More Margin with AI Agents: The Next ERP Shift in Multichannel Commerce

Every new sales channel brings a new tool. Every new requirement, a new integration. At some point, a growing commerce business stops running its business and starts running its tech stack. Brooklyn Soap knows this moment. And the company is showing where the road leads next.

At Xentral's K5 Masterclass, Co-CEO Domenico Cipolla sat down with Michael Wintjes, Head of Operations at Brooklyn Soap, to talk through exactly this shift. A journey that starts with clean data and ends with virtual teams. And the question that matters most to merchants: how do you turn the AI promise into real margin?

From a kitchen idea to a multichannel operation 

Brooklyn Soap launches in 2013 as a men's grooming brand. Three founders, one idea that takes shape in a kitchen and comes into focus on a trip to Brooklyn. The ambition: a different take on men's grooming, away from the clichés. In the early days, sales run on a lot of manual work across a handful of channels. Then comes the online shop, then the marketplaces, Amazon, and most recently TikTok. Alongside the B2C business, B2B grows too: getting into retail with chains like Edeka.

Every step adds complexity. And at one point, operations becomes the bottleneck. Michael Wintjes sums it up in a line every merchant knows if they've ever planned without clean inventory data: “Shit in, shit out.” No reliable data, no reliable forecast.

The result: one product unavailable for six months because production kicked in too late. Then a mountain of stock nobody wants. Gut feeling instead of facts. Manageable at low revenue. But as the business grows, every one of those mistakes eats directly into margin.

A centralized system is no longer a selling point 

This is where most ERP stories would end: centralized system implemented, data unified, problem solved. Brooklyn Soap did exactly that. Today, orders from every channel, both B2B and B2C, flow into the system largely on autopilot. Real-time inventory at every moment, pushed directly to Amazon. A lean customer support team instead of a whole department.

That's strong. But it's no longer what makes the difference.

Pulling data into a single place doesn't put you ahead anymore. It's the ticket to entry. Same goes for rule-based automation, for if-then logic from order-to-cash to logistics. That's standard now, not an advantage. The next step is different: intelligent AI automation. Not individual workflows anymore, but agents that take over entire areas of work.

The pattern is nearly the same in every commerce business. No matter how big the support team is, roughly half of tickets are “where's my package?” and another third are refunds. Work that an agent today does better, faster, and with continuous learning. That drops the cost per case and frees up people for the work where they actually matter.

How AI agents take over operational work 

This is exactly where Xentral's Agent Hub comes in: six agents that cover operational areas, not just individual clicks. They handle recurring product questions, work returns with the goal of preventing them rather than just processing them, and pull B2B orders out of emails to create them automatically. Every prevented return is margin kept. Every automatically captured order is one less source of errors.

The key is the human-in-the-loop principle. In the beginning, the agent suggests a reply or action, and a person reviews it. After a few months, once you feel confident, you automate further. AI with training wheels, as much as the team is ready for.

Sound like your next step? See what the Agent Hub can do for your operation.

Even more potential sits between the systems 

Brooklyn Soap is already thinking further down this road. For a commerce business, the ERP is only part of the picture. Marketing, purchasing, logistics all hang off it, a whole chain of operational processes. The interesting question isn't only “what can AI automate inside the ERP?” It's “what can we connect across system boundaries?”

Michael Wintjes makes it concrete with an example. During a Summer Sale, a product outperforms expectations. Stock runs low, and replenishment takes too long. Today, ad spend often keeps running on a product that isn't even available anymore. The budget burns, and demand goes nowhere.

Cross-functionally, it looks different. If an ad is performing well and the product is about to sell out, two things happen: purchasing gets a heads-up that a reorder is needed. And the ad budget automatically shifts to a comparable product that performs well and is actually in stock. Ad budget meets real inventory. Every step depends on a check: which product is strong enough next, and is there enough stock behind it?

That only works if data talks across system boundaries. This is where the best-of-breed model comes in: the best tool for every process, your own WMS, your forecasting, your customer care system. A layer on top brings the data together. Then, when you're running performance marketing, you don't just see conversion and traffic. You see the inventory behind it, the lead time, and the margin. An ERP has to open up to that kind of thinking and make its data accessible, instead of locking it away.

The uncomfortable prerequisite: clean processes first 

Michael Wintjes takes six lessons away from the journey. Centralized system as the backbone. Automation against cost and errors. Decisions based on facts. Data quality in one place. But one thing sits above all the others: solid processes are the future. Only when workflows are documented and stable can an agent build on top of them.

The reverse is true too. Layering an agent on top of a rigid, inaccessible data foundation doesn't solve your problems. It automates them. Data quality is the core prerequisite for being AI-ready, not the agent itself.

And a second point counts just as much: the people. The best system delivers nothing if the team doesn't have the skill, the will, or the appetite. That's the reason for the step-by-step approach, with a person reviewing the agent's suggestions at the start, until the trust is there.

What does this mean for you? 

The ERP category as we know it is breaking open. The system is going from data store to operating system, one that virtual teams run on. The days of six-month implementations that force the business into a rigid corset of processes are ending.

The road doesn't start with the most impressive AI feature. It starts with clean data and stable processes. Brooklyn Soap laid the foundation first, before the team started thinking about virtual teams. That order is where the edge sits, and at the end of it, the margin.

If you want to know where your operation stands on this journey: try it yourself, or talk to us.