Building AI Agents for B2B Marketing

Over the last couple of months, we have been developing prototypes for using AI agents in a B2B marketing context, specifically to identify, track, and act on engagement and buying signals in ways that support sales and business development teams. It has been a fun and insightful exercise, but a number of things have stood out.

The challenge is not lack of data. It is connecting it.

When it comes to AI, the issue is rarely too little data. If anything, the opposite is true. The real challenge is managing large volumes of data from multiple channels and combining them into a complete picture.

From a B2B sales and marketing perspective, that includes data from CRM systems, marketing automation platforms, social media, advertising platforms, and generative AI tools. Each can be valuable in its own right, but they do not always connect in the reliable and consistent way AI agents need.

In some of the initial workflows we created, we ran into blockers because the required data was not accessible. In other cases, the workflow threw errors because the agent could not pull the data it needed due to API integration issues. That will likely improve over time, but it highlights an important point: AI agents are only as effective as the depth, relevance, and consistency of the data they can access.

It is also not surprising to see many sales and marketing platforms reluctant to expose more of this data via APIs, whether to protect their position or until they work out how to monetise it in an agentic world.

Good outcomes depend on well-built skills

Another clear observation from the early prototyping is the challenge of building effective skillsets for AI agents.

The more clearly you can define a skill around an outcome, the more useful the results can be. But it is not enough to tell an agent what you want. You also have to design the steps it should follow.

That matters because in B2B marketing, no two companies or go-to-market strategies are exactly the same. There are common best practices, of course, but the results that really move the needle usually come from recognising signals and behaviours that point to something meaningful. What is a meaningful signal for one company may be meaningless for another.

AI agents will likely follow a similar path. Some B2B workflows and processes can be handled effectively, but human judgement will still be needed to interpret what the data is actually producing. The more complex the buying cycle or proposition, the more important that becomes.

Data ownership and accountability matter more than the technology

This is perhaps the most critical point, and one that often gets overlooked when companies start building AI agents.

You need to own your data and be accountable for the workflows and results the agent produces.

What does this mean in practice? It means:

Data ownership: The data feeding your AI agent should be under your control, not locked away in a vendor's black box. If your agent is identifying in-market accounts or tracking buying signals, you need visibility into how that data is being sourced, processed, and used. You need to understand the quality and reliability of that data. And critically, you need to be able to audit it, correct it, and improve it over time.

Accountability for workflows: Your AI agent operates within workflows you've defined. That means you (or your team) need to own those workflows. You need to understand what the agent is doing at each step, why it's doing it, and what the expected outcomes are. If something breaks or produces unexpected results, you need to be able to trace back to the root cause. A vendor can't do that for you.

Accountability for results: When an AI agent identifies a high-priority prospect or recommends a specific action, someone needs to be accountable for evaluating whether that recommendation was sound. Was the signal real? Did it lead to a meaningful conversation? Was the workflow actually working as intended? This requires continuous feedback loops and refinement. It's not a "set it and forget it" exercise.

The companies getting real value from AI agents are the ones who treat them as tools that require oversight and continuous improvement, not as black boxes that magically solve problems. That requires data ownership and a clear line of accountability for the workflows driving the results.

AI agents still need supervision

The third related point is simple: AI agents make mistakes, and sometimes those mistakes can be spectacular and delivered with confidence.

It can be tempting to treat them as though they think like humans. But reading the room is not their strength, and they are not accountable for the outcomes they produce. They can also brush over mistakes or errors as simple workflow changes, without any context for the impact those changes may have.

They are built to pursue particular outcomes through defined workflows. When those workflows break or are misconfigured, the agent can produce the wrong result, and do so with confidence.

That is why human supervision still matters. AI agents need oversight to make sure they are operating as intended. Continuous testing, feedback, and refinement are not optional. They are essential.

Final thought

AI agents can bring real value to B2B sales and marketing, but only when they are grounded in the right data, trained with the right workflows, managed with the right level of human oversight, and owned by teams who are accountable for the results.

The technology is just one part of the equation. The strategy, discipline, and governance around how you build and manage the agent matters just as much.