Why B2B Merchandising Needs a Merchant Agent

B2B merchandising looks simple until you actually try to do it.

You go to launch a promotion and realize it touches a pricebook. That pricebook is tied to a buyer group. That buyer group controls catalog visibility. And now you’re checking product data, account eligibility, and whether anything you change is going to break something downstream.

It’s all connected. And most of the time, you’re holding that model in your head while clicking through three or four different places to get anything done.

That’s usually the moment where people start asking if an agent could help.

The Appeal of an Agent (and Where It Gets Risky)

On paper, this is a great use case.

Instead of jumping between systems, you could just ask:

• Can you create a promotion for this set of products?

• Can you make this catalog visible to this buyer group?

• Can you update these merchandising attributes?

And the agent handles the coordination. That part is straightforward. The part that isn’t is everything underneath it.

Because in B2B commerce, those actions aren’t isolated. They’re tied to rules. Pricing rules. Eligibility rules. Visibility rules. And those rules exist for a reason.

So the question isn’t “can an agent do this?”

It’s “how do you let it operate without breaking the system?”

What Actually Changes with a Merchant Agent

The biggest shift isn’t automation. It’s how you interact with the system.

Instead of navigating objects, you’re working at a higher level:

• What’s underperforming?

• What should we promote?

• Where are we missing coverage?

The agent can look across the data, connect the dots, and suggest actions.

But there’s a catch.

It only works as well as the structure underneath it.

If your catalog is messy, you’ll see it.If your KPIs aren’t clear, you’ll feel it.If your data isn’t consistent, the answers won’t hold up.

In that way, the agent ends up being less of a shortcut and more of a mirror.

Where We’ve Seen This Go Wrong

The instinct is to give the agent broad access and let it figure things out. That usually creates more problems than it solves.

You end up in situations where:

• The agent has too much context and returns inconsistent results

• It tries to act across objects it shouldn’t be touching

• Or it produces outputs that technically work, but don’t follow how the business actually operates

There’s a line from one of our internal conversations that stuck with me:

“We open the aperture and say you’ve got everything in an account… that makes some things better and it might make some things worse.”

That’s exactly the tradeoff.

More access makes the agent more powerful. It also makes it less predictable.

The Architecture Is the Product

The teams that get this right don’t start with the agent. They start with boundaries.

A few things matter more than anything else:

Scope the data

Give the agent access to only what it needs. Not the entire commerce model.

Define the domains

Break the work into clear areas like promotions, catalog, and pricing. Don’t treat it as one open-ended assistant.

Control the actions

Be explicit about what the agent is allowed to do and how it does it.

This isn’t about limiting the agent. It’s about making it reliable.

Guardrails Are Doing the Work

In B2B, governance isn’t optional.

Pricing has to be right.Eligibility has to be enforced.Catalog visibility has to follow rules.

The agent should respect all of that.

If anything, it should make those guardrails easier to operate within. Not something you work around.

So What Is a Merchant Agent, Really?

It’s not a replacement for your merchandising team.

It’s not a system of record.

It’s an operational layer that helps you work across a complex system without having to manually connect every piece yourself.

And when it’s built the right way, it does something pretty simple:

It lets teams move faster without losing confidence in what they’re changing.

The Bottom Line

B2B merchandising isn’t getting less complex. If anything, it’s getting more connected. Agents can help. But only if they’re designed to work within that complexity, not ignore it. Otherwise, you’re just moving the problem around.