AI-Ready Data and Integration: the series
Every AI system we have put into production has started the same way, and the first few weeks had nothing to do with models. They went on connecting systems that were never designed to talk to each other, working out which record referred to which customer, and getting agreement on what the organisation would treat as true when two systems disagreed.
That work rarely appears in the case study. The demo shows a chatbot answering a question correctly, or a model predicting churn. What sits underneath is a set of pipelines someone has to run, monitor and pay for on the Tuesday after go-live.
This series is about that layer.
Why we are writing it
The AI pilots we see fail mostly fail on data. A pilot is built on a clean extract someone pulled by hand, it performs well, and then it meets live data: duplicate customer records, a field three teams fill in differently, a nightly export that silently stops on long weekends, an ERP with no usable API. The model was never the constraint.
We have been integrating enterprise systems since 2001, well before anyone was asking us for AI. The problems are the same ones, with less tolerance for being wrong. A reporting pipeline that drops 2% of rows produces a slightly odd chart. An AI system fed the same data gives a confident answer to a customer.
So the series works through the practical side: how to connect systems that resist connection, how to resolve entities across them, how to measure data quality in a way that means something, and how to govern access so an AI system can use the data without exposing what it should not. We will also cover Centazio, our open-source, code-first integration platform for .NET, which came out of writing the same plumbing repeatedly for clients and deciding to stop.
Who it is for
IT managers and executives in Australian organisations who have to approve this spend, defend the timeline, and carry the risk if it goes wrong. You will not need to write code to follow along, though there will be code where showing it is the clearest explanation. The aim is that you can ask sharper questions of your internal team and of vendors, and recognise when an estimate has skipped the hard part.
On approach: we write about systems we have actually run. That means real costs, including the ongoing ones, and real limitations. Where something is not worth building, we will say so, and where we have got something wrong in the past we will say that too. There are no vendor pitches here and nothing depends on you buying anything, including Centazio, which is free and always will be.
PicNet builds production AI systems for Australian organisations. Talk to us about what a first project could look like.