The Watson Weekly is sponsored by Avalara — the agentic AI platform automating global tax and compliance for leading eCommerce brands with simple plans, transparent pricing, and complete coverage across Shopify, Stripe, WooCommerce, BigCommerce, and more.

KIBO sells order management to retailers who are being told by everyone else in this market that agents will soon run it. Ram Venkataraman came on the show and spent his first three minutes arguing the technology is not ready.

I opened with accountability. An agent takes a customer's order, routes it to the wrong warehouse, and approves a return it should have declined. Who owns that outcome?

Ram said the shopper will blame the retailer, and that the burden sits with vendors to build systems that earn the retailer's trust. Then he got more specific than most people selling into this category are willing to be.

KIBO's routing runs on machine learning models, not LLMs. His reason is that LLMs remain too probabilistic to depend on for a decision that moves physical inventory. He credits the improvement in tool calling and prompt following. He does not think it has gone far enough. Where KIBO does use LLM technology is in two places. Configuration, so a merchandiser can write a routing rule in plain English instead of filing a ticket. And explainability, so someone can go back afterward and find out why an order got an 80% discount or split into five shipments. Any write operation in the platform still needs a human to approve it before it takes effect.

Notice that this costs KIBO nothing to say. The deterministic engine is the product they already sell. A guardrail argument defends existing revenue. That does not make the argument wrong, and I think he is right about the current state of the tech. It does mean you should ask any vendor making the same case what their answer would be if their own engine were the thing on the chopping block.

The second claim is the one I want proof on. Ram calls order management the margin layer of a commerce business, because that is where the fulfillment location, the cost, and the delivery promise get decided. He also calls it a conversion rate optimizer, since inventory accuracy and estimated delivery dates change whether the order happens at all.

The conversion argument I buy. Merchants already measure it. The margin argument is harder to verify from outside, because nobody publishes cost per order from before and after an implementation. What you get instead is a business case built during the sales cycle by the people selling you the software. If any operator on the other side of a two year deployment has audited numbers, send them to me.

The third thing is the most immediately useful, and it costs nothing to run. Ram's tell for a B2B distributor who does not know they have an order management problem is this. How often does your website say out of stock when the product is sitting in a warehouse? How often do you cancel an order you already accepted? Both mean the ERP and the front end are not talking in real time.

Underneath that sits complexity most retail people never deal with. Account hierarchies. Negotiated quotes. Project based pricing, where the same buyer pays two different numbers for two different job sites. Supply that has not landed yet and has to be allocated anyway. He described one customer running three separate versions of SAP ECC.

What did not get resolved is the part I keep coming back to. Saying the burden falls on vendors to earn trust is a position, and it is not a liability term. When an autonomous system ships the wrong thing to the wrong place, the contract decides who pays. I have not seen many of those contracts get written yet.