start here
A patient sees five or six services. None of them sees the others.
Nobody is responsible for the gaps between them. That's where people get lost.
The four things that break
None of this is about the quality of the medicine. All four happen in the space between services.
How this case study is organised
Three parts, in the order the work happened.
the idea
The patient is the only one who sees the whole thing
The patient lives one course of care, even when five people in three buildings deliver it. The coordination layer is what keeps it connected.
Built from AHRQ's care-coordination framework and the WHO's work on people-centred services. There are over forty published definitions and no agreed one, so this is our reading of it.
what has to stay true
What has to hold true, whatever the plan looks like
Read the sentence. Each one is written the way a patient would say it, with the formal term sitting underneath.
the method
How this care model was designed
We started with the evidence. Each stage got chosen after the one before it.
every stage answers a question, produces something, and changes the design
Select any stage to see what it produced.
the evidence
The research already says what good care has to achieve
There's a lot published on continuity, coordinated care, living with several conditions, stigma, alert fatigue. Our job was translating it into what a real service should do differently on a Tuesday morning.
Seeing the same clinician isn't a nice-to-have
You can't build the plan around BMI alone
A mental-health trigger changes the route, it doesn't raise a score
Prediabetes is a whole-body warning, not a sugar problem
More detection isn't more care
people
Who benefits, who does the work, who carries the risk
Rarely the same person, and that's the whole problem. A design can test beautifully and still fail, because the extra work it creates lands on whoever can't say no.
the patient's route
Most of what fails happens between appointments, not inside them
backstage
Half the work that keeps a service running is invisible
where it breaks
What gets recorded as a patient dropping out usually started as something else
human factors
Every idea we had made work for somebody
So we worked through it in order. Who does the job. What they're trying to get done. What they need to know. What they decide. Where it breaks. Technology came last, on purpose.
| what we looked at | what we found in the current work | what we changed because of it |
|---|---|---|
| Workload | Picked up by whoever notices first, usually a nurse, and it's in nobody's job plan | Made it a visible queue with a counted volume, so it can be staffed |
| Thinking time | Rebuilding four months of a patient's story from three systems, before a fifteen-minute appointment | The summary is ready before they walk in, changes first, source one click away |
| Interruptions | Generic alerts arrive mid-consultation and get dismissed unread | A daily limit per role, with items batched into two review windows |
| Searching | The one thing that would change the decision is in the record, several clicks away | Built the screen around the decision being made, not the data that happened to exist |
| Remembering | Open referrals live in someone's head and a personal list, and vanish at handover | Open items became objects with an owner and an age |
| Handovers | Quality depends on who wrote the note and how tired they were | Drafted from the record, edited by a person, never sent unread |
| Trust | Clinicians ignore a score they can't see inside. They'll argue with reasoning | Evidence sits above every recommendation, and disagreeing takes one tap |
| Who is responsible | Nobody owns the patient between appointments, so nothing happens and nobody's at fault | Every open item has one named owner. Items with no owner escalate on their own |
| Workarounds | Spreadsheets, messaging apps, and a nurse's personal follow-up list | We treated these as requirements. The nurse's list became the coordinator queue |
discovery People invent workarounds exactly where the system fails them. So the last row is a map of the failures.
the workshops
Technology came in session six, not session one
Same core group throughout, so each session built on the last. Every one opened with published evidence. When a session opened with speculation it stalled, and we lost an afternoon that way. When it opened with a finding, the specialists argued with the finding and then committed to something.
| # | question we asked | exercise we ran | artefact it produced | decision it changed |
|---|---|---|---|---|
| 01 | Where does information stop moving? | Rebuild real cases step by step: what happened, what was known, what had to be decided, what was decided, what followed | Map of information gaps | We aimed AI at gaps inside the work instead of bolting automation on top |
| 02 | Which signals mean something only in combination? | Work through ten kinds of data over time, then deliberately add signals that contradict each other | Signal map | Weight improving while engagement drops may mean nothing, so the room measured change instead |
| 03 | Can AI explain why, and not only detect that? | The same event in four different situations, one question: would you step in, and why | How context changes the answer | Detect, explain, suggest what to look into. Every prediction ships with its reasoning |
| 04 | What should AI never do? | Thirty decision cards sorted into five groups, each challenged with: what if the AI gets this wrong | Decision boundary map | A standing rule for the decisions that stay with people, whatever the technology can do |
| 05 | Can it recommend the right next action? | Ranked fifteen situations, then asked what would change the answer | What to do next, ranked | Changed the goal from predicting dropout to recommending the next care action |
| 06 | Where could AI remove work or improve decisions? | Every activity sorted into understand, decide, create, coordinate, predict or personalise | Where the opportunities sit | Coordinating and understanding carried the most value. Predicting the least, which surprised everyone |
| 07 | Which of these is possible with the data? | Each idea checked against the data it needs: source, owner, how often it arrives, quality, how sensitive | What the data can support | Kept the scope to what the data supports |
| 08 | What should we deliberately kill? | Five tests: is the problem real, is AI better than a simple rule, does the data exist, will someone act, is the risk acceptable | Kill list and survivors | Every survivor had a named owner for the workflow |
| 09 | Where can software act on its own? | Map the loop from notice, decide, contact, book, follow up, escalate, close, and find the safe steps | Where software can act | We separated AI that tells you something from AI that moves work along. The value sat in the second |
| 10 | What will patients accept? | The same message written four ways, from a gentle nudge to a blunt statement of risk | Acceptability principles | AI improves care behind the scenes instead of speaking to patients |
| 11 | What would make clinicians act on it? | The same recommendation with more and more explanation. Which would you act on | What has to be explained | Evidence, then interpretation, then recommendation. Never a score followed by an alert |
| 12 | What should be funded? | Scored on value, readiness, risk, likely adoption and evidence | The funding list | Build, pilot or research, each traceable to a factor |
discovery The method and the shape of each artefact, drawn from the sessions.
artefacts
What the work produced
These are the things we worked in. Most of them can be picked up and reused as templates on a completely different problem.
what can be inspected
Principles tell you what good looks like. Factors are the things you can go and check.
Each one is something a patient's care either has or hasn't. Our working model, and we say so everywhere it turns up.
How the factors connect to the principles
The eight above are the qualities you want. The twenty-four below are what we can measure and redesign to get them.
assessed against the model
What was still hard once the care model existed
Each one assessed against a building block in the model above.
The team can already spot more than it has time to act on. That's the limit every idea below had to survive.