It depends on the type of AI being applied and the area of the business it is being applied to.

Where AI is searching for commonly understood definitions - invoice number, invoice date, purchase order number - it can already be trained on that understanding and hit the ground running. Innovations like Invoice Capture AI are a good example: because the fields are relatively well structured and consistent, AI can deliver value quickly, without waiting for a perfect data foundation.

AI used in reporting can also help improve data sets over time. Where AI needs to be trained on information that is very specific to your business - coding structures, chart of accounts, internal categorisation - then that information does need to be accurate for the AI to make reliable predictions.

The practical approach is to understand the quality of data you have, identify where gaps or inaccuracies exist, and prioritise improvements according to the business problems you are trying to solve. Waiting for perfect data could mean waiting indefinitely. The important point is understanding your data’s limitations, not assuming AI will resolve them for you.

Not necessarily. The opportunity exists to apply AI in certain use cases before full automation is in place - invoice capture is a good example of this.

 However, as AI is applied to more complex tasks with greater layers of complexity, the dependency on broader automation increases. More complex tasks require a greater breadth of data to resolve, and wider automation is what creates that data foundation.

 The practical conclusion is that automation creates a stronger foundation for AI adoption at scale - even if you can begin to see value from AI in specific, well-defined areas before that foundation is fully in place.

That depends on the process and business objective.

 Relevant measures can include automation levels, first-time match rates, purchase order compliance, invoices per FTE and other operational KPIs. Betfred, for example, achieved a 47% improvement in AP efficiency following adoption of the Proactis Invoice Capture AI Accelerator.

 The important thing is establishing a baseline and then measuring progress against it consistently over time.

There is not one universal answer. Different types of AI are suited to different scenarios, and the business problem should determine the appropriate technology - not the other way around.

The more useful third question is not which type of AI to use, but what strategy you are deploying. That means understanding the need to automate, gaining quality adoption, and then enabling the tracking and iterative improvement against best practice performance that allows AI to deliver sustained value over time.