AI Is the Easy Part. Your Data Is the Work.

Last week Experian released a survey of 102 senior Australian lending decision-makers. 72% are already running agentic AI to help make lending decisions. Only 3% say their data is fully ready for it.

67% of respondents say their data is not ready or only partly ready. 45% blame fragmented systems, 42% point to poor data quality, and 31% do not trust the outputs. Only 2% are comfortable with fully autonomous decisioning. And 92% say they would pilot or adopt a vendor that solved the data problem for them.

Without data that is current, correct and understood, the AI system that consumes it will produce results that are wrong or unexpected, create risk to your customers and your business, and potentially undermine your AI investment at speed.

The big banks have already solved, or are a long way down the path to solving, this problem, and so have the AI-native lenders and aggregators now resetting broker turnaround times. Now the next part of the value chain needs to consider the implications. This is not just relevant to brokers. It applies to every business.

OpenAI studied how enterprises put AI to work and found the firms getting the most out of it connect AI to their own context, tools and repeatable workflows. The top users produce 8.3 times the output per active user of a typical firm. The difference is that they have invested in the plumbing, not just the AI software licences.

Getting AI to produce work you can rely on is the hard part, and that hard part is mostly about the quality of what you feed it and how you check the result (source).

This is not just a broking problem.

In real estate the portals now generate buyer intent signals for agencies. Domain is rolling out 3D tours nationwide to identify high-intent buyers before inspection, and REA's research shows buyers have restructured how they search around conversational AI. The unautomated step is what happens to that signal inside the agency: routing, prioritisation and speed to contact.

The 3rd National Policy Roadmap for AI in Healthcare names back-office administrative automation as the clearest near-term win, with a human in the loop as the non-negotiable. That is the safe first step precisely because it cleans and moves data without touching clinical judgement.

Six steps to get your data ready

Scope: you do not need a data governance project. Focus on one workflow and get it right before moving to the next, and keep it in a setup where you can still prove where the data went.

1. Pick a single workflow that matters, one with a clear start and finish. Understand and document what data is used.
2. Decide where the truth lives. This could be a system of record, or which version of a spreadsheet is the real one.
3. Clean only the fields that feed the decision. You are not tidying everything. You are making sure the handful of things the AI reads are correct and current.
4. Connect the two systems that need to talk to each other, so the file moves without a person copying and pasting between screens.
5. Keep a person on the checkpoint for anything hard to undo. The agent does the repetitive work, you sign off on what carries weight.
6. Make sure someone owns the data and is responsible for keeping it that way.

The newest problem in AI turns out to be the oldest discipline in business: knowing where your information is and keeping it in good order. Sort that, and the AI you are already paying for starts to earn its place.

That fifth step is the same judgement call behind deciding what an AI agent should be allowed to do on its own, applied to your data instead of your workflows.

If you want a hand working out whether your data is ready for the AI you are already paying for, email me.