1. Test a real clinical workflow, end to end

Start with a fictional or authorised test admission. Enter the presenting history and findings, complete a clinical round, document a procedure, record an important event and prepare a case or discharge summary. Check whether information moves between the appropriate parts of the record without forcing clinicians to re-enter it. Do not place identifiable patient information into an unapproved evaluation environment.

2. Distinguish the clinician's words from AI suggestions

Dictation, transcription and AI-assisted drafting are separate steps. Ask to see the original input, proposed text, review controls and what happens when AI cannot process a note. An acceptable result preserves the medical meaning, flags uncertainty and requires the responsible professional to verify important findings, laterality, medications and the plan. Smooth grammar is not proof of clinical accuracy.

3. Follow authorship, review and correction

Record a draft, verify it, amend an error and identify who made each change. Examine how consultant co-sign or other review requirements are handled where configured. The record must distinguish a proposed draft from a verified note and preserve an intelligible change history. Ask how incorrect entries are corrected without silently replacing what was previously documented.

4. Test hospital and role boundaries

Evaluate at least two separate institution contexts with authorised test users. Confirm that a staff member sees only the patients and functions permitted by their role and institution, and that a hospital administrator is not automatically a platform-wide clinical reader. Ask which permissions the institution controls and which functions require additional approval.

5. Review the discharge document against its sources

Check diagnoses, operation name and date, investigations, the course during hospitalisation, condition at discharge, medications and follow-up instructions. Trace important details back to their source notes. Verify that absent facts stay absent rather than being invented to fill a section. Test different dispositions, not only uncomplicated discharge.

6. Examine governance, security and patient-safety responsibilities

Request the applicable privacy notice, institutional data-processing terms, support arrangements, hosting and processing information, access-control details, retention policy and incident response responsibilities. Ask explicitly how voice and AI providers process information when their features are enabled. Do not infer that a feature implies a particular certification, data residency guarantee or legal compliance claim.

7. Fit the product around your existing hospital systems

List the systems already responsible for patient registration, billing, pharmacy, laboratory reporting and other hospital operations. Define whether documentation will be standalone, connected by approved integrations, or part of a broader institutional rollout. Anvaya is designed primarily as a clinical documentation and workflow companion—not a promise to replace an entire hospital management or ERP stack.

8. Agree on pilot acceptance criteria before rollout

Choose an authorised test team and a small set of measurable process indicators: time spent documenting, proportion of notes requiring correction, completeness of required fields, discharge-summary review workload, user feedback and access-control exceptions. Establish a safe downtime and escalation route. Evaluate outcomes during the pilot; do not accept advertised time savings or error-reduction figures without study evidence.