Remediation is the expensive half, and the order it is done in changes both the cost and the exposure. Fixing alphabetically by page, which is what a raw audit export invites, is the most expensive possible order. Fixing by template, then by customer path, then by content, is the cheapest. On the worked example the whole programme is 401 issues, 167 hours and $15,872 at a $95 developer rate, and 126 of those issues sit in nine templates that between them reach every page on the site.
Templates first, because they multiply
Nine templates carrying fourteen failing criteria each is 126 issues, and each template fix reaches 26 pages. Nothing else in the programme has that leverage. Do them first even when the audit lists them below a dramatic single-page failure, and re-scan afterwards so the page-level list shrinks before anybody starts on it.
Then the customer paths, because that is what gets cited
Search, the product or service page, the form, the checkout, contact. These are what a complaint names and what a customer abandons. They are also where a keyboard trap or an unlabelled control costs actual revenue, which is the argument that gets the work scheduled.
Then content, on a rate you can sustain
Alt text, headings, link text and captions across the back catalogue. On the statement sheet, four issues closed a month against 34 open leaves ten at the six-month review, which is a publishable position. Doing this as a one-off sprint and then stopping is how a site returns to where it started within a year.
Questions people ask about accessibility remediation
In-house or a remediation service?
In-house for anything in your own codebase; a service for documents, video captioning and legacy systems nobody can change.
How long does remediation take?
On the worked example 167 hours, which is about a month of one developer, spread over a quarter alongside other work.
What if a fix breaks the design?
Almost none do. Contrast and focus indicators are the two that touch visual design, and both have solutions that keep the brand.