Committees set the pace
A new tool passes clinical safety, security, privacy, procurement and change control before anyone uses it. Bain found 35% of provider AI proofs of concept reached production in its April 2025 report.
Hospitals, health systems and NHS trusts
Close the gaps between your EHR tools, your access center and your revenue cycle
We work alongside your EHR vendor and your governance committees to design workflows, connect systems and drive adoption, one service line at a time.
Short answer
How can AI help hospitals and health systems?
Hospitals get the most from AI in documentation, patient access and revenue cycle work, where tasks are bounded and review is built in. EHR-embedded tools handle what happens inside the record, while gaps between systems, teams and committees remain. Our role is workflow design, integration, governance support and adoption, not replacing clinicians or your major vendors.
Who this covers
Organizations in this segment
- Community hospitals and acute-care hospitals
- Multi-hospital health systems and academic medical centers
- Patient access centers, central scheduling and referral management teams
- Revenue cycle, physician relations and service-line marketing teams
- NHS acute trusts and integrated care boards
Challenges
The business problems that hold this segment back
The EHR covers the record, not the gaps around it
Embedded assistants help with notes, inbox messages and chart summaries. Referral faxes, call center queues and payer portals sit outside, and integration is the barrier leaders cite most when scaling.
Access is split across front doors
Central scheduling, service-line phone lines, online forms and referral desks each run their own rules. Patients are transferred, referrals wait for a match and nobody sees the whole journey to a booked visit.
Revenue cycle work spans many payers
Coding, denials, prior authorization and status checks run across dozens of payer rules. HFMA reported only about one in five providers used AI for denials management.
Service lines compete for referrers
Cardiology, orthopedics and oncology each need referring physicians to know what is available and how to refer. Marketing teams must do that without sending patient data to ad platforms.
Where AI fits
Suitable AI, and how mature each application is
Established means widely deployed and well documented. Emerging means real deployments exist but results vary with your data and workflow. Early means pilots or research, not a safe default.
Ambient documentation at scale
June 2025 deployment data showed 62.6% of Epic-using US hospitals had adopted or were implementing ambient AI documentation, which is not a figure for all hospitals. The study calls for research on clinician outcomes, care quality and equity, so we help with rollout and measurement, not outcome claims.
Patient access center support
Bounded routing, scheduling help and after-hours coverage relieve contact centers when escalation and audit trails are designed in. NHS England expects its App triage tool to reach more than 200,000 patients over 12 months, with all app users by April 2028, while traditional routes stay open.
Revenue cycle task support
Coding support, documentation queries, denial triage and status checks are proven at task level. Agents working claims across systems with audit trails are emerging, and the returns quoted in industry scorecards are self-reported, so we test against your own numbers.
Fax and referral packet intake
Document intelligence can classify incoming referral packets, extract fields and route them to the right team. A large US health system announced such work in September 2026, which shows direction, not proven results. Urgent or unmatched items need a human queue.
Physician and service-line outreach
AI can draft referrer updates, summarize which referral sources are growing and help marketing teams test messages. Compliance reviews every claim, and no patient or condition-level data goes to advertising platforms.
Locally built agents inside the EHR
One major EHR vendor was due to begin training its first agent-builder early adopters in October 2026, with broader availability planned for 2027. A marketplace listing is not evidence of clinical effectiveness, so these stay in supervised pilots.
What stays human
Responsibilities that stay with your people
CMIO and clinical safety officer
Approve clinical uses, set review rules and own the safety case, including DCB0160 deployer duties in NHS organizations and review of the supplier's DCB0129 evidence.
AI governance chair and CISO
Tier uses by risk, run vendor and security review and decide what is stopped. Joint Commission and CHAI guidance of 17 September 2025 is guidance, not a binding rule.
Patient access director
Owns escalation, language access and hours of coverage. AI agents support schedulers and never replace the people patients reach in a hurry.
Revenue cycle director and compliance officer
Own appeals, coding audits and any contested denial, and decide what a human must always review.
Improvements
Operational changes you can expect to work on
A shorter path to a booked visit
Measure abandoned calls, days from referral receipt to scheduled visit and referrals lost to other systems.
Healthier revenue cycle metrics
Track days in accounts receivable, first-pass claim acceptance and denials worked per coder, by payer.
Fewer swivel-chair tasks
Count the steps where staff re-key data between systems, then remove the highest-volume ones first.
Documentation adoption you can see
Report scribe use, note turnaround and review rates by department, so leaders know where adoption has stalled.
Commercial value
Growth for a system centers on access and referrals
Hospitals grow when patients reach the right service quickly and referrers trust the handoff. We tie AI work to service-line GTM strategy and access performance, with measures set first.
- Attract: service-line pages, clinician profiles and accurate campus listings through local SEO and GEO.
- Convert: a central access center that answers fast and keeps context when it transfers a caller.
- Retain: follow-up scheduling after discharge and recall for ongoing specialty care.
- Grow: referral source tracking by physician group, so liaisons spend time where referrals stall.
- Protect revenue: cleaner documentation and faster denial work, within payer behavior.
Rules and limits
US and UK rules that apply, and what is not ready yet
General information for planning, not legal advice. Statuses and dates are as published at the time of writing.
United States
- HIPAA applies to every AI data flow that involves PHI. A court partly vacated the OCR tracking bulletin on 20 June 2024, but portal, scheduling and symptom-entry data can still be PHI, so audit pixels on those pages. The Security Rule proposal of 6 January 2025 is not final.
- Section 1557 duties on patient-care decision-support tools began 1 May 2025, subject to court orders. HTI-1 (1 January 2025) added decision-support transparency to certified health IT, a certification requirement, not FDA clearance of an algorithm.
- CMS-0057-F binds payers, with hospital electronic prior authorization reporting from 2027. The claims attachment rule (final 20 March 2026) has a compliance date of 26 May 2028, so faxes are not gone yet.
- California AB 3030 (1 January 2025) and Texas HB 149 (1 January 2026) set AI disclosure duties. Texas SB 1188 requires covered EHRs to be stored in the US from 1 January 2026, which matters for offshore AI workflows.
United Kingdom
- NHS England's ambient scribing guidance (first issued 27 April 2025, updated through 29 July 2026) expects a DPIA, supplier assurance and clinician review. The January 2026 supplier registry is not NHS endorsement. MHRA's 29 July 2026 clarification separates clinician-reviewed drafts from tools that diagnose or act.
- NHS App AI triage followed a trial and is rolling out, not universal. Evaluation of outcomes and operational impact is still called for.
- DTAC covers clinical safety, data protection, security, interoperability and usability for NHS-market products. DSPT is an annual assessment (2025-26 deadline 30 June 2026), and DCB0129 and DCB0160 split clinical safety duties between supplier and deployer.
- The Federated Data Platform had 139 trusts live and 170 hospital trusts signed up at end-May 2026, and signed up is not live. The AI commission's recommendations, accepted by government on 6 October 2026, are policy direction, not enacted requirements.
Not mature or not suitable yet
- Treating scribe installation as proven patient-outcome or equity improvement. Safety, capacity and cash-releasing savings are not yet established.
- Agents that place orders, triage unsupervised or complete a cross-organization care pathway without oversight. Routine agent-to-agent clinical workflows across EHRs are not verified.
- Autonomous medical-necessity denials or resolving every payer dispute without human exception handling. Appropriately qualified clinicians must make those decisions.
Services
The services that fit this segment
01AI StrategyVendor-selection support and use-case sequencing so you add the tools that fit your EHR roadmap, not another disconnected pilot.02Data & IntegrationsClose the gaps between the EHR, referral desks, phones and payer portals with scoped permissions and audit logs.03AI AgentsSupport for access center teams with chat, routing and after-hours coverage, with handoffs designed with schedulers.04AI GovernanceInventory, risk tiering and evidence packs that your committees can review quickly.05AI TrainingChange management, champions and role-based training so adoption is measured, not assumed.
Where to start
Where we would begin
- 01
Map one front door or one service line
Document how a request or referral actually moves across systems, committees and handoffs, and agree the measures before choosing any tool.
- 02
Fill the gap your vendor does not cover
Pick a bounded workflow outside the EHR, such as referral packet intake built with AI automation, and prepare the security and privacy evidence in parallel.
- 03
Build adoption before scale
Run a supervised pilot with department champions and role-based training, report to the governance committee on agreed measures and widen only what holds. We complement your major vendors and do not replace them.
Questions
Frequently asked questions
Should we wait for our EHR vendor's AI tools?
Use embedded tools where they fit, since they sit closest to the record. They do not cover referral faxes, call center queues or payer portals, and adoption still needs owners. We help decide which workflow belongs where.
How common are ambient scribes in hospitals?
In June 2025 data, 62.6% of Epic-using US hospitals had adopted or were implementing ambient AI documentation. That is not all US hospitals, and uptake varied by finances, location and ownership. In the UK, a procurement or pilot is not universal deployment.
What does an NHS trust need before deploying AI?
For supplier products, DTAC assessment, DSPT assurance, the supplier's DCB0129 clinical safety evidence and your own DCB0160 work as the deploying organization, and a DPIA. Ambient tools also need the NHS England guidance steps, and MHRA's boundary applies if a product diagnoses or acts.
Why do hospital AI pilots stall?
Usually for organizational reasons: unclear ownership, integration gaps and slow review cycles. Integration was the most-cited barrier to scaling in a McKinsey survey of 150 US leaders. Planning for committees and plumbing from day one helps.
Can hospital marketing use AI without breaking HIPAA?
Only with conditions: drafts reviewed by compliance, no patient or condition-level data reaching advertising platforms, and a valid HIPAA authorization where PHI is disclosed for advertising. Audit pixels, event names and form fields, not only the cookie banner. In the UK, PECR consent rules apply to tracking and direct marketing.
From the blog
Further reading
Article
A plain-English map of healthcare AI rules for US and UK businesses, October 2026
A dated list of the main US and UK rules that affect healthcare AI in October 2026, with each item's status (in force, guidance, draft or policy direction) and what it means…
Read the articleArticleAmbient scribes: questions to ask before you sign a contract
Ambient scribes are the furthest-along provider AI use case, but the evidence covers draft notes, not capacity or savings. Questions to ask a vendor before you sign.
Read the articleArticleAmbient scribes: questions to ask before you sign a contract
Ambient scribes are the furthest-along provider AI use case, but the evidence covers draft notes, not capacity or savings. Questions to ask a vendor before you sign.
Read the articleStart with one front door
Bring your access center, referral desk or revenue cycle problem. We will map it across your systems and committees and suggest a supervised first step.