1,788 to 26, and what that bought

The question

The marketplace deserved its own consumer-facing brand, separate from the firm operating it. Could I afford to give it one?

What I assumed going in

That a second brand means a rebrand, and a rebrand on a shipped product is expensive enough that you do it once, late, and reluctantly.

What I actually did

Rather than estimate it, I counted. The design system had 1,788 usages of role tokens — names describing a job, like surface or signal — against 26 usages that named a brand colour directly.

That ratio is the whole answer. 26 is a morning's work. It meant the second identity could be one block of CSS refilling the same slots, with zero component edits, and hostname routing picking the brand once at the edge.

What I learned

The expensive work had already been done months earlier, by a decision that looked like over-engineering at the time: styling by role instead of by colour, and an earlier pass converting 94 files to variables. Nobody asked for either. For most of the first year they were pure cost.

The transferable part is not 'use design tokens'. It is that a strategic option had a measurable price, and I could have found that price at any point by counting instead of guessing. I nearly did not — the instinct was to argue about whether a rebrand was worth it.

What is still uncertain

Whether the ratio holds. Every new component is a chance to hardcode a colour, and I have no automated check that the number is not drifting back.

What I would test next

A lint rule that fails the build on a raw colour in component code, the way the later platform does it, and a count printed on every build so the ratio is visible rather than remembered.