Decisions in SQL with ai_decide()

Every operations team has small judgement calls someone makes by hand, hundreds of times a week, and eventually the data team is asked to automate them. Databricks now has a SQL function built for exactly that, and here's how it works and where it's safe today.

Picture a distributor's order desk. Two hundred orders arrive overnight from Business Central, and someone in customer service reads every note before ten. Does the warehouse need to see this one? Is the customer asking for net sixty again? Is a line down waiting on it?

Keyword rules miss most of it, because customers don't write in keywords. A chat model reads the note fine, then answers in a paragraph, when you need a team, a yes or no, and a priority to sort by.

Instead of a paragraph, you get answers you can sort. Ask whether a note wants special terms, and it tells you how likely that is. Give it your three teams, and it picks one. Ask how urgent the order is, and it puts it on your own scale.

It's one query. You write the three questions once, run it over last night's orders, and every order comes back with a team, a yes or no, and a priority, each with how sure it is.

Then you decide what happens. Say anything over eighty percent sure goes to credit, anything near the top of the urgency scale gets expedited, and the rest goes out. Your team can read and change that rule any time.

It's in Beta, so we'd run the first month as a pilot on your orders. We write the questions with the person who reads those notes today, compare their calls with its answers, and only then let it route anything.

We set these up for teams on Databricks. The questions get written with the people who make the calls today, and the routing into Business Central comes once the pilot proves it. If calls like these are piling up in your backlog, send us an example, and one of our senior engineers will show you that pilot on your own data. We're a Databricks partner, with 80 Databricks-certified engineers.