Omni talk interview: Why an AI agent is like a new employee on day one

This is an edited version of the Omni Talk episode "The AI Employee You Should Hire on Day One". Chris Walton of Omni Talk talks with Arber Sejdiji, Co-Founder and CEO of Zenline AI, about how merchandising teams should think about deploying agentic AI. Watch the full video on Omni Talk.

Chris Walton: Arber, you said something to me at Shoptalk that I'm never going to forget, and I've told everyone I can about it. You said an AI agent is like your new employee on day one. What exactly did you mean?

Arber Sejdiji: The concept is quite simple. Think of graduates who join a company. On day one, they are not productive. Why not? They have five years of graduate and postgraduate education, they are highly intelligent and highly capable. The answer is context.

They don't know the company's abbreviations. They don't know which products you already offer, what the processes are, or what the workflow is to come to an assortment recommendation. They have to learn all of that. That is why you have junior positions and senior positions, and at some point you become Head of Merchandise Planning. Over time you accumulate context, or what I now call learnings.

An agent starts on day one too, and it knows nothing about your organization. So one needs to understand the learnings that a Category Manager or the Head of Merchandise Planning has gathered over five, six, seven years, across all the promotions, and build a company brain from them. We call it "context engineering". With that context, the agent becomes as productive as a Senior Merchandise Planner after a few years.

Chris Walton: That's such a great analogy. As a former Harvard MBA, I wanted to think I knew everything on my first day at Target. I had absolutely no idea what I was doing. What in your background led you to think about the problem this way?

Arber Sejdiji: I'm right in between an engineer and a business guy. I studied engineering at ETH Zurich and did half a year of research on AI in Cambridge. Then I wanted to dive into the business world and spent three years at Boston Consulting Group. Our entire work there was about bringing technology and AI into business processes and strategy, so companies could make faster decisions and become more digital.

As an engineer, you think very technically: how do I solve this problem with technology? At BCG we always said that every digital transformation is 80% processes and people and only 20% tech. That was our main advantage when we started Zenline. However good the models are – and Claude and GPT are very good – without the proper context of how an organization works there's no way to improve it.

Chris Walton: It's the same with consultants. They come in without any knowledge of how the organization works and have to learn it too. So what is Zenline AI, and what are you trying to do for retail?

Arber Sejdiji: At BCG, we always said the two most important questions are: what do shoppers want, and what are they willing to pay? At most retailers these questions are answered in a very unanalytical way.

Through Anthropic or OpenAI, for the first time, you can use numerical and tabular data together with TikTok posts, transcripts of TikTok videos, social media and everything else on the web. Agents gather it, structure it and analyze it to find out what shoppers actually want, so retailers can make these decisions very quickly. We built Zenline to help retailers with exactly these two core decisions. And we sell outcomes: not just an analysis, but a recommendation of what to act on in pricing and merchandising.

Chris Walton: So of all places, you're going into merchandise planning. Fair play to you, because I've seen a lot of companies come and go over the last 30 years trying to do this. Why is now the right time, and why start there?

Arber Sejdiji: You're right, it is hard. But these are the two most important questions a retailer has to answer, and getting them right improves margin by two, three, four percentage points.

For the first time, it's possible. With classic machine learning you could analyze numerical data, but everything had to be pre-processed. Data cleaning and pipelines took weeks or months, sometimes half a year, just to get the data right. Today you can work with much more unstructured data. We realized this in conversations with CEOs and CCOs of the largest retailers in Europe when we started. They all told us the same thing: these are our most important questions, and we still solve them in a cumbersome, manual way. If an AI company could solve them, they would pay for it.

Chris Walton: When I was a buyer back in 2005, I could have done merchandising so much better and faster with the tools generative AI gives us now. Last question: hard problems are hard for a reason. Where does the average retailer see ROI first when deploying AI in assortment planning?

Arber Sejdiji: Within the first hours of talking to a retailer, we show them the gaps in their assortment and what shoppers are actually looking for. We combine the products on their website with what competitors offer internationally. We look at retailers in South Korea and marketplaces in the US and Europe to find where trends come up first and why shoppers ask for them: which shopper need they want to fulfill, and which young brands have just started to serve it. Then the retailer can close that gap and give shoppers what they want very quickly.

Within the first two hours we can discuss the actual gaps in their assortment. Retailers with fast processes have these products on their shelves within weeks.

Chris Walton: So a gap can be two things: a hole in the assortment, online or in store, or not buying enough of something that's about to trend up. Is that right?

Arber Sejdiji: Absolutely. The first one is strategic: what do my shoppers want, and am I serving their needs? The second one is operational, and that can be very complex. Once you decide you want a product, inventory is one question, the number of facings is another, and where it sits on the shelf is a third. These operational questions often take longer than the strategic decision to offer the product at all.

Agents are already very good at the first question, and we work with many customers on it. For the second, we are teaching agents to think about planograms and operational merchandising, and we're making good progress toward answering it in a productized way in the near future.

Chris Walton: That's a good point to end on, because it shows why speed matters in merchandise and assortment planning. You have to get into the products quickly, because the operational decisions are hard to implement down the line. The slower you are, the longer it takes to act and get products to market. Thank you, Arber.

Arber Sejdiji: Thanks for having me, Chris.

Watch the full episode on Omni Talk.

About Chris Walton: Chris Walton is President and CEO of Omni Talk, "the retail blog for retailers, written by retailers", and a Senior Contributor at Forbes. He has nearly 30 years of experience in retail and retail technology, including executive roles at Target and Gap Inc.