When an event supplier proposes AI, start with the task and the evidence it will provide. Then test it in a controlled pilot before wider adoption.
Separate the use cases
Generating text, recommending meetings and forecasting attendance are different tasks. NIST’s generative AI profile covers systems creating synthetic content and identifies risks including confidently incorrect output and privacy leakage. Make that distinction before buying a package advertised simply as “AI”.
- For draft communications, request an editable output and a named reviewer.
- For meeting recommendations, ask what data is used and how participants can decline a suggestion.
- For forecasts, request the evaluation method, error history and fallback procedure.
Design a controlled content pilot
Start by drafting visitor answers from an approved information sheet. Prepare reference answers first. Include questions about hours, admission and accessibility, plus questions the sheet cannot answer. Record invented information, omitted conditions and correction time.
Test Arabic and English separately. Include the dialects you actually need and have competent reviewers assess the results. Judge dialect fluency from those results rather than a polished demonstration, and check any speaker biography the tool drafts against approved information.
Review data and supplier arrangements
NIST presents its AI Risk Management Framework as voluntary guidance for evaluating trustworthiness. Use it as a reference for questions about accountability and evaluation. Before uploading attendee records, ask who can access them, whether they are used for training, where they are processed and how deletion works. Map those data flows first with our guide to event data privacy and security.
Measure the complete workflow
Compare the pilot with the current process: preparation time, correction time, accepted outputs and total supplier cost. For matchmaking, ask participants whether a suggested meeting was relevant. Treat claimed savings or better leads as a hypothesis to test against a defined baseline.
Decide whether to extend the pilot
Agree acceptance conditions before testing. Keep a person responsible for approving public information and stopping the system when needed. Extend use when the evidence supports the decision. Refer any crowd-safety use to a separate competent review.
