When AI Doesn’t Know, Should It Still Spend Your Budget?

AI agents are moving from creating marketing content to making marketing decisions. That sounds promising until the information they are working with is incomplete. New Appier research raises a question marketers will increasingly have to confront: can an AI recognize when it does not know enough to act?

Imagine asking an AI agent to identify your most valuable audience. It analyzes the available data, produces a convincing segment and recommends where to put the budget. The campaign goes live; results start appearing on the dashboard and nothing immediately looks wrong.

There is just one problem: the data covered only part of the period you asked it to analyze, and the agent never mentioned the gap.

This is the kind of problem behind two research papers announced by Appier this week. As AI moves further into campaign optimization, personalization and customer interactions, the question for marketers is no longer simply how much work an agent can take over. It is whether the agent can recognize when the evidence is not good enough to make a decision.

What if none of the answers are right?

In its first study, None of the Above, Less of the Right, Appier tested 28 language models using multiple-choice questions in which none of the available answers was correct. According to the company, accuracy fell by 30% to 50% when “none of the above” was the correct response. Training with Direct Preference...