In 2021 I wrote an article titled Public sector IT projects are hard because some people live in goddamn vans. I argued that these projects are difficult for a reason that few people outside the sector understand. The government is unable to choose its users. A commercial business can decide that the most difficult 2% of its potential customers do not justify the engineering costs. These businesses can focus their work on the profitable customers instead. A government department that serves citizens cannot do this. The cost of this difference is very high.
The example I used was vehicle registration. Build a form asking how many wheels a vehicle has and you’ll be broadly right, until you meet a trike, a quad bike, or a Reliant Robin. Ask for a home address and you’re fine, until somebody registering on the system actually lives in the van they’re registering.
Five years on I still think that argument’s right. I want to revisit it anyway, because something has shifted underneath it.
What I said then, and what’s still true
The core claim was that complexity in public sector delivery comes from accessibility rather than incompetence. Not accessibility in the narrow sense of screen readers - though that too - but universality. The service has to work for people with no fixed address, no English, no literacy, no smartphone, and for people who are recently bereaved and dealing with probate while barely able to concentrate.
I said the edge cases really do matter, and that getting them right takes an amount of careful research that looks absurd from outside. I also argued that departments outsource partly to move reputational risk somewhere else, which is why the contracts are enormous and why nobody wants to own the outcome.
My position hasn’t really changed. The NHS National Programme for IT still stands as the monument to all this, at over £12.4 billion before it was abandoned. The uncomfortable part of my own argument was that the incentives guarantee this continues, and I’ve seen nothing in recent years to suggest otherwise.
What I got wrong was treating it as a permanent distinction between two sectors.
The exemption was always temporary
On 28 June 2025 the European Accessibility Act became applicable. It covers e-commerce, banking, transport booking, electronic communications, e-books and a good deal else. Crucially it works by market rather than by address: the obligations attach to placing a product or providing a service on the Union market, so where you happen to be incorporated is beside the point. A firm in Leeds selling to consumers in Dublin is in scope.
That said, there is some good news.
Current commitments have a long transition period. According to Article 32, service contracts signed before June 2025 remain valid until they expire. These contracts are limited to a five-year maximum duration, which ends on 28 June 2030. No equipment or services will be removed during this calendar year.
And “microenterprises” providing services are exempt outright: fewer than ten people, turnover or balance sheet under two million euros. Read it carefully though, because it’s stricter than the version doing the rounds currently. The exemption is for services. A microenterprise dealing in products is fully bound by the rules, and gets only a paperwork concession under Article 14(4) excusing it from documenting its disproportionate-burden assessment. It still has to do the assessment itself.
So the assumption I was leaning on in 2021, that commercial software gets to ignore the long tail while public services can’t, has an expiry date on it. Not because government delivery got easier. Because the private sector’s freedom to choose its users is being reduced.
Yes, but you’re in the UK
Fair point - and if every one of your customers is British, the European Accessibility Act really is somebody else’s problem. It isn’t retained law, its transposition deadline fell after we left, and it isn’t listed in Annex 2 of the Withdrawal Agreement, so it doesn’t reach Northern Ireland either. legislation.gov.uk hosts a copy purely so you can cross-reference it.
This situation is difficult because the domestic position is not a lack of obligation. Instead, it is an obligation that no one has ever been required to fulfill.
The Equality Act 2010 has required service providers to make reasonable adjustments since it commenced, and section 29(7) applies that duty to anyone concerned with providing a service. The interesting bit is Schedule 2, paragraph 2(2), where the duty is owed not to a particular person but to “disabled persons generally”. Courts read that as an anticipatory duty: you’re expected to have thought about it before anyone turns up. Compare the employment version in Schedule 8, owed to “an interested disabled person”, individual and reactive. Services got the broader one.
So the duty predates the EAA by fifteen years. What it’s never had is a standard, a deadline, or much in the way of consequences for non-compliance.
There’s no UK statute requiring a private-sector website to meet the Web Content Accessibility Guidelines (WCAG) or anything else. Search the entire statute book for “WCAG” and you get exactly one instrument, and it governs the public sector. Search every published UK judgment and you get nothing at all. No court here has ever cited it (so far).
And when somebody did take a private provider to court over a digital service, they lost. In Maynard v ITV in 2019, a deaf claimant wanted subtitles on ITV Hub. It was rejected, because simulcast counts as a “content service” and Schedule 3 of the Equality Act excludes those from section 29 altogether. He was ordered to pay £11,000 in costs.
That’s the real picture. Fifteen years of duty, no standard attached, no case law behind it, and a cautionary tale for anyone fancying themselves as the test case.
The government knows, incidentally. Its own research, published in July 2025, found the single most requested change among respondents was making accessibility standards mandatory rather than voluntary. I’m guessing that they don’t commission that sort of review about a settled area of law.
So the EAA matters to a UK company for two reasons that have nothing to do with Brussels having any jurisdiction over you. It supplies the one thing the Equality Act never had, which is a date. And your customers who do sell into the EU will push the requirement down the contract chain, because nobody sane builds two versions of the same thing.
Has any of that landed where you work? Go and search your own backlog. I’d be surprised if you find a feature request, and more surprised if you find a budget.
Nobody has budgeted for this
Ask a product team what accessibility work costs and you’ll usually get an answer about colour contrast, alt text and keyboard navigation. All real, all worth doing, and all comfortably inside a sprint or two.
That’s the cheap part. The expensive part is what I was pointing at with the vans.
The expensive part is discovering your address model can’t represent a real address, because 0 Egmont Road exists and your validation rejects a house number of zero. It’s discovering your identity flow assumes a smartphone. It’s discovering your form asks a bereaved person to select the nature of their relationship to the deceased from a dropdown, and the dropdown is missing an option that’ll make somebody’s day materially worse.
None of that gets caught by an automated audit. Do you know how your service behaves for someone with no fixed address? Not in principle. Actually. You find that out by research, with real people, several of whom are guarenteed use your service in a way you’d never have predicted.
The tooling finds the contrast ratios. It doesn’t find the man living in the van.
And that research is the line item that gets cut, because it produces no feature and it’s hard to defend in a roadmap conversation against something that does.
The public sector already solved this, badly and expensively
The organisations with the most experience of building services for people who can’t be designed around are precisely the ones the private sector spent twenty years calling slow and wasteful.
The Government Digital Service ran copywriting audits for dyslexia. Departments built research programmes around users in genuinely difficult circumstances. It looked like gold plating, and some of it was. But a great deal of what read as public sector bloat was the cost of a constraint commercial teams simply didn’t have, and are now acquiring. I said so at the time and I’m obviously pleased to have been right, so discount me accordingly. The procurement bloat was real, but the research wasn’t the bloat.
If you’re a commercial engineering leader looking at this for the first time, my honest advice is to go and read what government published rather than buying a compliance product. The GDS service manual is free, and it was written by people who had to make this work for eighty-odd million users with no ability to pick the easy ones.
So what does this actually change?
For public sector delivery, nothing. It stays hard, expensive and politically radioactive, for the reasons it always did.
For everyone else, I’d put it like this. The next time somebody on your team argues a particular edge case is too rare to be worth handling, that used to be a commercial judgement and it was frequently the right one. If you sell into the EU it’s now a legal position. If you don’t, it’s a position on a fifteen-year-old duty that nobody has yet been made to honour, which is a different bet but still a bet.
Either way it’s being taken by somebody who hasn’t checked how rare the case actually is, because nobody ever funded the research to find out.
Five years on, I’d only add one line to the original piece.
The edge cases were never really edge cases. They were just the users nobody had met yet.