How Britain’s digital transformation is leaving millions behind and why fixing it makes services better for all of us.
There is a version of the AI revolution that should excite every policymaker, public servant and technologist in the country. Smarter services. Earlier intervention. Systems that can spot a person in crisis before they ever pick up the phone. Systems that catch people before they fall.
We are not living in that version yet.
Instead, we are living in a world where the tools being built at speed and scale are, in too many cases, quietly making an already stubborn problem worse. What was once called the Digital Divide, the gap between those who can access and navigate technology and those who cannot, is turning into something the analysts at Transform have started calling the AI Abyss. And unlike the Digital Divide, which at least had a name everyone agreed was a problem, the AI Abyss risks becoming invisible. Because if you are not in the data, you are not in the model. And if you are not in the model, the system is not designed for you.
Research suggests that a substantial minority of people across the UK lack access, skills, or financial means to engage with digital services at a basic level. The economic cost of failing to close that gap has been estimated at over £13.6 billion. But behind that figure lies a person – someone whose benefits claim fell through because a form was confusing, whose mental health crisis escalated because the service only existed online, whose housing application was rejected because nobody ever looked at the full picture of their circumstances.
The 80/20 problem we don’t talk about
Most digital transformation programmes are built around the majority. This is not bias. It is logical. The standard framework for designing any service, public or private, is to reach an MVP that delivers the best outcome for the greatest number of people in the most efficient way. The 80:20 rule. Get the majority right, then fix the edges.
But the edges do not wait. When a service fails someone in financial hardship, or someone who does not speak English as a first language, or someone navigating the system through grief, addiction or disability, the harm does not sit quietly in a backlog. It escalates. It becomes a crisis. And crises cost more than early intervention ever would.
The more energising realisation is that designing for the edges does not make services worse for the majority. It makes them better. More compassionate and more resilient. The subtitles originally designed for the deaf made video accessible to people watching in loud environments. The kerb cuts designed for wheelchair users became indispensable for pushchairs, delivery riders, and everyone carrying luggage. Design for the hardest case, and you solve for everyone.
Why is this so hard
Three things stand between good intentions and meaningful progress, and none of them is primarily about technology.
1. Visibility
Signals of vulnerability already exist in almost every system: in arrears data, in missed appointments, in repeat contacts, in the notes that frontline workers write, and nobody ever analyses at scale. The problem is not a lack of data. It is that the data lives in silos, and understandable concerns about privacy and consent have, in many organisations, led not to responsible data use but to paralysis. The insight is there. The picture is not.
2. Trust
You cannot design with communities you have never sat with. Co-design is a phrase that is common, but authentic participation, the kind that gives marginalised people genuine influence over the services that shape their lives, remains rare. It requires time, psychological safety, and a willingness to share power that procurement cycles and delivery timelines rarely accommodate.
3. Coherence
AI investment without data readiness is not progress, but the amplification of existing bias. Too many transformation programmes are a collection of pilots and platforms that do not talk to each other. The result is organisations that feel innovative but cannot demonstrate sustained improvement for the people with the most complex needs.
None of this requires abandoning ambition. It requires redirecting it. The question is not whether to invest in AI. It is about building AI that can see the whole, not just the comfortable middle.
What would it look like to start not with the assumed average user, but with the person most likely to fall through the cracks? To ask, before a single line of code is written: who is this going to fail, and why? Designing from the edges in, rather than the centre out, is not a harder way to build services. It is a smarter one.
