Artificial Intelligence (AI) is potentially the most powerful technology businesses have ever had access to. AI is not only transforming how businesses operate, but also the services and products they offer to their customers.
In our experience, having worked with countless organisations on their AI and digital transformation strategies, there are barriers to successful adoption that we see time and time again. Here are the essential elements businesses must have in order to overcome the obstacles standing in the way of leveraging AI to its full potential.
Access to the right data
Make no mistake – good data fuels good AI. Trained on the right data, machine-learning algorithms can do amazing things, such as see (machine vision), read (natural language processing), speak (natural language generation), walk (autonomous robots), act creatively (generative design), and much more. However, while businesses have access to more data than ever before – both internal and external – that data often isn’t ready to feed AI applications.
There are often issues around data integrity and complexity, siloes, metadata, information architecture and so on. To address these challenges, businesses need to approach them strategically and foster a culture where data assets are valued, defined, mapped, classified and well-governed.
So, to lean into AI, good data is a must, just as a map application needs good data to plan the best route. Any path on the map needs to consider all possible routes, travel preferences, waypoints, real-time traffic, roadwork data etc. If any of these factors are missing, out of date or incorrect, the outcome is bound to be less than perfect. Similarly, if the data you put into your AI algorithm is incomplete, inaccurate, inappropriate, or out of date, then AI will never achieve its potential.
So, for AI-specific data sets, it’s vital that organisations develop a complete understanding of the data involved. In short, in order to assess which data is appropriate and optimum, companies need good comprehensive metadata libraries. Metadata helps describe and define data and should include both technical and business detail. Good metadata catalogues describe data characteristics such as type, ownership, sensitivity, relationships, lineage, business definitions, quality and more.
Access to the right talent and skills
AI skills are in huge demand and companies are competing for top talent. The solution here is to develop a skills and data-literacy strategy that acts as a pathway to acquiring and building the relevant people and teams. The key to getting this step right is upskilling existing employees, boosting data and tech literacy across the board and finding the right partners.
Sometimes, businesses use external partners such as IBM with their AI developments to have the right mix of expertise. To do this and promote success, be prepared to start a collaborative journey with the business delivery partner, so you’re working side by side to co-create and co-develop AI. Both sets of staff need to work as a team as AI is a voyage of discovery – you may find insights you weren’t expecting, or realise you need to pivot your approach or method. As a team, you should approach this in an agile manner and work co-operatively so it’s easy to change course depending on your findings.
To lay the foundations of this approach, your team will need a handle on business insights, good analytical skills, especially if they aren’t data scientists, and a vision for the outcome. It’s also worth checking that your business partner is happy to move forward with paired development and co-creation. To fast-track AI and data-science skills, building hybrid teams that bring in external talent is fine initially, however, your priority should be rapidly building key in-house expertise to collaborate with delivery partners, manage outcomes and scale across the organisation.
It has never been easier to upskill your own workforce via (often free) online training courses for AI and machine learning – so-called massive open online courses or “MOOCs”. However, companies need more than data scientists or machine learning engineers. Data science is a team sport and therefore requires a supporting team or a data squad that usually includes data curators, data stewards, data owners, data cleansing and transformation specialists and more. The data squad’s job is to manage, shape, govern, cleanse and classify the data, to ultimately create a metadata map of the organisation’s data assets. The talent for the squad probably already exists in part, but the team’s remit needs to be across the organisation and not limited to looking after pockets of data.
The right technology
The uses of AI are diverse – from “ask-me-anything” chatbots to machine learning that predicts propensity of a shopper to buy. This means that AI can and should be delivered in different ways, using the right data, processes and technology. For instance, many AI algorithms have to ingest real-time data from event streams and processes, which requires a modern technology stack. For some projects, you can get an off-the-shelf “as a service” algorithm, while other projects require teams to craft unique machine learning approaches.
Getting the technology right can be a challenge, particularly for established businesses that have invested heavily in platforms a decade ago and are now struggling to bring their legacy system into the AI age. AI often requires an update to the technology used to collect, store and process data. Digitally native companies often have a huge advantage here as they can build nimble and scalable businesses on cloud with modern apps, APIs to collect and stream real-time data and modern ML/AI like IBM Watson.
Ethical and governance frameworks
Like any technology, AI can do amazing things, but, if misused, can have negative outcomes. Biased and intrusive AIs are one such example. To combat this, businesses need to remove biases from their systems and ensure data privacy and security are at the forefront of everything in their approach. Organisations should have clear ethical and governance frameworks in place to ensure that AI systems are in compliance with legal requirements. If you need guidance on what these frameworks entail, consult best practice guidelines, such as the Partnership on AI, Ethics guidelines for trustworthy AI by the European Commission, or the Principles on AI published by the OECD.
It’s also essential to ensure that your AI is unbiased and isn’t making decisions no-one can understand. But, creating unbiased and fair AI is a definite challenge. Biases in AI are often a result of the training data set, which may be skewed or favour certain attributes over others. It’s therefore important that any data selected to train and evolve your AI is continually checked and validated. It’s equally important to validate AI results to ensure algorithms don’t unintentionally favour one set of attributes over another. Thankfully, it’s possible to use technology to help automatically identify bias and understand how an algorithm has utilised its training data. All of this is essential to creating fair AI.
The right organisational culture
The culture of an organisation – or lack thereof – is one of the most underestimated barriers to successful AI adoption. Progress has to start from the top, with good leadership and open conversation to dispel fears and misunderstandings about the technology. Companies that are serious about AI need an executive sponsor – someone like a Chief Intelligence Officer or a Chief Data Officer to oversee AI strategy.
The next step is bringing people along on the journey. Including AI champions, who can span the world of business and technology and help engage with business users and techies alike, can be extremely useful. Having a better understanding of data science and AI across the business will also help promote data governance and ensure data is treated as the vital business asset it is. If everyone in the business understands the need for good-quality data, then that can be a big step towards better data, and therefore better AI.
A strategic approach
Finally, avoid AI for AI’s sake and ensure you approach AI strategically. It is absolutely vital that your AI initiatives align with your business imperatives and your technology strategy. Too many companies experiment with AI instead of first questioning how AI can help them tackle business challenges and uncover opportunities. In an ideal world, your business strategy should guide your data strategy, which in turn, should align with your AI strategy. Understand the purpose of AI for your business, and then develop the approach based on the goal.
These elements are the foundations for businesses attempting to adopt AI successfully. No matter what stage of the journey to AI your business is on, it’s worth frequently checking in and re-evaluating the barriers that are limiting the success of your AI and pivot where necessary. For, when successful, AI has the potential to offer enormous return on investment for businesses today.
To find out more about overcoming barriers to adopting AI, join us at our Data and AI virtual Forums. Register here.