£410
Estimated daily cost per GP of avoidable, hidden administrative workload: system errors, repeated referrals, and form-filling that adds no clinical value.
Source
Royal College of General Practitioners and Apollo Innovation, December 2025. [1]
Healthcare
AI agents for the admin behind healthcare
We design, build, and run AI agents that take administrative work off practice managers, federations, and operations teams. Built for governance-led healthcare, where the work has to stand up to CQC, the ICO, and patient trust.
The Problem
Administrative work has eaten the day.
Healthcare runs on documentation. Referral letters, discharge summaries, pre-authorisation submissions, appointment chases, regulatory returns. The work is necessary, but it has crowded out the work that only a clinician can do. For independent providers and primary care groups without large internal digital teams, that gap has only widened.
8m
NHS hospital appointments missed in 2023/24, at an estimated cost of £1.2 billion in wasted clinical and administrative time.
Source
NHS England data, with cost analysis by DrDoctor and Deep Medical, 2024. [2]
81%
Of NHS staff who support AI for administrative tasks. That's meaningfully higher than the share who support its use in patient care.
Source
The Health Foundation, national survey of NHS staff and public attitudes to AI in healthcare, July 2024. [3]
The work isn't going away. The question is who, or what, does it.
The pattern is the same whether you're a GP federation chasing rejected referrals, a private hospital group managing pre-authorisation across four insurers, a dental group rebooking the same DNA twice a week, or a care home preparing for inspection.
What We Automate
Workhorse agents for the admin layer.
Below are some examples of workflow areas we can build agents for. The exact steps, integrations, and outputs are designed around your processes during a Discovery session. We work within the administrative layer: agents we build don't make clinical decisions and don't replace clinician judgement.
Referral handling and routing
This could look like an agent that takes inbound referral letters, checks them against your acceptance criteria, and routes them to the right pathway.
- Inbound referral intake from email, post-scan, or e-referral
- Triage against documented acceptance criteria
- Onward routing to the right specialist, pathway, or service
- Acknowledgement letters drafted
- Rejection workflows tracked with reasons captured
- Exceptions and edge cases flagged to a named clinical reviewer before anything moves
The Output
A triaged referral, ready for clinical sign-off, with the administrative trail attached.
Appointment management and reminders
We might build this as an agent that handles booking, reminder cadence, and rescheduling for clinics that lose revenue every time a slot goes unfilled.
- Booking via SMS, web chat, or voice
- Confirmation messaging
- Reminder sequences timed to your DNA pattern
- Risk scoring for likely no-shows
- Rescheduling automation when a slot opens
- Waiting-list backfill prompts
- Insurer pre-authorisation status surfaced where relevant to the appointment
The Output
A fuller clinic, with fewer wasted slots and less reception time spent on the phone.
Insurance pre-authorisation handling
An agent at this stage could pull together the clinical evidence pack, submit it to the insurer's portal, and chase the decision on a defined cadence.
- Pre-authorisation submission packs assembled from clinician notes and the patient record
- Clinical evidence collated against the insurer's requirements
- Portal interactions for Bupa, AXA, Vitality, Aviva, and others
- Status tracking across multiple cases
- Decision chasing on a defined cadence
- Patient and billing-team updates when authorisation lands, is queried, or is declined
The Output
An authorisation status that's current, with the billing team no longer working the phones to find it.
Regulatory documentation drafting
Or an agent that drafts CQC compliance documentation from your operational records, in a structured format your team reviews before it goes anywhere near an inspector.
- Fundamental standards evidence drafted from operational data
- Care plan reviews assembled
- Quality assurance audit packs structured
- Safeguarding documentation collated
- Outdated legislative references and missing personalisation flagged before the draft is sent
- Every output passes to a named human reviewer at the end of the workflow
The Output
An inspection-ready document, drafted in your team's voice, with a human reviewer at the end of every output.
We build only the administrative side.
Healthcare AI is split into administrative tools and clinical tools. Any workflow that involves clinical judgement, urgency assessment, or amendment to a clinical record is outside our scope and into MHRA medical device territory. We'll always tell you, during a Discovery session, where that line falls for your specific process.
How We Work
Design. Build. Run.
Healthcare administrative AI isn't a chatbot. It's a system that has to integrate with your patient management software, work within your governance framework, and produce outputs that survive review.
Free 30 minutes
Design
A 30-minute Agent Discovery session. We map the process: your referral inbox, your pre-auth workflow, your CQC documentation cycle. Then we tell you honestly whether an agent will make a genuine difference. Most processes don't need AI. We say so when that's the case.
Fixed scope
Build
Custom development by senior engineers. Integration with your patient management system (EMIS, SystmOne, Dentally, PMS, care home software). DPIA support. Caldicott alignment. Testing against your own quality bar before the agent sees live data.
Live operation
Run
Ongoing operation. We monitor performance, retrain as your processes change, and keep outputs above your agreed quality threshold. Failed outputs aren't billed.
We do that work end-to-end.
Pricing
You only pay when an agent delivers.
The commercial model is the same across every page, sector, and agent we run. It's reader-facing because it has to be. Your finance director will read it, your COO will read it, and your board will ask about it.
A refundable commitment per agent. Paid at the start of the build. Goes towards the work. Returned in full if we don't deliver to the spec we agreed.
Per-output billing once the agent is live. Unit price is set at scoping and scales with output complexity.
Failed outputs aren't billed. If an agent produces an output that doesn't meet your agreed quality standard, you don't pay for it.
No rate card, no seats, no minimum hours. Pricing is scoped to your processes. Caps available.
Skin in the game. We're paid when the agent works. Not before. Not for time. Not for promises.
Illustrative Scenario
What this looks like in practice.
The scenario below is illustrative, not a named client. It shows the shape of work we'd typically scope with a mid-market private healthcare group during a Discovery session.
01
Today
A private hospital group with four sites and a mixed self-pay and insured patient base spends a meaningful share of practice manager and billing team time on insurance pre-authorisation. Submissions to Bupa, AXA, Vitality, and Aviva each follow slightly different rules. Decisions arrive on different timelines. Patients chase the practice for status updates the practice doesn't yet have. A decent amount of authorisations get queried because the clinical evidence wasn't packaged correctly first time.
02
What we'd build
During a Discovery session, we'd map the actual flow: which insurers, which procedure types, where the evidence currently sits, what the standard exceptions are. We'd build an agent that assembles the submission pack, handles portal interactions, and tracks status across cases. Every authorisation decision still routes through a named member of the billing team before it touches the patient record. The agent doesn't make clinical or coverage decisions. It does the structured work that currently sits on a human's desk.
What changes. Billing team time reclaimed from status chasing. Submissions packaged consistently first time. Patients see authorisation outcomes faster.
No named client. Illustrative only.
Compliance And Governance
Built where the work has to stand up.
Healthcare data is special category data under the UK GDPR. CQC inspects against Well-Led criteria that now explicitly cover AI governance. The MHRA regulates anything that touches clinical judgement. We design every agent with this frame in mind: DPIA support, named accountability, evidenced human oversight, and clear boundaries between administrative work and clinical work.
Boundary discipline
We don't build agents that make clinical decisions. Where a workflow touches the clinical record, we structure it so a named clinician signs off every output. If a process can't be done administratively, we'll tell you it's outside our scope.
Documented accountability
Every agent is designed to be ready for the DPIA your DPO will need to complete. Caldicott alignment is built into the workflow design. Audit trails are produced by default, not retrofitted.
Human oversight by design
No agent we build operates without a defined human review point. The CQC's clear expectation is that AI supports rather than replaces clinical and operational oversight. Our agents are scoped that way from day one.
CQC
Care Quality Commission · Single Assessment Framework, Well-Led criteria, GP Mythbuster 109 on AI use.
ICO
Information Commissioner's Office · UK GDPR, special category data under Article 9, DPIA before deployment.
MHRA
Medicines and Healthcare products Regulatory Agency · medical device classification where AI touches clinical judgement.
NDG
National Data Guardian · the eight Caldicott Principles, Caldicott Guardian appointment expectations.
NHS England
Digital Technology Assessment Criteria (DTAC) · procurement baseline for NHS-contracted work.
DSPT
Data Security and Protection Toolkit · annual completion expected of organisations handling NHS patient data.
Common Questions
Questions we hear from healthcare operations leaders.
The questions below are most likely to come up when we scope agents with private healthcare groups, federations, dental groups, and care home operators.
Are we crossing into MHRA medical device territory?
Probably not, if the agent only handles administrative work: scheduling, communications, pre-authorisation logistics, referral routing, regulatory documentation drafting. The MHRA's medical device scope kicks in when software supports or makes clinical decisions, calculates clinical risk, or creates or amends a clinical record. Ambient scribing, for example, is a medical device. The workflows we typically build aren't, but we treat the classification question seriously and will get a specific opinion in scoping if there's any uncertainty.
How do you handle the DPIA and Caldicott requirements?
Patient data is special category data under Article 9 of the UK GDPR, so a DPIA is legally required before deployment. We don't sign the DPIA for you; that's your DPO's role. But we produce the technical and operational documentation it needs. The same applies to Caldicott alignment: we design agents to use the minimum necessary data, with a clear lawful basis and a transparent purpose. Your Caldicott Guardian remains the accountable role.
Will this integrate with EMIS, SystmOne, our PMS, or our care home software?
Usually, yes. Most healthcare patient management systems have APIs, integration endpoints, or supported data exchange routes. That covers EMIS and SystmOne in primary care, the major private hospital PMS providers, the main dental practice software, and the care home record systems. Where direct integration isn't viable, we work with structured exports or supervised file drops. The integration question gets answered in scoping, before any build commitment.
We're not NHS-funded. Do we still need to think about DTAC?
DTAC is the NHS procurement baseline. It isn't legally required for purely private providers, but it's a useful framework. The questions it asks about clinical safety, data protection, technical assurance, interoperability, and accessibility map closely to what your CQC inspectors, your DPO, and your board will want answered. We treat DTAC as defensible due diligence rather than a tick-box, and recommend it for any provider whose work might come under NHS contracting in future.
Will CQC inspectors accept AI in our operations?
The direction of travel is that they expect to see it. The CQC's GP Mythbuster 109 (July 2025) sets out what good AI governance looks like in primary care, and the draft GP assessment framework is moving toward an expectation that providers are actively using AI well. The flip side is that there has already been an enforcement action against a care provider for ungoverned AI use in compliance documentation. The question isn't whether to use AI. It's whether your governance evidence will survive inspection. We design agents to produce that evidence.
What happens if the agent gets something wrong?
Every agent we build runs against agreed quality standards. Failed outputs aren't billed. That's structural, not a goodwill gesture. More importantly, every workflow has a defined human review point, so a wrong output is caught before it acts on a patient record, a billing decision, or an external submission. We monitor agent performance continuously and retrain when patterns drift. If something does get through, our run team investigates and fixes the root cause.
Ready When You Are Workhorses,
Workhorses,
not experiments.
If your administrative work is reliable, repeatable, and currently sitting on the desks of people who could be doing something more useful, an agent will probably help. If it isn't, we'll tell you in a Discovery session. We'd rather be honest than busy.
30 minutes. No obligation. No sales pitch.
Sources
Stats cited on this page.
- Royal College of General Practitioners and Apollo Innovation, *Uncovering the GP Workload Burden*, December 2025.
- NHS England data on missed hospital appointments 2023/24, with cost analysis published via DrDoctor and Deep Medical, 2024.
- The Health Foundation, national survey of NHS staff and public attitudes to AI in healthcare, July 2024.