I have just finished a planning meeting for a transformation that will use AI to automate lots of data entry processes.
Something which is rarely understood or even considered, is programming AI and where the intelligence comes from.
I have just finished a planning meeting for a transformation that will use AI to automate lots of data entry processes. What struck me was the lack of recognition of the time needed to teach the software how to process the transactions; putting the intelligence into artificial intelligence.
There was a presentation that explained how little human interaction would be needed to complete a customer order, and how this would drive efficiency. What was missing was the effort required at the front end of the transformation to understand the current processing logic. Without this understanding, the rules for AI to follow (or improve upon) cannot be known.
I was very frustrated to find that the cost of working with staff to understand and capture this logic for standard and exceptional items had not been factored into the business case.
The business case describes the difference between current and future cost per transaction, so the benefits are very clear. But they are not truly reflective of what will happen on this transformation because the activities and the resource costs of teaching the AI to be intelligent have been ignored.
Having run a number of workshops for these types of transformation, what works well is to treat AI as a colleague! Effectively we must onboard our new colleague with an understanding of what we do, how we do it and why we do it the way we do it.
We must consider standard transactions, and exceptional items.
For standard transactions we need to define:
• Transaction types – driven by different products and services
• An end to end series of activities, from initial request for information to satisfaction of need
• Timings and frequency of these activities
• The data needed to carry them out and the data created by this processing.
In an ideal world, all this information would already by documented, and we could let the AI crawl through our procedures, standards and policies to devise a set of instructions. Too often, documentation does not exist that captures what we do. There might be some for the “standard” transactions but what about those transactions that do not happen very often, because they are triggered by a specific set of circumstances:
• What are those circumstances?
• What is likely to trigger them (so what should the AI be scanning for)?
• What should happen for each of these circumstances and how does this differ from a standard transaction?
As with the arrival of any new colleague, how quickly they become productive, and the accuracy of their work is a result of the quality of their onboarding experience. We also need to plan this when an AI colleague joins the team in order to get the most out of the technology.
I’ve been releasing my views and experiences with AI weekly – take a look at previous thoughts and opinions here.
