How to Evaluate Agency AI Propositions: A Procurement Guide

Procurement teams are being asked to assess more agency proposals built around AI services, proprietary tools and workflow automation. That creates new risks and new questions: how to validate data sources, test automation claims, measure business value, and make sure AI is used responsibly.

This guide draws on a recent ISBA Marketing Procurement Forum session led by mediasense. It explains how to evaluate agency AI propositions, separate hype from real value, and build a practical framework for buying AI-enabled services with confidence.

Three key questions procurement teams should ask when reviewing agency AI propositions:

Is the platform the value, or is the intelligence the value?

Many agencies now place AI platforms, data products and technology stacks at the centre of their pitch. That can make the offer sound more advanced, but it can also make the real value harder to see. Platforms change over time. What matters more is the intelligence being built for your business: the insights, models, decisioning logic and learning that improve performance over time.

Takeaway: Ask what intelligence is being created, how it improves outcomes, and whether your business retains the benefit if the platform changes.

Is the AI creating value for your business, or mainly for the agency?

AI systems optimise towards the goals, incentives and data they are given. If governance is weak, they may improve agency efficiency or margin without improving your business outcomes. Procurement teams should put independent auditing and oversight in place to test the impact of AI-driven decisions, monitor risk, and feed results back into future activity.

Takeaway: If you cannot independently measure what the AI is optimising for, you cannot be sure it is working in your favour.

Who owns the intelligence, and what happens when the relationship ends?

As agency tools become more deeply integrated and more data is accumulated, the intelligence created through the relationship becomes more valuable. That also increases dependency and switching costs. Lock-in is no longer only commercial or relational; it can sit inside models, workflows and learned behaviours. Yet many agency contracts are still too short and too lightly protected for that level of dependency.

Takeaway: Define ownership, access, portability and exit terms from the start so your business does not lose the intelligence it funded.