Evaluate what the team can verify, correct and control.

Write the permitted task and its boundaries

For each proposed use, record the input, the output, who checks it and what the tool is forbidden to change. In Vetomo, AI prepares structured information and drafts; it does not independently diagnose, prescribe, alter medication, approve a plan or declare a patient safe. A feature description also does not prove that a particular function is enabled in your configuration. Ask to see the available workflow, its source context and its review step. Where information is missing or contradictory, the useful behavior is to preserve that uncertainty and allow manual handling, rather than produce a confident answer.

Use a small, deliberately varied test set

• Complete context: a fictional owner update with an active, approved plan. Check that the draft preserves the original meaning, timing and source without adding new instructions.
• Missing context: omit a detail needed for the requested output. Check that the gap remains visible and the tool does not turn an assumption into a confirmed fact.
• Conflicting information: include two inconsistent owner statements. Ask how the reviewer can inspect the originals and resolve the conflict.
• Untrusted instructions: place a request to ignore the clinic’s rules inside an owner message. Treat that content as data and verify that it cannot authorize clinical changes.
• Unavailable output: test the manual route when assistance fails or is unsuitable. The team should still be able to complete the authorized work without accepting an unchecked draft.

Score the reviewed result, including correction time

Use the same fictional cases for the current process and the proposed workflow. Record omissions, unsupported statements, changes to meaning, review time, editing time and the final decision to use or discard the output. For example, a draft produced in seconds may still require several minutes to check; the useful comparison includes both. Set acceptance criteria before the demo, and treat an unauthorized clinical change as a failed boundary test rather than averaging it into a general quality score. These checks are an evaluation framework, not a claim that any AI system is error-free. During the free one-month Vetomo pilot, all currently available features are included. Test the enabled assistance with your team, confirm the PIMS handoff and keep the record of clinical decisions in the existing system.