The three biggest AI challenges businesses face

The three biggest AI challenges businesses face

Artificial intelligence (AI) is becoming pervasive in everyday life but many businesses still apply AI for the sake of AI, rather than for the sake of real business value. This was one of the main messages to emerge from the SAS Road to Artificial Intelligence event, held in Johannesburg on 28 June.

In his opening address, SAS Africa VP Desan Naidoo said businesses are officially operating in an analytics economy and the new currency is data-driven value.

Gartner predicts that global business value derived from AI will reach $1.2 trillion this year, a 70% increase from 2017, rising to $3.9 trillion in 2022. AI expert, Dr Jacques Ludik, said examples of business value include increased productivity and revenue, and reduced costs and risk.

To succeed in this economy and derive true business value, Naidoo said organisations need a blend of data, analytics and collaboration. But they face three main challenges when it comes to AI: knowing what to optimise, accessing and preparing data, and skills.

Optimisation conundrum

For many businesses, one of the biggest challenges with AI is knowing what to optimise, said Prenton Chetty, Senior Manager of Advanced Analytics at Nedbank. For Prenton, the current approach, where humans make decisions based on data outputs, is flawed because human biases quickly creep into the decision-making process.

Rather, he said we should empower the machines to make decisions based on data and facts. But AI has not yet evolved to this point, he cautioned. Until it does, organisations should start by using the data they have available to optimise processes and solve small, specific business problems.

For Nedbank, this involved using propensity, attrition and collection models to predict which customers were likely to churn and which were likely to take up a new product.

The fact that businesses collect and process different data means the application of AI will be different for each organisation, depending on the problems they need to solve, said Larry Orimoloye, Senior Business Solutions Manager for Advanced Analytics and AI at SAS.

For Orimoloye, AI means different things to different people and businesses, depending on their experience of it. For some, AI is a chatbot that assists with financial transactions. For others, AI is image processing, where a machine is able to identify cancerous growths. AI is text analysis when gauging sentiment on social media; it’s facial recognition to control access to restricted areas in a manufacturing environment. It’s anything and everything, and each application is different.

Orimoloye adds that while the application and customisation of AI are as diverse as the number of businesses and their problems, there is one common goal: to automate repetitive tasks and free up human resources. Despite the common goal, there’s no one-size-fits-all solution. Rather, algorithms are built for specific businesses and specific problems, based on the data available. But this presents new challenges.

Data

Speaking during a panel discussion at the event, Rajesh Duvooru, Senior Manager for Artificial Intelligence at Accenture South Africa, said there was no point implementing AI if businesses did not have the data or data governance in place.

For him, one of the biggest AI challenges lies in getting the data, cleansing it and using it. To thrive and differentiate themselves in the analytics economy, he said businesses need to start making decisions based on data and analytics, and to make those insights available to all business users – not just the C Suite.

Orimoloye agreed, saying that before businesses embark on an AI project, they need to think critically about their datasets. In any project, he says the majority of time and effort is spent massaging and preparing data so that it can be fed to machine learning algorithms to produce valuable insights.

For Ludik, the more data businesses make available for analysis, the more value they’ll get. But he stresses that the data has to be rapidly available and in a good state: bad data in equals bad insights out.

Evolving skills

Data and analytics skills are still in high demand and short supply, according to the panel. With so many different solutions, programming languages and analytics platforms available, Dr Benjamin Rosman, deep learning and robotics researcher and lecturer at Wits University and the CSIR, said educators were focusing on producing flexible graduates who can quickly learn and work with any tool, depending on the business requirement.

Prenton noted that graduate CVs are changing, showing not just university degrees but also links to additional online courses, Kaggle projects and a blend of open and proprietary software skills. It was encouraging, he said, that graduates were experimenting with AI and extending their learning beyond university because, at the end of the day, it doesn’t matter what language they use, as long as it integrates into the rest of the system and solves a business problem.

Echoing Naidoo’s thoughts, Ludik said collaboration was key to AI success. Businesses need to draw on the skills and experience of a mix of people who can work with many different analytics platforms to produce end-to-end solutions that can be integrated into processes and workflows.

Any business that wants to start deriving value from AI should start with a small, specific business problem that results in a quick win, the experts advised. AI is already disrupting successful business models and the wave of disruption has only just started, said Naidoo.

For Prenton, there’s only one approach to getting started: “Start somewhere. Start simple. Show the business impact.”