A business factory that does not learn repeats its mistakes at the speed of its agents. This chapter describes how experience is turned into structure, and how the model is improved without breaking what works.
Lessons become structure
When something goes wrong, the most common response is a reminder: one more line in the instructions («remember that…»). It works for a while and then gets lost among the other lines. The opposite practice is to turn each lesson into structure: a permission that no longer exists, a check that triggers on its own, a template that does not let the step be forgotten, a script that does the delicate part.
The signal to do so is concrete: a correction that had to be made twice is no longer a mistake; it is a lack of structure (chapter 17).
Periodic reviews against the principles
In addition to one-off corrections, it is worth reviewing the whole system against its principles from time to time: whether the rules are still short, whether the roles still have clear work, whether there are mechanisms nobody uses, whether some exception has become habit. The review is done by fresh eyes, without premises (chapter 18), and each finding comes back as a request with its path (chapter 12).
A role that experiments
New ideas need a place to be tried without putting what works at risk. A useful practice is to have a role dedicated to experimenting, with one strict rule: its experiments never touch production. Each experiment ends in one of two ways: it graduates, and becomes a design that follows the normal path (chapter 12), or it is archived, with what was learned written down.
The versioned blueprint
Since what is common to every business factory lives in a versioned blueprint (chapters 1 and 16), an improvement to the model reaches every business factory at once, without touching what each business defined. It is an important advantage of separating what is common from what is particular: each lesson learned in one business can improve all of them.
The business factory as its own first client
The most direct way to know whether a model works is to use it. That is why it is best for whoever designs a model like this to use it in their own operation, with their own roles, their own owner and their own rules. Each friction they find (a request that got lost, a rule that was missing, a screen that confused people) becomes an improvement, and what that organization suffers first, the others later use in improved form.
Understanding before building
A business factory is not set up on a business that is not understood. Before defining roles and routines for a business, it is worth following a path in three steps:
- a diagnosis of the business: how it works today, with evidence;
- a plan: what is worth changing, in what order and how it will be measured;
- the implementation: the roles, the knowledge, the connections and the routines that carry it into operation.
That path deserves a role of its own, the one that turns the diagnosis and the plan into each business's operation. That path, from diagnosis to plan to implementation, is the subject of another book by the author, in preparation: Growth Architecture.
What runs on its own, runs on its own
A routine that waits for a person to be present is not a routine. The practice is that what runs unsupervised has no presence gates and is never left waiting on a process: if it needs something from a person, it asks through the normal path (chapter 9) and carries on with the rest.
Stopping at a safe point
People rest, and so does the business factory. One practice that avoids surprises is a nightly stop at a fixed time:
- nothing irreversible is left half done: a release to production, a migration or a change of secrets is finished or rolled back, never left in the middle;
- all work in progress is saved and pushed, with a note of where it stopped;
- everything that runs on its own is closed;
- in the morning, each role starts clean, from its note (chapter 6).
Tasks with a time limit
A task that drags on unchecked ends up consuming what was saved by delegating it. That is why tasks have a time limit: for example, a notice to the conversation that opened it at 45 minutes, which can extend it once or close it, and an orderly close at 2 hours, in which the task saves everything, leaves what is pending in order and sends a summary so another conversation can continue it. A task with no activity for a while is reported as stuck.
To recap, a business factory improves when it turns lessons into structure, reviews itself against its principles, experiments without touching production, improves every business factory at once through its blueprint, uses itself, understands before building, lets what is automatic run on its own and stops at safe points. The epilogue gathers what is still unresolved.