The first two parts of this series describe a state. A Brand Ontology records what belongs to the brand, how things relate to one another and which rules apply: the digital twin of the brand. Brand Control makes sure that twin survives thousands of daily decisions and grows more accurate along the way.
The more awkward question remains open: how does an organisation get there? Such a twin is easy to describe. It cannot be ordered.
The obvious answer would be an upload. Collect the brand book, the style guide, the product catalogue and a few presentations, push everything into a system and wait ten days. It sounds efficient, and it is incomplete for one simple reason: the most interesting part of a brand's knowledge appears in none of those documents.
“We can know more than we can tell”
Michael Polanyi, a physical chemist and philosopher of science, wrote a line about human knowledge in 1966 that has proved hard to refute: “We can know more than we can tell.” We recognise a face in a crowd without being able to say what gave it away. We ride a bicycle without being able to explain the physics.
The same thing happens in brands every day. Someone looks at a layout and knows within a second that it is wrong. The justification follows afterwards, if anyone asks for one at all. Why a particular visual is never paired with a particular product. Why the brand phrases things more cautiously in one market than another. Why a formally approved statement is nonetheless never used. This is rarely written in the brand book. These are decisions that were made at some point and never recorded. Not out of negligence, but because it was never necessary while the people who knew were sitting in the same room.
Ikujiro Nonaka and Hirotaka Takeuchi examined in the nineties how such tacit knowledge becomes explicit inside organisations. Their conclusion is sobering for any well-ordered folder structure: tacit knowledge first travels through shared practice. It becomes explicit through dialogue and joint reflection. Documents sit at the end of that chain, not at its start.
A brand model therefore does not emerge from gathering material. It emerges while someone actually produces the four-thousandth ad alongside the brand team, asking questions until “we don't do that” becomes a rule a piece of software can apply.
An import is a photograph
The second common misconception concerns the sources. Brand knowledge lives in the style guide, on the website, in the PIM, in the DAM, in Figma files and in product catalogues. These places do not exist to be read out once. They change, and at very different speeds.
A one-off import is therefore a photograph. It describes a single day. Three months later the website has been rewritten, the DAM holds new visuals and the catalogue carries revised wording. The twin knows none of it and continues to claim it is current. An image that reliably knows the wrong things is worse than no image at all.
Sources need the status they were granted long ago in other system landscapes: a register rather than a folder. Which source is connected? When was it last read? Is the connection healthy? What has changed since? This is not the most exciting layer of a product. It does, however, determine whether anything above it can be trusted. It is the connection that turns an image into a twin.
A model does not improve by growing
Connect enough sources and the temptation to model everything arrives quickly. Every new source brings fields, and every field could be turned into an object. Six months later there is a model with three hundred types that nobody maintains and no software can meaningfully apply.
The core of an ontology should stay small and hard. Products, audiences, claims, markets, visuals, rules: a manageable number of dimensions that stays closed and stable. Not every new source may create new object types.
Nothing unstructured disappears in the process. The tone of a campaign, a brand manifesto, an argument about attitude: none of it maps well onto fields, and all of it is valuable. It belongs in a layer alongside, accessible and searchable, but not dressed up as structure. The most expensive modelling mistake is not a missing object. It is an object that pretends to be precise.
Contradictions do not need a referee
Part one argued that contradictions between sources can be made visible and that the decision about them remains a human one. In live operations, that sentence has a hard consequence.
Once a person has confirmed that a particular wording applies, the next sync must not quietly overwrite that decision. Automation that overrules curated knowledge makes curation worthless. Nobody curates twice.
The same principle explains why such a system should leave its source systems alone. Writing the consolidated truth straight back into style guide tools and catalogues is technically appealing. In practice it makes every connected system start working against the others. A clear division of roles serves better: the source systems remain what they are. Consolidation happens in one place, and that place stays able to say at any moment where its knowledge came from and who confirmed it.
Fifteen per cent
That leaves the question of who does this work.
Two answers suggest themselves, and both fall short. One is pure software: a product you buy, connect and use. It fails on Polanyi. A system can only read in what someone wrote down at some point. The decisive part was never written down.
The other is conventional consulting: a project with a start, an end and a day rate. People collect exactly that tacit knowledge, and what remains at the end is a document. The knowledge has been recorded, but it stands still. It does not keep learning, it does not notice when the website changes, and the next approval flows nowhere.
By far the larger share has to be product. The structure, the source connections, the control loop, the interfaces: none of it can be rebuilt for every client, or it is not a platform. As a rough ratio: around eighty-five per cent.
The remaining fifteen are people sitting inside the client's operation. They map the workflows that actually exist rather than the ones described in a slide, build connections to the idiosyncratic systems every company keeps somewhere, and push the first real use cases through until they run. That work produces exactly the tacit knowledge no upload delivers.
What matters is what happens to that knowledge afterwards. If it ends in an invoice, an agency has been built. If it flows back into the product, the platform improves with every client and the share that people have to carry gets smaller.
Depth beats the next model
The reflex to tie the quality of brand communication to the quality of the models in use is understandable and misleading. Models will improve regardless, at intervals no individual company influences.
What a system can do for a particular brand depends on something else: how deeply that brand has been described. For a functional brand, a shallow model goes surprisingly far. Products, claims, markets, rules. For a brand that works mainly through feeling and attitude, that is not enough. There the model has to capture things that resist formalisation: what the brand stands for, which registers it can play, when irony is permitted and when it becomes embarrassing.
That depth does not appear during onboarding. It accumulates over months, out of approvals, corrections and exceptions. This is why the three parts of this series are not three products but a sequence: a twin is established, a control loop keeps it current, and neither becomes deep enough unless someone accompanies the brand while it happens.
A brand that knows who it is has an image of itself. A brand that keeps that image current across thousands of decisions has control. Neither is delivered. Both are built in live operations, together with the people who have been carrying the brand in their heads.
Share it
The thinking behind the system.
NEW4 — Brand Judgment
When may a system decide alone?
When may a system decide alone? It depends on how new, reversible and risky a decision is.

2 — Brand Control
How does a brand manage itself?
How brand knowledge is applied the same way every time, to new assets and existing ones.

1 — Brand Ontology
What does a brand remember?
How a brand becomes something a system can know: rules, products, people and places.

