AI Strategy & Leadership Practical insights
Evaluate an agency’s AI claims during the pitch
Assess agency AI claims through reproducible work samples, transparent cost assumptions and demonstrated controls before accepting promised marketing savings.
An agency’s AI pitch should show how its method improves the work you need, under the conditions your organization actually faces. A polished demonstration can establish possibility, but not reliable delivery. Ask for evidence that connects the proposed process to quality, operating controls and a transparent commercial model.
Use a representative work sample
Provide a bounded brief with realistic source material and evaluation criteria. Ask the agency to explain what AI does, what people do and where judgment enters. Anthropic’s evaluation guidance supports defining tasks and criteria before assessing an agent. [1] The procurement equivalent is a comparable sample, not a competition between unrelated showcase projects.
Inspect the hidden effort
Request the time spent preparing inputs, reviewing drafts, correcting failures and managing tools. If the agency claims a saving, ask compared with which baseline and at what accepted quality. Distinguish lower production cost from a lower client fee. Neither automatically proves better campaign performance.
Ask for a difficult case
Use an ambiguous brief, conflicting source or unsupported product claim and observe the response. The agency should show when it stops, requests clarification or escalates. Review how client data is separated, how outputs are approved and who corrects a published error. A slide listing policies is weaker evidence than a demonstration of the actual operating process.
Put the evidence into the agreement
Define deliverables, quality expectations, permitted data use and reporting responsibilities in terms both sides can assess. Use qualified contractual review where needed. Consider a limited engagement with a clear continuation decision before a broad commitment. The strongest agency relationship does not require every production detail to be identical; it requires a credible explanation of where AI adds value and how the agency remains accountable when the work is wrong.
Sources and evidence
Sources checked on 4 October 2026. Proposed workflows and hypothetical examples are editorial analysis.
From insight to practice