Artificial Intelligence is changing business forever. But is the data powering it good enough?

The last few years have been auspicious for artificial intelligence. Once the preserve of Silicon Valley’s tech giants, AI has emerged as a tool that businesses small and large have become keen to harness for disparate use-cases in domains ranging from cybersecurity to customer service and analytics. Data scientists, tasked with creating machine learning-driven tools, are now ubiquitous in most industries.

As AI’s presence and influence grows across the business world, one key potential pitfall should not be overlooked: data. AI systems are generally powered by machine learning, a technique that creates smart systems through the recursive training of algorithms on vast troves of data – in the case at hand, we are talking about data that companies create and gather throughout their daily businesses.

Yet this data is not always usable, explains Fernando Lucini, managing director for Artificial Intelligence at Accenture UKI. Different kinds of data – client, financial, transcripts of customer service calls – can be scattered throughout a business, making it challenging to marshall in a concerted way. “Data is everywhere, so the question is how you bring it together,” Lucini says.

Another major issue has to do with the data’s so-called “dirtiness”, such as missing information, inconsistencies and errors. “If you have to use machine learning on a dataset where some of the entries are empty or inconsistent, that is problematic,” Lucini explains. “You can’t count on this data to be what it is supposed to be.” (Of course, data perfection is almost impossible to achieve, and a pointless objective; but data can be cleaned up to be good enough for a given purpose.) The possible results of relying on dirty data for fashioning AI tools can be summarised with the famous computer science saying: “garbage in, garbage out.”

“Businesses sometimes seem to feel that the answer is definitely there, somewhere in their data,” Lucini adds. “It often is, but it’s not right in front of them – they will need data scientists to fix the data. But if the data is incredibly bad, maybe they won’t be able to sort it out.”

How to solve this problem? Lucini says that the reason companies don’t tend to structure, clean and de-silo their data is that they don’t see a compelling reason to do so. Data hygiene takes a lot of time, energy and money – why would a company engage in such an effort, unless there is a clear reason?

Research suggests that 79 per cent of businesses base critical decisions on data without properly investing in its verification – which is risky. The way to “win hearts and minds of executives” Lucini explains, is to “clearly identify and explain the tangible value of doing it”. For example, inaccurate data caused United Airlines $1bn in missed revenue. Why? Its pricing models were built on obsolete data.

Finding that value comes from setting about this effort with a precise use-case in mind, rather than proceeding with a blunderbuss-like approach.

“When you understand the problem with the data, and know what you want to do with it, that’s when things go well,” Lucini says. “Things go very badly when your effort is open-ended.”

“Poor results happen when nobody builds a use case: nobody tries to understand what it will cost to get the data, why it’s needed, and whether it will be worth it in the long run. You have to be very clear about how the data will be used and whether it can be reusable for various business cases afterwards.”

In Lucini’s opinion, one of the key trends we are likely to encounter in 2019 is the progressive “industrialisation of AI”. Until now, he says, many companies interested in developing an AI capability have been acting in an unsystematic and unfocused fashion.

“A couple of years ago, these companies got a bunch of data scientists, who started learning about AI, and experimenting with it,” Lucini says. “Now, one or two years have been spent doing this AI thing, and the companies haven’t seen any return. This year the situation needs to change – it can’t go on as an experiment.”

According to Lucini, the forthcoming industrialisation will need to be comprised of several chapters: a value framework that helps a company discover how AI can deliver value while avoiding “use-case paralysis” (“you can do everything [with AI], so you don’t do anything,” Lucini says); a plan to involve all relevant parts of the company in the process; an approach that helps the right people – data scientists – get the data they need, and thus produce outcomes; and a reliable data structure.

“Keep in mind: the use cases will need to drive all the decisions,” Lucini says. “But as much as we can dream up different ways that AI can change the world or make a business hugely successful, they need to look at their data first. Without that fixed, there’s no AI.”

Source: WIRED

Leave a Reply