AI Underground: introducing AI-native Customer Operations

The room was full of the people this actually happens to: distributors, wholesalers and manufacturers, the ones whose order desks absorb whatever arrives. AI Underground is deliberately small and off the record, which is exactly why we used it to put a term on the table for the first time: AI-native Customer Operations.

What the term means

Most automation in this field is retrofitted. An order desk exists, has existed for twenty years, and AI gets bolted onto the side of it to take some of the typing away. AI-native Customer Operations starts from the other end: if you were designing the whole customer operation today, knowing what these models can do, you would not build a desk and then automate it. You would build something that reads, validates, decides and escalates by default, and put people where judgement is actually needed.

Why we can say it now

A term is only worth introducing if something stands behind it. The EDI order agent has matured — no longer the experiment it was eighteen months ago, but production software at companies who have stopped thinking about it. That is precisely when a component stops being interesting on its own and the question becomes what it is part of.

So the scope follows the term. Reading an incoming order is one step in a chain that is still largely manual after it. We set out where we are heading: the full order-to-cash flow rather than its first step, and from there into quote-to-cash and procure-to-pay — quotes, confirmations, invoicing, purchase orders, supplier documents. The same problem wearing different clothes in each of them.

Thanks to everyone who came, argued and stayed for the last drink. AI Underground runs on the last Tuesday of the month, in a small group, with no slides to hide behind.