Two findings landed in the same week this month. Read together, they are not a contradiction. They are a diagnosis.
The BMJ reported that although every hospital trust in England now has an electronic patient record in place, 79% still rely on pen and paper to some extent for clinical. Days earlier, The King’s Fund put the cost of storing NHS patient records at £239.5m in 2024/25, higher per trust than it was in 2017/18.
Twenty-five years of digitisation. Full EPR coverage. And the paper bill is still climbing.
So here is the question worth asking as the Health Bill returns to Parliament in September to lay the foundation for a Single Patient Record. If the systems are finally in place, why is so much of the value still missing?
The answer is not the software. It is what the software never sees
An electronic record only holds what reaches it. Most of a patient record was never structured data to begin with. It arrives as referral letters, consent forms, lab reports and scanned notes, moving between departments in formats the core systems cannot read. That is dark data, and it has nothing to do with the dark web. It is the information an organisation already owns, already paid for, and cannot see.
NHS England’s latest Digital Maturity Assessment found that although 93% of providers run an EPR, only 30% have fully integrated, bidirectional data flows between systems. Coverage is nearly universal. Connection is not. And you cannot automate what you cannot see.
None of that is new. What is new is the cost of ignoring it. AI did not create the fragmentation. AI made it undeniable. The gap that was a manageable nuisance in 2019 is now the reason a pilot quietly fails, because a model can only reason over what the organisation can actually see.
The bill is being paid in staff time
If this were only a technology problem, it could wait its turn. It cannot, because it is settled in the one currency the NHS has none of.
A national study of resident doctors found they spend four hours on administration for every hour with a patient. The finding that should stop any board sits further down the paper. Doctors working on electronic records spent more of their time on administration than those still on paper, 44% against 37%. Digitising a process is not the same as improving it.
There is no slack to absorb that. The NHS is committed to 2% productivity growth for three more years while deficit support funding is withdrawn and close to 100,000 posts sit unfilled. The next round of savings cannot come from asking people to work faster. It has to come from work that never reaches a person at all.
Ambient voice tools are the most visible answer to that today, and the time they return is real. But they capture the consultation. They do not capture the document that leaves the room.
Which makes the question specific. Where does information actually go dark?
Rarely inside the EPR. It happens at the boundary, where a digital record becomes a physical one or the reverse. Someone prints. Someone scans. Someone attaches a PDF and emails it on.
That boundary is the most heavily used interface in any hospital, and almost nobody treats it as one. File > Print is built into nearly every application ever written and needs no training. Each crossing weakens the link to the system that created it, and something searchable becomes a sheet on a tray or a file nobody can find. Thousands of times a day, across every department, that quiet handoff is where dark data is manufactured.
There is a reason it stays invisible. Print and scan volumes sit with facilities or IT operations, not with anyone accountable for data strategy, so the largest unstructured data pipeline in the building never reaches the roadmap.
Treat the crossing as a beginning, not an end
The instinct here is to procure something. That instinct is part of how twenty-five years of digitisation produced full EPR coverage and four fifths of trusts still on paper. Two questions are worth more than the next platform.
- What do we capture when a document crosses? If the answer is a page count, you are discarding the context you will later pay someone to reconstruct. Who created it, why, under what authority, and what happened to it next is the record an auditor will ask for.
- Which of these steps actually needs judgment? Most do not. Sorting, classifying, extracting and routing are rules, and rules are cheap. Reserve judgment, and the cost of the AI that exercises it, for the places where a person is making the same small call over and over because nobody designed a better way. A referral bounced back for one missing detail is one of those places.
Answer those honestly and a document problem becomes a data architecture problem. The sequencing then follows. Deterministic rules where they are sufficient. AI only where it earns its cost. A named person accountable for anything irreversible. An audit trail written at the moment of action rather than reconstructed afterwards, because governance added later is governance that can be switched off. That is the difference between governance you can prove and governance you can claim.
None of which matters unless the record reaches the person who needs it
That is the test the Single Patient Record will be judged against, and it is a patient test, not an infrastructure one. An app is only as good as the record behind it. A patient opening a referral on their phone does not care which system holds it. They care whether it is there and whether it is right. Every document that went dark at the boundary is a gap in that answer.
So this part of NHS digitisation will not be won in the systems you announce. It will be won in the workflows nobody talks about. Twenty five years of transformation asked staff to abandon what they knew, and four fifths of them are still holding a pen. The alternative is to take the action they already perform thousands of times a day and make it intelligent. Print is not the barrier to NHS digitisation. It is the entry point. The organisations that understand that will not have solved a document problem. They will have found the way into everything that comes next.
By Corey Ercanbrack, Chief Technology Officer at Vasion