The success of any machine learning (ML) and artificial intelligence (AI) project depends less on the delivery of a working ML/AI model, and more on the value it delivers to the business.
That’s the view of Guy Taylor, executive head: Data and Data Driven Intelligence at Nedbank, who told delegates at the 15th annual ITWeb Business Intelligence Summit in Sandton this week that many of the technical issues around the deployment of ML and AI had been resolved.
”What is much more difficult is extracting value from the use cases that have been created, but it can be done,” he said, citing recent projects being undertaken at Nedbank as examples.
In one, an anti-fraud project, fraudulent transactions to the value of R137 million had been prevented in just six months. The other, a pilot scheme, had brought in revenues of over R250 million over the past four months.
If the algorithm does not support the business process, it is nothing more than an intellectual exercise.
Taylor maintained that the only way to ensure the delivery of value from a ML/AI project was if it was the result of a 'bottom-up' approach, which demanded close collaboration between the data science team and the business, rather than a 'top-down' approach.
“Everything you are trying to achieve with your AI projects has to be aligned to the business – and business is actually on the ground, not in the C-suite. The strategy might come from the top, but the pain is on the ground and that is where you will find practical use cases,” he said.
It was important for data scientists to realise that AI did not matter outside of the context of the business. “If the algorithm does not support the business process, it is nothing more than an intellectual exercise. This is where the gap between business and technology really exists,” he added.
First step
Because the ML/AI project was designed to help automate decisions, it was essential to identify exactly which decisions needed to be automated to achieve three key objects:
- bring in money;
- prevent the loss of money; and
- help to manage risk.
“The data science team cannot know the answers to these questions, at least not at the same level of detail as the business people. The data science team can help the business people answer the questions, but the business people have to be responsible and accountable for managing their own pain points,” Taylor said.
It was also important to remember that ML/AI was not about software development.
We have to remember that what ML is measuring is behaviour and because of this, the model changes all the time.
“You don’t just get to put your model into production and walk away. The tools that we have for machine learning are not the same tools we have always used – we have to build out these tools in a new way, dive deeper and create further ability into the system. We have to remember that what ML is measuring is behaviour and because of this, the model changes all the time,” he said.
Another critical success factor was getting the right people and skills in place for the project. However, this was easier said than done as the demand for AI/ML skills had increased by over 650% (based on LinkedIn job posts) in the last few years.
In addition, many of those applying for these jobs did not have the requisite skills, regardless of what was stated on their CVs.
Taylor had therefore started to recruit team members based on aptitude – aptitude particularly around curiosity as well as a hunger to learn and explore data. Once recruited, these bright young individuals were closely managed and mentored by senior data scientists, and BI and data analysts.
“It’s important to remember that AI/ML is not about the model – it’s about people. The biggest successes come when AI makes the lives of the people on the ground easier, so they don’t get in the way of the change management process and the project is able to deliver the value it was designed to do,” he concluded.
Share