Take a single defect. A form field whose label is not connected to its input. The fix is one attribute, and it is the same attribute at every point in this post. What changes is when somebody notices, and that turns out to decide almost everything about what it costs.
In the Design File
A designer marks the label and the field as a connected pair. It takes seconds, nobody logs it as accessibility work, and no ticket exists. This is where the design checklist earns its money, and it is the reason three platform vendors have started shipping contrast and structure checks into design tools.
In the Pull Request
A reviewer spots it, the author fixes it before merging, and the whole exchange takes a few minutes. The context is still in the author's head, which explains most of the low cost. They know what the field is for, why it was built that way, and what else touches it.
In an Audit, Six Months Later
Now it is a finding in a report. Somebody has to read the finding, reproduce it, locate the component, work out whether the same problem repeats elsewhere, fix it, test the fix, and get it through a release. That person probably did not write the original code, so every one of those steps starts with reconstruction. The same one-line change has quietly become a small project.
It has also multiplied. A label problem inside a shared form component is one defect in your codebase and forty findings in the report if that component renders on forty pages. That is not a hypothetical shape. In WebAIM's 2026 survey of a million home pages, low contrast text averaged 34 separate instances on a single page, and almost all of those trace back to one decision about a color.
After a Customer Could not Use It
The fix is still one attribute. Around it now sit a customer complaint, a deadline you did not set, an internal escalation, and whatever transaction the customer was trying to complete. If the field was part of a checkout, the standard treats the whole process as failing rather than the one page, so the scope of what has to be re-tested is larger than the thing that broke. The engineering is not harder. Everything surrounding the engineering is.
Where the Multiplier Comes From
- Context decays. The person fixing it later has to rebuild an understanding the original author already had.
- Defects replicate. Between the mistake and its discovery, the pattern gets copied into other components by people who assumed it was correct.
- Process accumulates. What was a commit becomes a ticket, a review, a regression test and a release.
- Urgency removes choice. Work discovered through a complaint has to happen now, whatever it displaces.
What This Means in Practice
The useful conclusion is not that audits are expensive. It is that an audit is the worst place in the world to discover a missing label, because audit time is the only time you have with a person who can find the things no tool reaches. Spending it on defects a machine could have caught in a build is spending the scarce thing on the cheap problem.
So move the machine-checkable work earlier and the audit gets more valuable rather than less. A check in your build pipeline catches repeatable failures on every commit for the cost of setting it up once, and it reaches exactly as far as any other scanner, which by our own count is 10 of the 356 WCAG techniques and documented failures we graded. That narrow slice is precisely the part that should run in a build rather than turn up in a report you paid for. How to set that up is in accessibility testing in CI, and what keeps it working afterwards is in keeping a site accessible.
If you are starting from behind
None of this applies to a site that already has problems everywhere, and pretending otherwise would be useless advice. For that situation the order matters more than the timing. Fix the template before the pages, because one component usually carries the same defect on every page it appears on. Your first month is mapped out in where to start.