The third requirement to true success in leveraging AI: DATA

As per my previous article , there are three very simple principles business leaders can use to harness AI & Machine Learning (ML). We’ve already covered the first two; about off-the shelf AI tool use and letting strategy steer custom AI models for real benefit.

The final one, data and its necessary architecture only requires three things:

·       Appropriate and resilient data backbone – engineering, architecture and compliance for avoiding costly, cumbersome or fragile data ecosystems

·       Constant focus on data quality – a much more taxing and continuous effort than initially obvious, but absolutely key to stay ahead of the game

·       Attention and dialogue between “data team” and business leaders – true engagement and desire to understand and collaborate

The data backbone needs to be appropriate and resilient; the right data strategy, expertise and both internal and external data should be in place to optimise the working of AI tools. This applies to data used in off-the shelf AI applications as well as training and actual data feeds for proprietary ML models. A good starting point for confidence and familiarity is to read up on Data Engineering basics. I include compliance, ethics and regulatory aspects under this heading.  

A senior leader needs to consider that “data quality” isn’t a once and for all fix: it’s a constant challenge but also opportunity to enable better AI-based business. For example, it’s easy to understand that you need to train an ML model on data which is relevant for its later application. But have we thought about how often we need to refresh the training, check that it’s still compliant with regulation, ethical policies and above all still relevant from a business perspective? If we use a training set based on current customers, but manage to bring in a whole new customer segment or market next year, we need to retrain the ML model. Does this sound familiar? As a senior leader, a useful approach can be to think about AI as statistics on steroids – the pitfalls and insights possible are similar to those you may have learned in Stats 101 at uni.

The final requirement is a way of working which links business goals and evolution to data architecture and quality. The two factors above (data architecture/engineering and data quality) aren’t much use on their own, buried in the IT department without much oxygen or dialogue with senior business leadership. Ideas and risk management works best when data experts and business owners can communicate and regularly solve problems together. We need a governance forum or organisational set-up which fosters respect, desire to collaborate and creativity. The forum should include people with different backgrounds and experiences to ensure data topics are well understood by executives and business requirements are prioritised appropriately in data teams’ work.  

This article is about how to ensure both off-the shelf and custom AI tools use data to actually deliver quality output, driving great business. In previous articles I also covered how to use off-the shelf AI and what is required for successful in-house AI development; Harnessing available tools, as well as how our AI/ML efforts should be driven by business strategy.

As a senior leader, these basics will be familiar, and above all they can be addressed with well-known approaches; strategy refreshes, investment proposal assessments and change management. Good luck and enjoy the journey; I know I am!