Artificial intelligence (AI) can be transformative for business – but taking the wrong approach in implementing it can result in very costly failures.
So says Willem van Eeden, Technical Lead at Entelect, who notes that many organisations turn to Entelect for help on the back of failed AI projects.
“It’s generally reported that up to 80% of AI projects fail. Although it’s not clear where this figure comes from, what is true is that many attempts to deploy AI to solve a business problem have had disappointing outcomes. The AI itself works, but the business use cases, foundational technologies and correct strategies were not in place to properly harness AI,” he says. “You have to plan your AI journey thoroughly.”
Use cases for AI
Van Eeden says AI performs exceptionally well in high-frequency, high-volume tasks and areas involving huge amounts of data. Use cases that check these three boxes are a good bet for success.
“The best use of AI involves high-frequency things that humans can't actually do – for example, predicting risk in the insurance industry, or handling calls in contact centres,” he says. “AI and machine learning (ML) solutions are also desirable where the cost of a mistake is low. It's riskier to use AI for low-volume, low-frequency tasks which are often high risk areas – such as a bank deciding to freeze a customer’s account. We can use AI technology to screen existing customers and see whether their disclosed details match what information we can find about them. But the decision to actually freeze accounts, stop doing business with a customer or place a customer under a strict audit is a high-risk endeavour that could have significant reputational risk if done incorrectly.”
Planning to act on outcomes
Van Eeden says: “One key area organisations overlook is what to do with the forecasts and analysis gained from AI. People forget to ask ‘what then?’. If AI was a Magic 8 Ball producing true predictions, what are you going to do with that information? If a bank uses AI to predict which customers will churn – and when – they also need to have a plan for how they will react to this prediction and how this could change the business process. It is equally important to plan for what you will do if the AI’s prediction is wrong.”
He notes that it is crucial to plan for the impact of AI on the business, its stakeholders and customers.
Getting the basics right
Another common reason for AI disappointments is that AI can be too advanced to solve a particular problem. Van Eeden says: “Organisations often believe they need AI when all they actually need is a process change or a good software system, such as a good dashboard or business intelligence (BI) tool. If they don't have the basics in place yet, they should first implement those, and do it right, then bring in the smart stuff later.”
He notes: “Entelect specialises in getting to the core of the problem. In our view, AI sits at the pinnacle of the hierarchy of needs, and you need to build this hierarchy from the bottom up, to get to the next level. We generally recommend starting with statistics and basic engineering before moving up to AI. If you don't get the basics right, AI can be little more than a very expensive research project.”
Organisations cannot expect to leap directly to predictive or prescriptive analytics without having the basics in place. Analytics maturity begins with descriptive analytics (what happened?), then moves to diagnostic analytics (why did this happen?), before predictive analytics (what might happen in the future?) and the very advanced prescriptive analytics (what should we do next?) and the pinnacle – cognitive analytics – which simulates human thought.
“It’s important to benchmark your organisation and understand where you fall within the analytics maturity process, and move forward from there,” he says.
“Deeply understanding the use case and where your organisation falls in the analytics maturity process can be transformative, leading to real action and circumventing costly failures,” he concludes.
Share