Making machine learning fit for business

Machine learning is finding a role in all areas of business, but while some businesses are still experimenting, others are beginning to industrialise their approach. Lindsay Clark examines some of the lessons they have learned.

According to an O'Reilly survey of 11,400 data science experts, nearly half (49%) of organisations are at the early stages of adopting machine learning. Only 15% consider themselves sophisticated users of the statistical techniques expected to make such a big impact on business.

But even those with considerable experience in machine learning are still developing their understanding of how to apply it to real-world problems at a scale that benefits the business.

Lack of trust is a huge barrier to introducing machine learning in any business, according to Jagdish Mitra, chief strategy officer Tech Mahindra, a technology and business process outsourcing company.

He told a London conference hosted by machine learning software provider that the time it took to get return on investment from machine learning projects - on average more than one year - created scepticism among those expected to fund projects. As such, machine learning was more often applied to the problem of making existing processes more efficient, rather than transforming processes or helping the company find growth.

Walter Kok, director at ING, described how he had helped lead the introduction of machine learning to tackle the problem of money laundering at the global insurance firm. He said the key to gaining momentum was to be able to demonstrate benefits quickly, and directly relate that to customer experience or business performance.

"After we introduced and developed a model, there was resistance among the business team, but the model created a 70% reduction in false positives. That meant analysts could spend more time on real investigations," he said.

For a financial institution whose licence and reputation depends on its ability control fraud, this was a big benefit, he said. 

To continue reading this article register now