The question we get asked is how much accessibility adds, and the honest answer is that it depends entirely on when you ask it. Designed in from the first component, it's mostly discipline plus a handful of habits, and it disappears into the noise of ordinary front-end work. Retrofitted onto a finished product, it's a redesign with a compliance deadline attached. Same requirement, wildly different bill — and the difference isn't effort, it's how much already-built work has to be undone.
The cheap version is mostly restraint
Built in, accessibility is a series of decisions that cost nothing extra at the moment you make them. Use the element that already does the job: a real button element arrives with keyboard activation, focus behaviour and the correct role, none of which you then have to write, test or maintain. Check colour contrast while the palette is still being chosen, when changing it is a five-minute conversation about a hex value rather than a rebrand. Give form fields real labels. Decide what focus looks like once, in the design system, instead of deleting the default outline everywhere because it's ugly and never replacing it with anything.
The genuinely additional work is real but small: keyboard paths through custom widgets, a heading structure that reflects the page rather than the type scale, and testing with a screen reader often enough that surprises stay minor. On a project where those habits are in place from the start, none of it reads as a separate workstream. It reads as knowing how to build a front end.
The expensive version is everything downstream of a structure decision
What makes retrofits brutal is that the costly problems aren't styling — they're structural, and they were set a year earlier. A modal built as a plain div that neither traps focus nor returns it, now used in ninety places. A custom select every product surface depends on. A single-page app that never announces navigation, so a screen reader user is silently relocated without being told. A design language whose brand colour fails contrast against its own background.
None of those is fixed in a stylesheet. Each is a component rewrite plus a sweep of every call site, on a codebase whose test coverage probably doesn't reach the behaviour you're about to change. Add the copy that has to be rewritten because it relies on “click the green button on the right,” and the layout that has to change because nothing was built to survive text being enlarged, and you have a project rather than a task.
The overlay trap
There is a well-marketed shortcut here and it deserves naming: the drop-in script that promises to make a site accessible after the fact. They are appealing precisely because the alternative is the project just described. They don't work, for a structural reason — an overlay can only guess at meaning the markup never expressed, and it cannot invent a keyboard path through a component that doesn't have one. Worse, they routinely interfere with the assistive technology a user has already configured to their own preferences, which is why they are actively disliked by many of the people they claim to serve. Buying one converts an engineering problem into a legal position, and it is not a strong legal position.
Keeping it once you have it
The last thing worth saying is that accessibility is not a state you reach, it's a state you maintain. A product that was compliant at launch will drift within two quarters unless something holds the line, and the thing that holds it is not an annual audit. It's automated checks in the pipeline for the mechanical failures — missing labels, contrast, broken landmark structure — plus a short manual pass whenever a new interactive component appears, plus keeping these decisions in the design system so that fixing a component fixes the whole product at once.
That's the argument, and it isn't primarily a moral one. Accessible products are more usable for everyone, they're better structured underneath, and they survive the next redesign more gracefully. But the reason to build it in from day one is simply that it's the only version of the work that's cheap.









