A Speaker Guide for Event Organizers and Planners

Build a fair AI workflow baseline

Planning an AI event? Use this resource to shape your speaker brief or follow-up; it does not certify a speaker or predict event outcomes.

Before comparing tools, define the finished result and count all the work required to get there.

Choose a bounded task

“Use AI in support” is too broad. “Draft an internal response from an approved policy document for a reviewer” is testable. Specify the input, output, allowed data, reviewer and exception route. Avoid starting with autonomous payments, employment decisions or externally binding commitments.

Compare like with like

1. Choose representative examples, including difficult and incomplete inputs. Remove sensitive data unless its use is authorized.
2. Measure the current workflow through approval, not only the first draft. Record active time separately from elapsed waiting time.
3. Run the proposed process on comparable tasks. Include prompt preparation, source checking, review, correction and failed attempts.
4. Keep acceptance criteria unchanged between the two methods. Record who reviewed each result and what failed.
5. Repeat enough examples to expose variation. A small convenience sample is a pilot observation, not proof of organization-wide benefit.

Use an evidence log

Make the continuation decision explicit

Continue only when the workflow meets your stated quality and risk requirements. If work moves from one team to another, report that transfer rather than claiming it disappeared. If a failure is unacceptable even once, an average time saving cannot compensate for it.

Calculate a scenario after you have observations, then test answer quality separately.

Sources and scope

NIST AI Risk Management Framework provides a voluntary risk-management foundation. Microsoft guidance on evaluation methods distinguishes relevance, groundedness and completeness. The practical worksheets here are editorial recommendations, not NIST certification or vendor benchmarks.