Personalised care without the administration
Social care has lived through three technological generations. Each changed how organisations record what happened. None changed how they operate. The fourth one does.
Foreword
When I graduated with an economics degree, social care was not exactly what people had in mind when they asked about my plans. It is not a well-known sector for those of a graduate age.
But I had grown up with my great-grandmother being cared for by our family-run elderly care service. When I was starting out, I spent time with that team: the care professionals, the managers, the people who turned up every day to do work that most people never see. I watched the impact they made. And I thought: if I am going to spend my career in any sector, it is this one.
What followed gave me a perspective on social care that I could not have anticipated. Running and turning around services that were Inadequate. Helping to bring virtual reality training into the sector. Becoming an innovation fellow at an Integrated Care System, where I came to understand both Health and Social Care from the inside: the culture, the pressures, the extraordinary work that happens quietly and without recognition every single day.
Through all of it, one thought never left me: there has to be a way to use technology to put time and resources back into the team and the people they support. Not technology that digitises an existing process and calls it transformation, or a system that replaces a paper form with a screen version of the same form. Something that radically changes how the work gets done.
The shift from paper to digital was real, but it was narrow. The time saved was largely the delta between writing something by hand and typing it. In many cases the burden increased, because digitisation meant more data collected, and more data collected meant more oversight required, and more oversight required meant more of the care professional's time spent on the system rather than the person.
My co-founder, George, and I started Emma because we believe social care can be at the frontier of this technological change. For us, social care is not a passive recipient of AI. It should not get tools designed for other industries, or be a second thought for AI companies. Social care must lead the technology charge, because the nature of care, the richness of its language, and the depth of its human relationships make it one of the most powerful sectors for AI to genuinely change lives.
Finally, the sector has become harder to operate in. The financial pressures are real. The workforce challenges are acute. But the opportunity in front of us right now is unlike anything we have ever seen before.
This white paper is our attempt to set out what that opportunity looks like, and what it means for the people running care organisations today.
Charles Cross | Co-founder, Emma AI
Executive summary
Social care has lived through three technological generations: paper, on-premise systems, and cloud-based care management. Each changed how organisations record what happened. None changed how they operate.
Generation Four does. An AI-led operating model does not digitise the existing workflow, it replaces the logic underneath it. 92% of organisational data is in a format legacy systems cannot touch: voice. Turning the language of care into real-time intelligence means prediction becomes possible, and with it, automated action.
The effect reaches every function of a care organisation: frontline delivery, management, quality, HR, operations, finance. In every case the impact is the same. Less time processing, more time caring.
In a mid-size 500-resident provider, roughly 20% of labour cost sits in non-care-facing work. Emma does not cut that cost. She redirects it, back into the frontline, into coaching, into the people who actually deliver care.
The end state: no administration as a category of work. Every person either delivers care or supports the people who do. Emma does everything else.
This paper sets out what this looks like in practice, how to get there, and why the organisations that move first will have a structural advantage that compounds over time.
Four generations, one axis shift
Every generation of care software changed how the work was recorded. The model underneath stayed the same: someone provides care, someone documents it, someone reviews it.
Social Care Software
Small teams, handwritten notes. No oversight at scale.
Digital forms, local server. Numeric only, input-only.
Same forms, hosted online. The process never changed.
Same model underneath: someone provides care, someone documents it, someone reviews it
A different model entirely
The move from Generation Three to Four is bigger than the three before it combined. Organisations that start thinking AI-first and move now build an advantage that compounds. Everyone else spends the next decade trying to close the gap.
The 92% problem
Care runs on language.
Care notes, handover conversations, supervision discussions, the way someone describes how a resident seemed this morning. Every legacy system was built for numbers and forms.
That is not a failure of the people who built those systems. It is a reflection of what was technically possible at the time. Large language models have changed the conversation.
How much of care is language?
A care professional visiting the same person fourteen times in a fortnight generates maybe seventy data points in a legacy system: five fields per visit. In a care intelligence platform, those same fourteen visits generate fourteen rich, contextual accounts of how that person moved, spoke, ate and seemed.
At scale, that language surfaces what numbers cannot: a change in how someone talks about pain, a withdrawal from social contact three days before a fall, a shift in appetite before a UTI takes hold. These signals have always been there, but they were either hiding under layers of numerical data or lived in someone's head alone. A form could never hold on to this kind of detail.
The AI-first way: care intelligence platform
Emma is a care intelligence platform. Not a better form, and not an AI feature bolted onto a legacy care management system. The intelligence is the foundation. Everything else - data capture, workflow, reporting, compliance - is built on top of it, not the other way around.
And like any intelligence, it needs rules.
The Care Harness, or why Claude could never be Emma
Emma runs on the Care Harness: a proprietary orchestration layer routing multiple frontier AI models, including OpenAI, Google Gemini, Anthropic Claude and ElevenLabs, inside a framework built specifically for health and social care.
The Care Harness
Regulatory
CQC, Ofsted, CIW, CIS and local policy
Orchestration
Transcription, generation, analysis, agent queries
Validation
Every output checked and audit-trailed
Zero retention
Nothing trains on patient data, ever
Emma holds health and social care partnership agreements with every major AI provider it works with. Generic AI tools cannot obtain these agreements because their user base does not exclusively handle patient data. It is a contractual and regulatory distinction with material implications for compliance.
One organisation, one understanding
Each organisation using Emma builds its own picture over time: how it operates, what its standards are, what outstanding care looks like for the people it supports. That understanding runs from the organisation down to region, service and individual - the same context, held everywhere at once.
Input becomes output
Legacy systems only take. They wait for someone to compile the report, notice the risk, spot the opportunity. Emma gives back: insights, predictions, alerts, evidence packs, drafted documents, routed escalations, generated in real time from voice, records and documents alike.
Emma generates a complete, regulatory-aligned audit pack in under five minutes. It checks 100% of care delivery against an organisation's own quality standards, continuously, and it routes safeguarding concerns the moment they appear.
Emma does not wait to be asked.
What changes in an augmented organisation, role by role
The AI-led operating model touches every function of a care organisation. What follows walks through what actually changes in practice, using an illustrative example: a 500-bed residential and supported living provider operating across multiple services with approximately 800 employees.
We call this organisation Harmony Care. It is fictional. The numbers and the pressures are not.
| Function | Today | With Emma | Net change |
|---|---|---|---|
| Care delivery | Documentation interrupts every interaction with the person being supported | Carer speaks naturally; Emma transcribes, files and updates the record in real time, flagging what needs attention. Exceptional practice reaches a manager the same day, not the next appraisal | Outstanding care visible in real time |
| Deputy / RM | Half a day to compile a report | An intelligence briefing generates overnight, accurate to the morning, reviewed in five minutes. One manager can meaningfully oversee more without working more hours | Hours freed weekly. The 24/7 management gap starts to close |
| Quality | Frantic inspection prep, document chasing | Inspection-ready, always. CQC, CIW, CIS, Ofsted and local authority requirements are built into the Care Harness; evidence builds continuously and calls up in under five minutes | Reactive to proactive |
| HR | Scheduling supervisions, chasing documents; development only surfaces in a quarterly meeting | Supervisions captured and transcribed; development needs surface from daily practice. The case for promotion is already built before a manager sits down to make it | Focus on people, not process |
| Operations | Lagged reporting, reactive decisions | Real-time visibility and quality control across every service; the intelligence carries the load human processing used to | Larger patch, same oversight |
| Finance | Manual data aggregation | Referral patterns surface before they show up in occupancy; cost trends flagged early | Forward-looking |
| Marketing / BD | Manual case study creation | Good news stories, family feedback and community engagement surfaced and routed automatically | Continuous pipeline |
| Admin | About 10% of headcount, about 20% of labour cost | Transitions to care-facing, coaching roles | Cost becomes investment |
The end of administration
Administration, as a category of work in care, is at an end.
Not eventually, or at some point in the next few decades. It ends the moment an AI-led operating model becomes the operational foundation. Every task that exists only because a legacy system needs a human to process what it cannot, disappears.
What is left is people: delivering care, supporting the people who deliver it, coaching, developing, connecting.
AI-led operating model
At Harmony Care, the 80 people in non-care-facing roles move into care, coaching and business-development work. Some go frontline. Some become the invested, continuous people-development the sector has always wanted and never had the capacity to deliver. The 20% of labour cost that funded the back office does not vanish from the budget. It becomes the frontline's pay rise.
non-care-facing roles in a 500-resident organisation
of labour cost transitions to care-facing investment
target frontline pay, made achievable by redirecting administration cost
The frontline pay argument
Care professionals are paid, in most cases, at or close to the National Living Wage. That is up 22% in three years, against employer National Insurance now at 15%. The cost base is rising faster than commissioned rates.
Meanwhile, people inside care organisations are paid significantly more than frontline carers to do work a care intelligence platform now does automatically. Redirecting that cost, rather than cutting services or headcount, creates a viable path to £18 an hour on the frontline.
This is the most direct answer the sector has had in a decade to its own recruitment and retention crisis.
The financial reality
Adult social care spending is up 47% since 2015, while care hours delivered are down 9% after inflation. Local authorities face a combined £4bn gap for 2025-26. Only 6.7% meet the Homecare Association's minimum price. The sector needs 470,000 additional posts by 2040, more than the entire NHS workforce in Scotland and Wales combined.
The mathematics of doing more with less are simply becoming impossible.
An AI-led operating model does not patch over the funding gap. It changes the unit economics of delivering care, and the surplus that creates goes back into the frontline: better pay, more hours, real investment in the people doing the work.
Making the transition
The shift happens in stages, and each stage delivers value on its own terms.
The 5 stages of transitioning to an AI-led operating model
-
1
Voice replaces the form.
Care professionals start recording at the point of care by speaking into the app instead of typing after the fact. Administration time falls immediately. -
2
The intelligence starts learning.
Every conversation, every record, every document builds a live picture of how the organisation runs, and Emma keeps updating it constantly. -
3
Manual work becomes automated.
Reports generate themselves. Audit packs compile in ninety seconds. Inspection prep stops being an event and happens in the background. -
4
Care becomes proactive.
As the data deepens, early warning signals get sharper. Proactive measures start being what a normal Tuesday looks like. -
5
People move to where the work actually is.
Non-care-facing roles shift into care, coaching, and growth. Administration stops existing as a category of work here.
The organisations that begin this journey now will have an intelligence advantage that compounds with every passing month. The language data accumulated in year one feeds better predictions in year two. The intelligence built across the first twelve months becomes a proprietary asset that cannot be replicated quickly by a competitor who starts later.
The accuracy question, or why the real risk is not using AI
Care leaders rightly ask about accuracy and AI hallucinations. In regulated environments, documentation errors are not a minor inconvenience. They are a compliance and safety risk.
Emma's evaluation framework was co-developed with Northeastern University London, producing what we believe to be the UK's first evaluation and recursive improvement framework for AI-generated documentation in regulated domains. The methodology operates at field level, assessing every individual question in an AI-generated care document against the source material, classifying errors by type and severity.
The validated results:
- 87.8% of Emma outputs are unchanged by the social care professional after review
- 6.2% is missing information that was not provided
- 6% was changed by the social care professional
Critically, the framework drives recursive improvement. Emma continuously identifies where its outputs need refinement and improves them without retraining on patient data. To our knowledge, no other care platform has an equivalent framework validated in a regulated domain.
Closing: the first honest answer
The sector has spent years being told to do more with less: efficiency drives, digital transformation programmes, technology that promised to change how care is delivered and just changed how it is documented. None of it answered the real question: how do you stop so much of a care organisation's cost and time going to things that are not care?
An AI-led operating model is the first honest answer. By eliminating the category of work that should never have needed people in the first place, we can, for the first time, redirect that capacity into growing care organisations to meet the increasing demand of the future.
Personalised care without the administration is possible.
It is here today, and I cannot wait for the transformation of this sector for the betterment of the people caring and those being supported.
Charles Cross | Co-founder, Emma AI