Audit your brand before automating production
Two logo versions, near-identical colours and templates only the designer knows how to adapt become recurring errors when production is automated. An audit establishes what you can delegate to a tool and which decisions still need a designer.
In this article
Audit the assets that take up the team’s time
Start with recurring marketing requests: product announcements, social content and sales presentations. Select approved examples alongside work that needed several rounds of corrections. For each one, retrieve the brief, the source file and the designer’s feedback.
Look for decisions that consume time. A wrong colour needs an authoritative reference. An overflowing headline needs a composition rule. A disputed product claim needs content approval. These problems have different causes; calling them all “inconsistency” makes it harder to resolve them.
Rank the assets by frequency and the effort required to revise them. Focus the audit on the family whose improvement would help the team most often, then extend it to the rest of the brand.
Trace decisions to their sources
The same logo may appear in several folders. Two components may have identical names but different spacing. Distinguish deliberate variations from accidental divergence before generating code. Otherwise, automation can preserve an unresolved mistake as a permanent rule.
A simple audit record is enough: resource, purpose, chosen reference, owner, conflict and required action. “Approved”, “needs clarification” and “retire” are more useful than an impressive overall score. Keep resources without an explicit decision outside the initial automation scope. A team should know which version to trust without comparing folders manually.
Check tokens and their meaning
A token connects a value to a name and a purpose. I separate primitive values, such as a colour, from semantic roles, such as secondary text or a surface. That makes it possible to change appearance without asking every component to reinterpret the palette.
The Design Tokens Community Group format describes an exchange structure for tokens, their types and references. It does not decide which visual choices suit a brand. Contrast, modes, ambiguous names and appropriate usage still need assessment.
| Audit area | Control question | Expected decision |
|---|---|---|
| Colour | Decorative value or functional role? | Name and allowed uses |
| Typography | How does it adapt across channels? | Scale and overflow rules |
| Spacing | Shared system or isolated values? | Common scale and exceptions |
| Component | Which inputs and states exist? | Operating contract |
| Composition | What happens when a title doubles in length? | Adaptation or review request |
Look beyond the component library
At Gladia, the library connects foundations, components and compositions to serve marketing, product and sales. That structure gives us three levels to examine: the base value, the reusable element and its use within a complete asset.
A transcript card contains more than colours and corner radii. It establishes a hierarchy between speaker, text and status. Check what remains legible when text grows, information is missing or the format narrows. Valid variations should be visible in examples, not only described in an instruction. This matters when a person or an agent has to choose among them.
Test with difficult briefs
Build a small set of cases before connecting an agent so that system versions can be compared. Include an ordinary request, incomplete information, a longer translation, an unfamiliar format and a constraint incompatible with the template. The expected result is not always a visual. Sometimes it should be a precise question to the user.
For every case, distinguish automatic checks from judgement. Code can check dimensions, required fields and token references. Message relevance, hierarchy and fidelity to intent need review. A technically valid output can still be poor design. Record the reason for rejection so that the next version addresses the actual problem rather than an arbitrary preference.
Deliver a production decision
At the end of the audit, you should be able to name one family of assets ready for automation, the issues blocking it and an owner for each unresolved decision. A cleaned-up list of styles is not enough to start production.
I recommend tracking each issue alongside its effect on deliverables and an example that can verify the fix. Inconsistent spacing in a shared component should take priority over a rarely used exception. Marketing supplies the actual frequency of requests; design assesses which decisions can be standardised.
Finish with a validation brief. Marketing supplies the content, design reviews the result and the technical owner checks that the source remains editable. That test shows exactly where to invest before coding the brand.
