The Challenges of Implementing AI: Part 2

Taking a deep dive into two of the six biggest challenges when it comes to implementing artificial intelligence (AI).


Over the last 24 months, the adoption of AI has accelerated dramatically, and it hasn’t been easy. Organizations face a range of challenges supporting next-generation applications, ranging from insufficient performance and increased Capex to new security vulnerabilities. 

Our recent roundtable discussions, with executives from industries including Banking, Manufacturing and Automotive identified six major challenges. In part one of this three part series we explored the challenges around recruiting and retaining the right talent and implementing natural language processing (NLP). This week we’re going to take a deep dive into getting and using data to its fullest and data security regulations in AI. 

These two challenges are crucial because IT leaders recognise the value of AI in the acceleration of business developments, and they explore effective digital transformation strategies that leverage IT in modernising data centres. However, according to research by Accenture, CIOs continue to describe their AI transformation projects as complex and overwhelming and, as a result, 76% of businesses regard infrastructure as an obstacle to advancing the adoption of AI. So, let’s explore the challenges further, and discover some solutions. 

Gathering and Using Data to Its Fullest 

Many businesses are not using their data to its full potential. Why? Often there is poor alignment with business strategy resulting in not knowing how to use it. An executive from Sandvik says “the statement ‘data is the new oil’ has been the worst statement for our industry, because now no one wants to share any data – they think there’s a lot of money in it. But there is no money in the data, the money is in the insights.” They are right. AI algorithms’ knowledge and experience comes from those large, varied, high-quality datasets, but such datasets have traditionally proved hard to come by. Access to anonymous data is hard, and can take weeks or months, and by that point it’s last year’s problem, not tomorrows. 

A senior solution architect at NVIDIA explains “once you have that data, training the model is fairly simple, but the documenting, labelling and preparing of the data is what takes so long.” It’s such a huge task – and if you get it wrong, the long-term effects will be significant. So, is this the main reason that so many industries are hesitant to implement AI more quickly?  

When talking about using this data correctly to optimize deliveries, Scania Group says, “…we’re using Machine Learning (ML) for optimization, and by using these basic building blocks of data streams to find where we can make the most difference” and in fact, since implementing that technology they “have reduced kilometres driven by 10%.” So, we know when implemented correctly, it works well and saves time, energy, and resources. 

Data Security Regulations  

Arguably the biggest challenge discussed at recent Meet the Boss roundtables was around ensuring organisations comply with data security regulations. A Director at Xerox explains “we’re driven by the regulations, and it is a big hurdle, because it changes all the time across all countries.” A Raiffeisen Bank executive shares, “we can’t get the data we need to reach the level of AI we want because of North American legislation, GDPR in the EU and the banking sector being very protective.”  

Regulations should of course protect consumer rights, but they must also still work for businesses. You shouldn’t be afraid to use the building blocks of others. There are so many resources already out there, it would be wrong to re-invent that yourself. A director at NVIDIA advises you should: 

“Take systems that have already been developed and then apply them to your specialist environment. That’s how we make fast progress together. It also allows you to focus on your domain expertise instead of the heavy lift of creating an AI development platform.” 

While it is true that external regulations, internal governance, and organizational structure significantly slow down time to deliver data, unfortunately there is no simple solution to the topic of sharing data to every branch or store. Why? Because data security regulation differs from country to country. But is there a solution that can handle the regulation requirements that businesses can trust?  

Federated learning can be the way forward, as it allows for multiple clients, each with their own data, to collaborate on training together without having to share their actual data. The components employed are designed to allow users to bring their own components in a modular way. 

For more information on all of these topics around AI, join us at our CIO summit in June, and look out for the upcoming part 3 of this series. Next up, we will look at on-premise vs cloud and using AI to supplement business strategy. We’ll see you then!