We don't call what we do AI.
Here's why that matters.

Everyone is calling everything AI right now. Procurement tools, delivery trackers, invoice processors. If it has an algorithm, it's AI. If it makes a suggestion, it's agentic.

We understand why. The label opens doors. It attracts investment and signals modernity.

But in the supply chain, it also obscures a more important conversation.

The real problem isn't intelligence. It's data.

AI models — whether they route trucks, predict demand, or flag anomalies — are only as good as the data they run on. And in the formal retail supply chain, much of that data doesn't technically “exist.”

It lives and dies on paper. It sits in someone's head. It exists in a supplier's or retailer's ERP and spreadsheets — disjointed. It appears in a delivery note with items crossed out by hand, reconciled manually at a receiving dock, and filed in a folder somewhere.

You cannot build intelligence on top of that. Not reliably, and certainly not at scale.

This is the problem we chose to solve first.

Before we wrote a single line of machine learning code, we spent years inside the supply chain — servicing retail fulfillment deliveries, handling documents, resolving disputes, and sitting in receiving queues. We saw, firsthand, where the data breaks — not in theory, but in practice.

In 2022, we came across a transformer-based document-reading technology called DONUT. We trained a model on top of it to understand the specific documents our industry produces: purchase orders, pick lists, delivery notes, GRNs, returns, and more. The documents that are inconsistent, unstandardised, and full of exceptions.

That work was not glamorous. It took over a year and a half, done quietly on the side, with limited resources, hiccups, and learning. But it produced a starting point for something most supply chain AI platforms don't have: a clean, structured, reliable data layer built from the ground up on real operational documents and workflows.

What we call it.

We call it automation. We call it intelligence.

What we don't call it is AI — not because we're being modest, but because the label has become a substitute for thinking.

When someone says “we use AI for supply chain,” the next question should always be: what data is it running on? Where did that data come from? How was it structured? Who validated it?

Those questions matter more than the models.

What this means for you.

If you are a retailer or supplier evaluating platforms that promise AI-driven anything, we'd encourage you to ask one question before any other:

Where does your data come from?

If the answer involves only ERP exports, or assumptions about what suppliers and retailers are doing, the AI is built on sand.

SengaOS starts with the data. We surface what's hidden, structure what's inconsistent, and connect what's been disconnected.

The intelligence then follows naturally from there.

That's not a limitation of our approach. It's the whole point of it.