Accenture logo


Ray Eitel-Porter

Ray Eitel-Porter

Global Responsible AI Lead at Accenture

How can you identify whether a problem or opportunity – whether it’s from industry, third sector, government or academia – is a suitable candidate for an AI-driven approach? Data Study Groups are a method devised and established by The Alan Turing Institute that aims to answer that question. 
I've seen first-hand just how effective a Data Study Group can be at taking a problem and turning it into an academic challenge, which then delivers real benefits for the organisation concerned and potentially to wider society.

So what are Data Study Groups? One way of thinking about them is as a week-long, very well-prepared hackathon. During the preparation, the ‘client’ organisation and the Turing team work together on selecting and refining the business problem so that it's formulated in a way that the Turing team can tackle during the Data Study Group. 

The Data Study Group team will take a problem and work on it using a range of different data-science skills. At the end, a report illustrates the different techniques which have been tested and what the results were. 

But before a problem or opportunity can be selected for the Data Study Group approach, it’s important to understand the ingredients of a good challenge. This requires three things: a suitable dataset; an interesting, but sufficiently defined, problem formulation; and then its potential for learning and impact.  

The right data

It’s not uncommon for a prospective challenge to arrive with questions, but limited data. But data quality and a good understanding of how the data has been collected is vital. There is no point, for example, in having a machine-learning model that requires data which cannot be made available to the Data Study Group participants. Contrary to popular thinking, there’s no AI ‘magic’ that can be applied here. The longstanding observation of: garbage in, garbage out remains true.

What exactly is the question?

Working out what the question is and expressing it with as much clarity as possible is harder than it sounds. A particularly good example was a recent challenge that a Data Study Group undertook, looking to improve the prediction of how quickly and under what circumstances stroke patients would recover their speech. This was able to make use of brain scans, evidence of cognition capabilities, as well as patient data and the severity of strokes that patients had suffered. The resulting insights were used to build a model to help clinicians prescribe treatment plans.

Start simple, go open-ended

It’s also important to start with a question for which there is a clearly defined answer, be that:  a simple yes or no, a number or a category classification. With that established, it’s then possible to progress to more open-ended enquiries. 

In the case of an energy organisation, the question was seeking to predict the levels of energy demand that would be seen in one, three, or seven days’ time? By addressing this, and expanding the question into more open-ended areas of enquiry, it was possible to improve the forecasting of renewable energy production by over 50 percent.  

To gain attention, it’s important of course to have a challenge that will spark the interest of a Data Study Group. If, for example, resolving a problem will have wider positive impacts for society, that can make it more attractive. For instance, Accenture sponsored a programme that used a Data Study Group to analyse hate speech against female politicians on social media. The Turing has also worked with others to understand how a product’s formulation could be adjusted to influence the size of its carbon footprint. 

Above all, formulating the question is an iterative process. It starts with a rough diamond that can be gradually refined and polished to become a sparkling gem.

What will we learn?

The third element is value, in terms of outcomes and/or the learning that will arise from the Data Study Group. This cuts two ways. The first is for the organisation to acquire insights that will enable it to operate more effectively. But it’s also about being able to work with leading figures in the industry or sector and data science, fostering the collaboration that is essential to continue making progress in how AI can help solve some of the toughest challenges out there.