to start. They are the reasons to decide before the first build, not after the third.
07
Where to start
Before the next system is funded, write down what it must leave behind: which questions, corrections, decisions and outcomes it records, who owns them, and which system they could feed.
If nobody can answer, the system is a cost line. If somebody can, it is the first turn of the loop.
Related use cases
Differentiation and defenseAn asset competitors cannot buy.Decision intelligenceOne answer across systems that disagree.