Please Stop Building AI Agents Before You Know What Problem They Solve
Nick Craig, head of go-to-market at Rokt mParticle, said on the eMarketer podcast 'Behind the Numbers' that AI is powerful but problematic when you start with the technology. His prescription is to start with use cases and value, then let AI serve as the enabler. The discussion focused on how marketing AI agents depend entirely on the quality of the data underneath them. The source is emarketer.com/content/marketing-s-ai-agents-only-good-data-they-re-built-on.
The technique here is use case first, technology second. Most marketing teams do the opposite. They procure an AI tool, then hunt around for something useful to do with it. Craig's framing forces you to identify the business problem, define the value of solving it, and only then determine whether AI is the right enabler. The corollary is that AI agents trained on messy or incomplete data will produce messy and incomplete outputs.
Nick Craig leads go-to-market at Rokt mParticle, a customer data and commerce technology company. He shared these views on the eMarketer podcast 'Behind the Numbers,' which covers marketing and commerce trends.
Step 1: Write down one specific marketing task you want to automate, such as segmenting email lists or writing product descriptions. Expected outcome: You have a concrete use case stated in one sentence. Step 2: List the data inputs required to do that task well today, such as customer purchase history or product attributes. Expected outcome: You see exactly what data quality issues might undermine an AI agent before you build one. Step 3: Ask a consumer AI assistant to perform a small version of that task using a sample of your real data pasted into the prompt. Expected outcome: You observe whether the AI output is useful or garbage, which tells you whether your data is ready for a production AI agent.