Jeff Bezos has hailed this era as the “golden age of AI”. However, a quarter of companies are still reporting a 50 per cent failure rate for AI projects, pointing to a lack of AI skills and unrealistic expectations as the two main roadblocks. Other issues raised were high costs, a lack of data readiness and the risk of bias. In such a rapidly growing market, businesses on the road to AI need a clear plan to overcome these challenges to improve their chances of success.
AI on the rise
Whether it’s through the growing level of investment in AI – which is set to skyrocket to $98bn by 2023 – or the fact AI projects are set to double over the next year, organisations are improving performance, efficiency and analytics capabilities with AI to solve real world problems faster. What’s more, McKinsey found that in two-thirds of use cases, AI – specifically deep learning – can improve performance beyond that provided by other techniques.
All too often, when AI projects fail, it is because organisations are rushing to deploy them. It’s only after putting in the time and money that they realise they don’t have the right skills and resources, or discover their data isn’t properly collected, labelled or categorised. Instead, before undertaking AI projects, organisations should follow these four steps to reap the benefits AI has to offer.
Step 1: Identify a clear use case
One of the most common slip-ups when adopting AI is rushing into deployment before considering exactly where it can be applied in the business. This approach can be costly and waste serious time and money on ill-planned projects.
Part of the problem is a lack of understanding at a company level: EY found that 48 per cent of business leaders think a lack of managerial understanding and sponsorship plays a role in slow AI adoption. This means it’s vital for organisations to secure buy-in from both IT and business leaders. Additionally, businesses should identify what areas could benefit from the efficiency, analytics and performance benefits that AI has to offer, setting clear goals for the project from the offset.
Step 2: Make sure company data is AI ready
In the 1960s, programmer Wilf Hey famously coined the term “garbage in, garbage out”: 50 years on, it’s never been more applicable in the data science world. If input data is poor, then even a perfect algorithm can produce complete nonsense. Without high-quality data, the risk of failure is much greater, and organisations could end up basing decisions on inaccurate, biased or unintelligible insights.
Once a clear use case has been set out, precise data foundations need to be established. Organisations must review the relevant data sets for each project to ensure that everything from tabular data to image, audio and unstructured text can be stored, categorised and labelled effectively before it’s analysed.
Additionally, it’s vital for IT to ensure the required data can be continually collected and fed into AI models, so that data sets remain relevant, trends can be spotted and models continue to improve their outputs as time goes on. If the business doesn’t have these capabilities and can’t store and categorise data effectively, this needs to be remedied before making a start on any AI projects.
Step 3: Offset the talent gap by empowering your workforce
Just last year, LinkedIn reported a shortage of 151,717 data scientists in the U.S alone, as demand continues to rise and skills gaps widen. Not only is this a problem for businesses, but it’s slowing the overall progress of AI development: EY recently reported that 80 per cent of business leaders agree.
While there has been an increase in the number of programmes like MOOCs (Massive Open Online Courses) to train users in digital skills, we need to lower the bar for entry to ensure the most powerful technologies – like AI in general and deep learning in particular – are put in the hands of a wider group of users.
To get more people using AI, organisations need to find ways to reduce the learning curve. This can be done through partnerships with AI experts, or by utilising a platform to remove the underlying complexity for even the most challenging types of AI like deep learning. By automating workflows and offering a visual development environment, organisations can help reduce the skills gap by enabling more junior data scientists, developers and domain experts to run and operationalise complex deep learning projects. This will take the pressure off senior data scientists and move AI projects from concept to production much faster, improving their chances of success.
Step 4: Operationalise AI to accelerate and achieve your AI goals
Even once organisations have a working AI model, the project still needs to be operational before it can be deployed in the business. This means that AI projects must also be auditable, scalable and efficient enough to work in today’s business environment, where regulatory requirements, cross-departmental collaboration and scalability are vital considerations for any project. However, creating custom algorithms from scratch can cause problems for most businesses, as they require a huge amount of time, money and resource to integrate with other enterprise software, scale efficiently and overcome departmental IT silos.
Instead, organisations need to seek out the right partnerships so that the technology can be deployed in a way that supports their business model, easing common pain points and increasing the chance of project success. Deploying an operational AI platform that takes away some of the core challenges of deployment can reduce the need for specialist skills and remove the cost and complexity required to run AI projects. This will help organisations to move from concept to production much faster without the need for major investments.
Fail to plan…
The AI industry is booming, and organisations are set on reaping the benefits, but success isn’t assured. In order to overcome the cost, complexity and skills challenges AI projects present, organisations need to be planning carefully, gaining buy-in from IT and business departments from the offset. Then, introducing an operational AI platform can remove complexity and empower workers to start using the technology efficiently today to guarantee project success. If they don’t, a golden opportunity may pass them by.
Kye Andersson, AI expert, Peltarion