Safe Space, NHS mental health care when it moved home
What a private diary for teenagers, a summary for parents, a simple app for older adults, and one clinician tool for callsFor an NHS trust’s young people’s and older adults’ mental health teamsProblem COVID sent clinicians and patients home, and the notes system was built for a room
In 2020 COVID sent an NHS trust’s mental health clinicians home, and the people they cared for were already there. Care that had happened across a desk had to happen through a screen, for the first time, at both ends of the age range. This built the connection in between: a diary private enough for a teenager to use, an app simple enough for someone making their first video call, and one clinician tool made for calls rather than rooms.
24 min read · 16 min skimmed
Three weeks a mark per entry written, one shared by choice. Scrub it.
Write an entry on the left and choose who reads it, and how much. In the middle, what his mum sees: that he wrote, which way the week went, never the words. On the right, the other end of the same service: Margaret, 78, reaching her nurse through a screen for the first time.ONE SERVICE, TWO ENDS OF IT
Open a record and it starts on the brief for the call: what matters to them, their body beside their mood, what you agreed last time. Write a session note and leave one thing to remember. Switch the team to see the same tool at the older adults’ end.USE IT, DON’T READ ABOUT IT
Accessibility: The clinician tool, the three phone apps and the notes system before all pass axe-core’s WCAG 2.1 AA rules at desk and phone, in both settings, and in the 16 states this file links into. Not yet tried with a screen reader. The statement
Weight on a first visit: the clinician dashboard, 349 KB of script and 13 KB of style, compressed, against budgets of 365 and 15 KB; the phone apps, 293 KB of script and 16 KB of style, compressed, against budgets of 310 and 18 KB. The build fails if a product passes its budget.
The states around it, each one a link: a caseload search that finds no one. Each says what is true, names the next step and offers one action.

The clinician tool works on a phone, as it had to once the clinic was a kitchen table: the day’s calls as cards and each record in full. Open it full screen for the whole tool.
Open Safe Space ↗- I owned
- The discovery research, the privacy model, and the design of the diary, the parents’ summary and the clinician tool, through to the beta; and the older adults’ app, proposed on the same model
- With
- The CAMHS clinicians, a youth mental-health nurse who co-facilitated the teen interviews, the trust’s clinical safety officer, digital governance, and two developer leads
- Theirs
- The constraints, set by the trust’s clinical safety officer and digital governance in the feasibility session, and the beta, run by the clinic as part of real care
- Constraints
- Care that moved home overnight in 2020, a teenager’s privacy under Gillick competence, the GDS Service Standard, the trust’s clinical safety and digital governance, and not one minute of added workload for clinicians
- Friction
- Parents wanted to see what their teenagers would not show them, by forty-six points; parents get a summary, never the words
Two ends of one service, sent home at once
I came to the trust’s work through an agency, with access across its projects. The largest was this one. In 2020 COVID moved its mental health care out of clinics and into people’s homes, and the software it ran on was built for a room.
The young people were fluent on their phones, on Instagram and the rest, and guarded about how much of themselves left them. Plenty were not at home on Zoom. They needed somewhere that felt safe enough to write in, like a diary, with full control over how much of it reached a clinician or a parent.
The older adults were the other end of the same spectrum: people who needed to talk to their nurse more often than ever, could not see them, and were asked to use a computer or a phone to do it, often for the first time. They needed the lowest barrier to entry that could be built.
The clinicians were at home too, with four or five people to see an hour. In each call they had to build a connection through a screen, and write down enough of what mattered that the next call, with no waiting room and no corridor to warm it up, could start where this one ended.
And the two were joined. These are services where the body and the mind show up in each other first: a teenager’s sleep before their mood, an older adult’s meals and pain before theirs. Shut in at home, both got worse together, and the old system recorded neither.
Try to record today’s video call. There is no such contact type, a room is mandatory, a phone call is refused as administrative, and the history is a printout to an office nobody is in. This is the kind of system the clinicians had, rebuilt from memory to show the problem; it is not a copy of it.

The old system is a fixed 1024 pixels wide, as it fitted an office monitor. On a phone it is a picture, which is the point.
What the old system assumed, and what the new one does instead. Each line opens the tool at the place it does it.
| The old system assumed | The new tool does |
|---|---|
| A contact happens in a room, and a room is mandatory | A contact is a call, by video or by phone, and a room is the exception. Book one |
| One appointment at a time, in a building | Four or five calls an hour, with the hour on screen: how many, how many done, who is next. See the day |
| The history is a printout | Each call opens on a brief: what has changed since you last spoke, what you agreed, what matters to them. Open Josh’s |
| A note is a free-text box for the record | A note keeps one thing to remember for next time, and the next call opens on it. Open Hannah’s |
| Mood, if anything, typed up afterwards | The body beside the mood, from the person’s own check-ins: sleep, meals, getting out, pain. Open Margaret’s |
| One kind of patient, at a desk in an office | One tool for the young people’s team and the older adults’ team, on a laptop at a kitchen table or on a phone. Switch team |
The system drawn first, then the screens
Before the dashboard: the service as a system, the one gate everything crosses, the four numbers it was built to move, the evidence under each decision, and the spec the rebuilt product is held to on every release.
Read this plate as text
The service drawn as five stations on one line: the diary on the young person’s phone, the record held by the trust as data controller, the clinician dashboard, the clinic and the service’s governance. Between the diary and the record sits the consent gate, crossed entry by entry and withdrawable. A return leg carries recommendation chips and the next booking back to the app.
The dashboard is the fourth thing in that drawing, not the first. The diary is private on the phone, the record holds only what crossed the gate, and the clinic reads the record. Draw it in that order and most of the product decisions make themselves: a clinician cannot see an entry the young person did not share because there is no path on the drawing by which they could.
The return leg is the part the first sketches missed. A session ends in recommendation chips and a booking, and those go back to the app in the same thread as the conversation, which is the only reason the dashboard adds no admin step of its own.
Read this plate as text
A tree with one root and four branches. The root is the beta itself: 21 people used the service as part of real care at one clinic for about three weeks. The branches are the four rollout metrics agreed before it ran: 94 per cent would keep it, four entries a week per patient, all eight parents opened every weekly summary, and five of five clinicians navigated the prototype.
Read this plate as text
Four rows, each running input to finding to decision. Nineteen interview sessions found that teenagers under-report when reported on, so the diary is private by default and shared entry by entry. Seventy survey responses found that added admin kills a clinical tool, so the admin went inside the clinical flow. Three workshops found the teens rejected peer messaging, so it was refused. One feasibility session found that consent settled late costs rewrites, so the privacy model was written first.
The spec below is the part of this that is new. The rebuilt dashboard on this site is held to six statements, each written as an acceptance test before the screen was rebuilt, and each verified by a check that runs before every release of this site. The right-hand column names the check. Spec-driven in the plain sense: the sentence came first, the screen had to satisfy it, and something other than my eye confirms that it still does.
SS-01No diary entry is visible to a clinician unless the young person shared it.
Given a record, when the conversation opens, then only entries marked as shared appear, each stamped with who shared it, and the settings page says who can see an entry.
The shared-entry card in every thread, and the consent list under Settings.THE PRODUCT, EVERY THREAD
SS-02A risk alert cannot be switched off while a clinician holds a caseload.
Given Settings, when the clinician reaches the risk-flag preference, then it reads Always on and carries no switch.
The check “the risk alert cannot be switched off”.THE INTERACTION SUITE, RUN BEFORE EVERY RELEASE OF THIS SITE
SS-03Escalation is one act, is logged, and notifies the safeguarding lead.
Given a flagged record, when Escalate is pressed, then the flag rises to High, the risk history gains a dated entry and a notification is raised.
The checks “Escalate raises the flag to High” and “the escalation is confirmed in a toast”.THE INTERACTION SUITE, RUN BEFORE EVERY RELEASE OF THIS SITE
SS-04The dashboard opens on the next appointment, not on a menu.
Given the day loads, when nothing has been clicked, then the first appointment’s record is on screen with its episode and its risk flag.
The check “opens on the first appointment’s record”.THE INTERACTION SUITE, RUN BEFORE EVERY RELEASE OF THIS SITE
SS-05A booking reaches the young person’s app.
Given Book, when an appointment is saved, then it appears on the week, in notifications, and the confirmation says it was sent to the app.
The checks “Book opens the appointment dialog” and “booking confirms in a toast”.THE INTERACTION SUITE, RUN BEFORE EVERY RELEASE OF THIS SITE
SS-06The whole dashboard works on a phone.
Given a 390-pixel viewport, when any screen opens, then four tabs are on screen, nothing scrolls sideways, and the record reads in full.
The check “nhs phone: 4 bottom tabs, no sideways scroll, no errors”.THE INTERACTION SUITE, RUN BEFORE EVERY RELEASE OF THIS SITE
One hour a fortnight, then the teenager disappears
Read this plate as text
A fortnight drawn as fourteen days. The first is filled, the appointment, and the thirteen after it are empty outlines that run off the edge of the drawing. Between appointments the clinical team had no signal at all, no mood, no sleep, no incident and no contact, unless the young person volunteered it at the next session and remembered it accurately. One hour to see all of it, diagnose against it and decide what happens next, against thirteen days a fortnight reconstructed from memory. The clinical decision was made on a fifty-minute snapshot of a fortnight, and closing that gap without taking the teenager’s privacy to do it is what the whole service is.
The team called it the Triangle of Care, and every corner of it was blind.
Read this plate as text
The Triangle of Care. The young person lives the week seven days a week and says almost none of it out loud, and cannot see that the worry downstairs is about them rather than at them. Parents and carers have no reliable signal between appointments and worry in the dark, able to see nothing at all unless it is volunteered, which it rarely is. Clinicians get about fifty minutes a fortnight and make a decision inside them, with no sight of the thirteen days in between, which is where the week actually happened. Each corner held something the other two needed.
The answer wasn't one app but two connected ones: a mood diary a teenager would actually trust, and a clinician dashboard that surfaces what matters before the session, named "Clinician Data Dashboard", in the GDS habit of saying what a thing does.
The older adults’ service, sent home in the same weeks, needed the same connection with almost none of the screen: one app, one button to join the call, and a phone call always offered instead.
Four jobs, one per person at the table
Written from the interviews, not quoted from them: what each person was trying to get done when the service went home.
- The young personWhen I have had a hard day and cannot say it out loud, I want to write it down privately, so I can choose later whether my clinician sees it.
- The parent or carerWhen I am worried about my child between appointments, I want to know how their week went without reading their diary, so I can help without breaking their trust.
- The clinicianWhen I have five minutes before the next call, I want to see what has changed since we last spoke, so the call starts where the last one ended.
- The older adultWhen the nurse is due to ring, I want one button that connects me, or a phone call if it will not, so I can talk to my nurse without learning a computer.
I asked all three corners before drawing anything
19 interviews With 12 people: clinicians, young people and parents
3 workshops The smallest block, and the one that moved the design
Two months of discovery, qualitative first, checked against numbers, then three co-design workshops: a half-day with 12 clinicians, two hours with 8 teens, and a feasibility session with the safety officer, digital governance and two developer leads.
Read this plate as text
The discovery and validation programme, counted. Nineteen one-to-one interview sessions with twelve people across the three corners: clinicians, teenagers and their parents or carers. Seventy survey responses, thirty from clinicians and forty from young people and parents, run after the interviews rather than before so the questions came from what people had already said. Two co-design workshops, half a day with twelve clinicians and two hours with eight teenagers. One feasibility session with the safety officer, digital governance and two development leads, which is where the constraints were agreed rather than discovered late. Then a live beta with twenty-one people in one clinic over three weeks.
The programme above is how the research ran. This is what it found, and it is the reason the product has the shape it has: two of the three relationships in the triangle are uncontested, and one is not.
Read this plate as text
What each side of the Triangle of Care will allow, from the youth and parent survey, forty responses. Sixty-four per cent of teenagers are comfortable with a clinician reading their entries. Thirty-four per cent are comfortable with a parent seeing their activity. Asked the same question, eighty per cent of parents want exactly that. So two of the three relationships are uncontested and one is not, by forty-six points, inside the same family and about the same screen. That gap is not a preference to be averaged out; it is the thing the product had to be built around, because a diary that satisfies the eighty per cent loses the thirty-four and then has nothing in it to read.
Three scripts, not one
The three corners do not share a problem, so they were not asked the same questions: one set would have produced a false consensus. Each interview was semi-structured, forty-five minutes to an hour, and the young people’s were co-facilitated by a youth mental-health nurse.
The young people
- How they express what they think and feel
- What is hardest about talking to parents, and to clinicians
- How much privacy matters
- What an ideal support system would look like
Parents and carers
- How they see their child’s mental health
- What makes it hard to talk about
- How they balance privacy with wanting to help
- What tools would help
Clinicians
- How they prepare for a session, and what they have or lack
- What gets in the way of a young person’s trust
- How they keep confidentiality and keep parents informed
- What a digital tool would need to do
Both surveys, one scale
Run after the interviews, so the questions came from what people had already said. Each bar is the share who agreed. Three rows became the product’s shape, and the findings below open the product where each shows.
Six findings, and where each one lives now
The deck’s insight cards, grouped by who said it. Each carries its evidence and opens the product at the place it changed.
The young people
88% of 40 worry about who can see what they write. “I don’t mind my therapist seeing my mood tracker, but I’d panic if my mum read my actual journal entries”, a 16-year-old.
Sharing per entry and per reader, the mood alone by defaultAll the young patients said it must not feel like homework or another clinical questionnaire, and that they would stop if it felt cold or like a form.
A diary that opens on today, not on a questionnaireA 17-year-old said it is sometimes easier to write down what they feel than to say it out loud.
Chat instead of talking, offered before every sessionParents and carers
Most would accept a summary rather than the entries, and pushing for detail is what ends the conversation. 34% of young people would let a parent see their activity; 80% of parents want to.
A weekly summary, never the wordsClinicians
82% of 30 struggle to track progress between sessions; 73% rely on what the patient or a parent remembers.
The record opens on what changed since you last spokeThat is all the time clinicians said they had, and 45% of 30 feared information overload.
Highlights first, one chart, the rest foldedThe teens rejected the feature adults assumed they'd love
Everything from the workshops went through a value-versus-complexity matrix, with point 2 of the GDS Service Standard, do less, as the tie-breaker. Peer-to-peer messaging died in the bottom-right quadrant: the teens themselves killed it, on privacy grounds.

High impact · low complexity, kept
High impact · high complexity, deferred
Low impact · low complexity, parked
Low impact · high complexity, rejected
Privacy wasn't a checklist at the end; it shaped the product:
"My Mind Diary, your private space to share, only if you want." THE WAITING-ROOM POSTER LINE, PRIVACY AS THE PITCH, NOT THE FOOTNOTE
Read this plate as text
The privacy boundary drawn through the Triangle of Care. Above the line sits the young person and a diary that is private by default, where nothing leaves on its own. The boundary runs the full width and is solid except for a single gap, which the young person opens entry by entry as an explicit choice. Below the line sit parents and carers, who otherwise worry in the dark, and clinicians, who otherwise decide on a fifty-minute snapshot; both receive only what came through the gap. Three things never cross the line at all: peer messaging, notification theatre, and anything shared silently.
Private by default, shared only by choice
Try the shipped privacy model below, log a week, then decide, entry by entry, what your clinician sees.


The young person's diary
What the clinician sees


The clinician's side opens on today, not on a database: the day's appointments with change chips ("stronger symptoms", "improved mood", "no episodes"), a patient timeline of quality-of-life indicators, and the entries a patient chose to share, right in the communication thread where the conversation happens. Session notes end with recommendation chips (micro-goals, DBT, thought journal) and a follow-up booking, so the admin happens inside the clinical flow.




Where AI does not go
A position rather than a finding: nothing a young person writes is read, summarised or scored by a model, and the products say so where the data is set, in the diary and in the clinician’s settings.
- Why notThe research’s first finding was that privacy is everything to the young people, and 88% of them worried about who could see what they wrote. A model is one more reader, and the one they could not see.
- What it would takeNICE’s evidence standards ask a digital health tool for evidence in proportion to what it does, and a tool that shapes care asks for the most. The constraints this beta ran under were set by the trust’s clinical safety officer and digital governance; a model reading entries would go back through both before it was built.
- Where it might belongIf a model ever helped, the place it would have to earn is the clinician’s five minutes before a call, over what the young person has already chosen to share, and it would say it was a model wherever it spoke.
Tested mid-build, then run in a real clinic
Scenario-based sessions with 5 clinicians and 5 young patients changed the build where the evidence said: an entry-tagging system was added; a "your clinician read your entry" notification was postponed. QA ran on real iPhones and Androids.
- 1 · WireflowsEvery key task in sequenceThe dashboard: log in, the patient list, a patient’s detail with charts. The app: consent, the daily mood, a journal entry, past entries, the sharing settings.
- 2 · ReviewToo many chartsReviewed in the team and with a few friendly stakeholders. The dashboard became one main chart and a few key metrics, after a clinician said: “just show me the highlights”.
- 3 · Mid-fidelityClickable, with real-looking dataFigma for the app, a web prototype for the dashboard, the trust’s components, and fictional records: Alex’s month of moods, a journal line about a rough day at school.
- 4 · UsabilityFive clinicians, five young patientsEntries gained tags and a mood scale that says which way it runs; the read notification was postponed.
- 5 · BetaA live clinic, three weeksThree clinicians, ten patients and eight parents, on the build the four passes left.
| Task | With | What it found |
|---|---|---|
| “Imagine you’re about to see patient Alex. Use the dashboard to get ready and tell us what you gather.” | 5 clinicians | 5 of 5 got ready from the dashboard and found the data useful 95% interval 51% to 100% |
| Reading the mood graph, labelled “Wellbeing Score over 4 weeks” | 5 clinicians | 2 of 5 could not tell whether higher was better 95% interval 12% to 77% Relabelled Mood (1 = low, 10 = high), and the rebuild gives every measure on the record’s chart its direction |
| “You’ve just had a hard day. Can you log that in the app? Now imagine a week passed, can you see how you were feeling over the week?” | 5 young patients | Fed the two changes the sessions made: entries gained tags, and the “your clinician read your entry” notification was postponed |

Two apps, one lane each, and a gate between them
The young person's week and the clinician's day run as separate lanes, and nothing crosses from one to the other without an explicit choice.
Read this plate as text
Two-lane service flow for Safe Space. In the young person’s lane a gentle daily prompt leads to a mood log of one to ten, about ten seconds of work, then a private diary in text, voice or tags, and a weekly chart that is theirs to read first. Between the lanes runs the consent boundary, solid except for a single gate, operated one entry at a time and only ever from the lane above. In the clinician’s lane the caseload opens on today rather than on a search, the pre-session view shows only what was actually shared plus the trend, the session is the hour it was always going to be, and a same-day flag surfaces before that session rather than after it.
Where the service broke, and what closed it
Six things the discovery work found, the design decision taken on each, and what happened next.

A CAMHS team sees a teenager for one hour, then loses sight of them until the next appointment, so clinical decisions were made on a fifty-minute snapshot of a fortnight.
Two connected apps rather than one: a mood diary with an optional prompt, and a dashboard that opens on today's appointments with change chips, a quality-of-life timeline and the entries a patient chose to share.
21people used it as part of real care: 3 clinicians, 10 patients and 8 parents, with the rollout metrics agreed before rollout.BETA · 1 CLINIC · 3 WEEKS
Teenagers will not volunteer how the week really went, and anything that reports on them quietly makes them report less.
Private by default, shared entry by entry. Nothing syncs on a schedule and nothing leaves the diary without the young person choosing it; the consent gate is the product, not a setting inside it.
Privacy became the pitch rather than the footnote: the waiting-room poster led with “your private space to share, only if you want”.THE POSTER LINE, VERBATIM
The adults in the room assumed teenagers wanted to message each other, and a feature list built on that assumption would have swallowed the budget.
Asked all three corners of the triangle first, then ran every workshop idea through a value-versus-complexity matrix with point 2 of the GDS Service Standard ("do less") as the tie-breaker. The teens killed peer-to-peer messaging themselves, on privacy grounds.
19 / 70interviews and survey responses before anything was drawn, and three features refused rather than built.19 SESSIONS WITH 12 PEOPLE · N=70
The mood graph was labelled “Wellbeing Score over 4 weeks”, and two clinicians could not tell whether a higher number meant a better week or a worse one.
Renamed it in plain language as “Mood (1 = low, 10 = high)”, in the GDS habit of saying what a thing does, which is also why the dashboard is called the Clinician Data Dashboard.
5 of 5clinicians then navigated the prototype successfully, naming the timeline graph and the alert flag as the parts they would use.CLINICIAN PILOT
Any tool that adds administration to a clinic does not get used, however good the data in it is.
Put the admin inside the clinical flow: session notes end in recommendation chips (micro-goals, DBT, thought journal) and a follow-up booking, in the same thread where the conversation happened.
The dashboard asks for no separate write-up step, which is the only reason a clinician with a caseload of twenty opens it before a session.DESIGNED WITH THE CLINICAL TEAM
Consent, data controller and identity are usually settled after the screens exist, which is when they start costing rewrites.
Wrote the privacy model first with the trust's clinical safety officer and digital governance: trust as data controller, Gillick competence, documented parental consent, encryption in transit and at rest, email and PIN aligned to NHS identity standards, and adoption by conversation.
The model was fixed before the first screen, so the beta ran inside a real clinic as part of care rather than as a trial bolted onto it.FEASIBILITY SESSION · SAFETY OFFICER & GOVERNANCE
The decisions that mattered were refusals
| It worked if | Result | Status |
|---|---|---|
| It is used as part of real care | 21 people, three clinicians, ten patients and eight parents, at one clinic for about three weeks, with every approval in place | |
| The people using it want to keep it | 94% of beta users said the service should continue; the beta report gives the share, not the number it is a share of | |
| Teenagers write without being nudged | 4 entries a week per patient, in a product with no streaks and no escalating reminders | |
| Parents read the summary | 8 of 8 parents opened every weekly summary | |
| Clinicians can find what they need | 5 of 5 clinicians navigated the prototype |
The beta ran with three clinicians, ten patients and eight parents at one clinic site for about three weeks, with every approval in place. The beta report puts those who said the system should continue to be used at ninety-four per cent; it gives the share, not the number it is a share of. Patients wrote an average of four journal entries a week each, unprompted, in a product that was deliberately built with no streaks and no escalating reminders, and all eight parents opened every weekly summary, which is the one that carries a trend and no entries. The deck carries the full beta report.
What I'd tell another designer taking this on: the product decisions that mattered most were refusals, no peer messaging, no notification theatre, nothing shared without an explicit choice. Trust was the feature. Everything else was UI.
What I would do differently
- Test the riskiest assumption before it reaches the build. Peer messaging was in the build when the teenagers killed it. A paper version in front of them would have cost an afternoon.
- Keep the research records with the work. The breakdown behind the nineteen interview sessions survives only in this portfolio’s own earlier text, which makes it the weakest figure on the site.
Source notes
- Trust identity, staff and patient details anonymised throughout; personas in the full deck are labelled fictional.
- Research figures (interview counts, survey Ns, workshop rosters, beta cohort) are from the project's research plan and rollout deck, shown above.
- Full interactive case study: withcontexts.com/artifacts/safe-space











