Skip to main content

Project guide / Healthcare practices

Make the next care-team step clearer.

A useful practice project helps staff receive, prepare, or review work without losing the patient, source, or accountable owner. Start with one handoff: an intake response, referral packet, draft note, or patient communication. Test the errors that could change care, and measure the work through accepted use rather than counting generated outputs.

Start with the handoff

Name who receives the work and what they must check.

An assistant can prepare a work item without establishing that it is clinically correct or that someone has acted on it. Specify the patient and encounter, original information, unresolved questions, receiving role, and evidence of acceptance. Keep an accessible staff-assisted route available when the system cannot complete the task.

Administrative does not mean harmless
A wrong recipient, lost referral, or incorrect appointment detail can affect access to care. Judge the actual failure path, including delay and disclosure.
Documented does not mean verified
A generated note can omit a negation, confuse speakers, or add an unsupported finding. The clinician needs an inspectable draft and a clear correction path.
Routed does not mean resolved
Putting a task in a queue is not evidence that staff received it, reviewed it, or completed the next step. Keep those states separate.

Four project candidates

Define the useful output before the automation.

These designs need local clinical, privacy, and operational assessment. They assume approved data and named reviewers, and do not provide autonomous triage or care advice.

Prepare an intake response for staff

01

Useful handoff: Help staff find unanswered questions and conflicting patient-provided information.

Starting situation
Patients repeat details across forms and staff must reconstruct which answers are missing, current, or supplied by a representative.
Required sources
Approved questionnaire, patient and respondent identity context, original answers, prior record provenance, and routing rules.
Assisted preparation
Suggest field mappings and a source-linked handoff summary. Preserve declined, unknown, unanswered, and not-applicable states instead of filling them in.
Retained authority
Staff confirm identity and representative authority; qualified clinicians assess clinical content and urgency through the practice's established process.
Review artifact
Original answers beside proposed fields, unresolved conflicts, patient corrections, and an attributable receiving queue.
Failure to test
A blank allergy field becomes none, a representative's answer is attributed to the patient, or symptom text bypasses the approved escalation process.
Trial measure
Compare repeated questions, incorrect field acceptance, unresolved handoffs, accessible completion, and staff effort through review.

Assemble a referral preparation packet

02

Useful handoff: Help the receiving team see what is present, missing, or inconsistent.

Starting situation
Referral details and attachments arrive separately and staff repeatedly check the same packet before it reaches a qualified reviewer.
Required sources
Patient and referral identifiers, original request, approved document checklist, source versions, and destination requirements.
Assisted preparation
Match candidate documents to the checklist, suggest a packet summary, and flag missing attachments or conflicting identifiers without deciding clinical priority.
Retained authority
Staff resolve identity and completeness questions; clinicians decide appropriateness, urgency, and required clinical follow-up.
Review artifact
A document manifest, checklist status, source-linked summary, open questions, and separate receipt and review states.
Failure to test
A similarly named patient's document is included, an old result appears current, or a complete-looking packet is treated as an accepted referral.
Trial measure
Compare wrong-document incidents, repeat requests, missed gaps, receipt-to-review time, and total preparation effort.

Prepare a draft encounter note

03

Useful handoff: Give the clinician an editable record of the encounter to verify.

Starting situation
Documentation preparation is burdensome and the practice can support approved capture, source review, and correction within the clinical workflow.
Required sources
Authorized recording or dictation, exact patient and encounter context, participant information, and approved record sources.
Assisted preparation
Transcribe and structure a draft while marking uncertain speakers and statements. Keep imported history distinct from what was said or observed in the encounter.
Retained authority
The clinician checks material content, edits and signs the exact note. Orders, medication changes, coding, and patient instructions have separate approval paths.
Review artifact
An attributable draft, permitted source references, visible edits, clinician attestation, and a record of later amendments.
Failure to test
A negation or dose is changed, a caregiver's statement becomes a clinician finding, or template completion invents an examination that did not occur.
Trial measure
Compare material omissions and additions, attribution errors, review time, later amendments, and performance across the intended users and settings.

Draft a message from an approved care plan

04

Useful handoff: Help staff prepare a clear communication without changing the clinician's instructions.

Starting situation
An approved plan exists but preparing patient-facing wording and checking the intended recipient takes repeated staff work.
Required sources
Current approved instructions, intended patient or authorized representative, communication preferences, and approved language support.
Assisted preparation
Suggest wording and structure with links to the approved plan. Flag additions, omissions, changed numbers, and questions for review.
Retained authority
Qualified staff approve clinical content and recipient, then authorize sending. The system does not invent advice, reassurance, or an answer to new symptoms.
Review artifact
A marked draft, source plan version, reviewer approval, delivery state, and an owned path for responses.
Failure to test
Simplification changes an instruction, a message goes to an unauthorized representative, or delivery is mistaken for patient understanding.
Trial measure
Compare substantive corrections, recipient errors, clarification rounds, unowned replies, and total effort through approved communication.

Keep the boundaries visible

Preparation, decisions, and actions need separate evidence.

This matrix is a workflow design aid. Regulatory status depends on the actual function, intended use, and jurisdiction; calling software an assistant does not settle the question.

Work stagePossible supportNot establishedAccountable ownerEvidence before progression
Information receiptOrganize answers and attachmentsIdentity, truth or clinical completenessReceiving staff confirm contextOriginal information and unresolved gaps
Referral preparationAssemble a source-linked packetAppropriateness, urgency or acceptanceStaff and clinical referral teamCorrect patient, documents and receiving owner
Clinical documentationProduce an editable draftVerified finding or signed recordAttesting clinicianMaterial content checked and corrections retained
Patient communicationDraft from approved instructionsNew advice or patient understandingAuthorized clinical teamExact content, recipient and response path approved
Decision supportPresent a defined evidence basisDiagnosis, treatment or device exemptionClinical and regulatory ownersIntended-use assessment and appropriate validation
Clinical actionPrepare an authorized work itemPermission to order, prescribe or close careAuthorized clinician and workflow ownerExact authorization, execution state and follow-up

Run a bounded trial

Test the handoff under the conditions that make it difficult.

Use suitable synthetic or properly authorized material first. Define what the trial must show before it can affect real patient work.

  1. 01

    Choose one receiving team

    Name the task, patient group, intended use, exclusions, clinical owner, and operational owner. Specify who acts when the output is incomplete.

  2. 02

    Assess the data and function

    Map capture, storage, retrieval, suppliers, retention, and disclosure. Establish applicable privacy, recording, clinical-safety, and device requirements for the actual function.

  3. 03

    Define accepted work

    Specify the source evidence, material-error categories, reviewer action, and receipt or attestation needed. Keep acceptance separate from a successful API call.

  4. 04

    Exercise failure and downtime

    Test wrong patients, missing records, ambiguous speakers, changed numbers, unavailable reviewers, interrupted delivery, and manual fallback. Verify that work remains owned.

  5. 05

    Review the whole result

    Compare baseline effort, material errors, corrections, delays, and accessible completion. Monitor after changes and narrow or stop the trial when the evidence does not support use.

Before patient use

A useful draft still needs a working safety process.

NHS England's ambient-scribing guidance addresses local adoption, review, and monitoring. HHS and FDA sources address different US obligations. The technical checks below need validation against the practice's own setting.

Bind every item to the right context
Preserve patient, encounter, source version, author, and purpose. Make uncertain identity a stop condition and test access across patients and representatives.
Make review practical
Show the source, uncertainty, and proposed changes where the reviewer works. Allocate time to inspect material content; an approval click alone does not demonstrate a meaningful check.
Keep a staffed alternative
Provide an accessible route when someone declines capture, cannot use the interface, or the system fails. Preserve task ownership and approved escalation arrangements during downtime.
Trace corrections and changes
Keep draft, accepted, amended, and released versions distinct. Assign incident review and downstream correction owners, and reassess affected uses after model or workflow changes.

Questions for the practice

Resolve what the product demonstration leaves open.

These questions support project selection with qualified clinical, privacy, safety, and regulatory owners. They are not a determination of a product's suitability.

Should the first project be administrative?
Often an administrative handoff offers a clearer starting boundary, but assess the actual consequences. Wrong-patient information, missed referrals, and delayed responses can still affect care. Choose a task with an owned fallback and measurable acceptance.
Does a business associate agreement make an AI tool compliant?
No single contract establishes that conclusion. For applicable US HIPAA relationships, HHS describes business-associate obligations alongside risk analysis and safeguards. Confirm the actual service, settings, data flow, and responsibilities; other jurisdictions have different requirements.
Is a clinician review step enough to avoid device regulation?
Do not assume so. FDA's January 2026 guidance describes multiple criteria for certain non-device clinical decision-support functions. Assess the actual intended use and functionality with qualified regulatory owners. Other jurisdictions apply their own frameworks.
Can a draft note update medications or issue instructions?
Keep these as separate actions. A documentation draft has not established the authority or correctness of a medication change, order, or patient instruction. Require the appropriate clinician decision and an attributable action record.
What would make the trial unsuccessful?
More material errors, hidden review work, unowned tasks, inaccessible completion, or delayed corrections can outweigh faster preparation. Include those outcomes in the comparison and stop or narrow the trial if staff cannot reliably inspect and recover the work.

Source basis

Sources behind the control model.

  • 01

    NHS England

    Guidance on AI-enabled ambient scribing products

    Version 3, updated 29 July 2026, for England. Informs adoption, clinician review, monitoring, and local safety assessment; no claimed productivity or patient outcomes are adopted here.

  • 02

    NHS England and MHRA

    Medical device regulation for ambient voice technology products

    29 July 2026 communication on intended functionality, oversight, and local deployment responsibilities. It does not classify the illustrative projects on this page.

  • 03

    US Department of Health and Human Services

    Business Associates

    Reviewed 30 July 2026. Explains HIPAA business-associate relationships, agreements, and exceptions, including a patient-portal AI example.

  • 04

    US Department of Health and Human Services

    Guidance on HIPAA and Cloud Computing

    US guidance on cloud-service responsibilities, risk analysis, agreements, and safeguards. A contract or encryption setting alone does not establish compliance.

  • 05

    US Food and Drug Administration

    Clinical Decision Support Software

    Final guidance issued 29 January 2026. Explains FDA's approach to clinical decision-support functions and the criteria discussed for non-device CDS.

[ WORKFLOW / SYSTEMS AUDIT ]
THE FIRST ENGAGEMENT

Start with one real workflow

A Systems Audit is the usual starting point. If the opportunity is already clear, we can move directly into a focused build.

Show Us the WorkflowStart with the free automation readiness checklist

OBSERVEQUANTIFYDECIDEBUILD