Why Does AI Need Business Context to Improve an Ecommerce Site?

AI needs business context because its output is only as good as what it knows about your ecommerce store. Mory Katan, Chief Technology Officer at LiLi Tile, calls it context equals output: an AI system that reads session data, heat maps, customer conversations, inventory and financials makes recommendations that land, while one starting from a blank prompt guesses.

That is the core idea of episode 150 of The Ecommerce Toolbox, where Noibu President and Co-Founder Kailin Noivo talks with Mory Katan about building a self-improving ecommerce site. This recap covers why AI needs business context, what context to give it, and what LiLi Tile saw when that context was in place. Listen to the full episode.

Who is Mory Katan, and what did LiLi Tile build?

Mory Katan is the Chief Technology Officer of LiLi Tile, a brand that sells handmade tiles online. He started LiLi with his sister around 2018 as a branch of a family construction materials business, after five years leading a development team at JPMorgan.

LiLi Tile's team built an AI system called Owl that connects the store's data and acts on it. Owl reads LiLi's Shopify site data through Noibu, along with the CRM, inventory, financials and warehouse workflows, and uses all of it to recommend, test and launch site changes.

Why does AI need business context to improve an ecommerce site?

Context equals output means an AI system's recommendations can only be as accurate as the information it has about the business. Mory Katan opens the episode with the rule:

“context equals output. If you don't have a robust context set, you don't have a really good output.”
— Mory Katan, Chief Technology Officer, LiLi Tile

Most AI tools start every conversation as a blank canvas. A merchant asks a question, pastes in some data, and gets an answer that is sometimes right and sometimes not, depending on what was pasted. The model is not the weak link in that setup. The missing context is.

What to do: before judging an AI tool on its answers, check what it can see. If it cannot read your shopper behavior, products and operations, its answers will stay generic.

What context does AI need before it can improve a store?

At LiLi Tile, the first step is to let the system learn. Owl reads the tools the business already uses, the past conversations with customers and the team's working cadence, so it understands the nature of the business before it makes recommendations. It also logs corrections, such as a formatting preference, so it does not repeat the same mistake.

For an ecommerce store, the context that matters most falls into four groups:

• Shopper behavior: sessions, heat maps, page speed from real traffic and add-to-cart data. This is where the site's problems show up, and shoppers rarely report them: over 90% of website errors are never reported.
• Customers: CRM records and past conversations.
• Operations: inventory, lead times and purchasing. LiLi's system weighs page visits and add-to-cart counts when it suggests what to reorder.
• Money: financials, so recommendations account for margin and not just clicks.

Why do handoffs slow down AI-driven site changes?

Handoffs slow AI-driven site changes because every step that moves between a tool and a person adds waiting time. Mory Katan describes the usual pattern: the AI returns a packet of code, and someone has to take it, upload it and test it. His team set out to remove those steps:

“you want to go from evaluation to decision-making, to execution, to deployment, all in one study, in one flow. You shouldn't be going back and forth and back and forth.”
— Mory Katan, Chief Technology Officer, LiLi Tile

The same applies to a store's own team. On the episode, Mory Katan says that historically a site redesign took a developer, a designer, a project manager and the owners at least two months, with many decisions made on gut feel. For another brand, a project manager and the owners took a redesign from design to live on Shopify in two days.

What to do: map how a site change moves today, from idea to live. Count the handoffs. Those are the steps where context gets lost and time disappears.

What happened when LiLi Tile connected its data to its Shopify site?

LiLi Tile onboarded Noibu on a Thursday afternoon. By Saturday, Owl had made its first set of decisions, and the team launched the updated site on Sunday night. Mory Katan shared the results on the episode:

“Our site speed went from 3 seconds to 1 second. Our add-to-carts over 30 days went up 500%, and our general conversion went up almost 200%.”
— Mory Katan, Chief Technology Officer, LiLi Tile

These results are reported by LiLi Tile. The LiLi Tile customer story covers the rebuild in more detail. The point Mory Katan returns to throughout the episode is that the quality of the context drives the quality of the output.

How can a lean team query data spread across platforms?

A lean team can query scattered data by connecting the platforms to one system that can cross-reference them. Mory Katan describes customers with CRMs holding hundreds of thousands of records and no quick way to ask simple questions, such as which customers did not receive an email in the last 30 days.

At LiLi Tile, Owl combines signals that live on different platforms. For custom tile, the team cross-references location, whether an area has a lot of pools, and how many times someone visits the site. Large retailers have analysts for this kind of work. A team of a handful of people usually does not.

Where should an ecommerce team start?

Mory Katan's advice to listeners is to begin with the work that takes the most time:

“Let's identify some workflows that are taking you a long time and can be repeated by AI. And if you're spending more than an hour a day doing admin work, you should probably evaluate that, right?”
— Mory Katan, Chief Technology Officer, LiLi Tile

Five steps to give AI the context it needs:

1. List the slow, repeated workflows that take more than an hour a day.
2. Connect shopper behavior data first: sessions, heat maps, real-user speed and add-to-cart data.
3. Add the business data behind each workflow: CRM, inventory and financials.
4. Cut the handoffs so a recommendation can move to a test without being rebuilt by hand.
5. Measure every change and feed the result back in as context for the next one.

How do Noibu's AI agents use store context?

Not every team will build its own system the way LiLi Tile did. Noibu's supervised AI agents start from the same kind of context: real shopper sessions, heat maps, errors and page speed on your own store. The Performance agent measures Core Web Vitals from real shopper traffic, ranks the slowest page groups, identifies causes and stages changes. The A/B Testing agent builds and runs experiments, measures lift and reports the winning variant. The CRO agent finds underperforming product pages and rewrites them.

Every Noibu agent runs the same loop: find, build, your approval, ship, measure. Nothing ships without approval. The agents work in the Noibu console and in the AI tools teams already use, such as Claude, ChatGPT and Microsoft Copilot.

For the bigger picture, see what AI for ecommerce operations means, why recursive, self-improving websites are already here, and whether to wire Claude to your store data yourself.

Noibu is AI for ecommerce operations, the always-on optimization system that keeps online stores running at their best. A team of supervised AI agents finds what's broken, slow, or leaking revenue, does the work, and hands results back for human approval. Trusted by retailers like Free People, Guess, Liquid I.V., and Ariat.

Listen to the full episode

Hear Kailin Noivo and Mory Katan in episode 150 of The Ecommerce Toolbox on the episode page, Spotify, Apple Podcasts or YouTube.

Frequently asked questions

Related reading

• Shopify has native A/B testing. What does an AI agent add?
• What is AI for ecommerce operations?
• Build vs buy: should you wire Claude to your store data?

Ecommerce Toolbox listeners: get your podcast-exclusive website audit →