McKinsey: Why some industries gain an AI advantage, while others lag

McKinsey’s Global AI survey has found that the use of artificial intelligence (AI) in automating business processes has increased by 25% year on year, yet a number of sectors are struggling.

In a global survey of 2,360 businesses, McKinsey found that 58% of organisations have embedded at least one AI capability into a process or product in at least one function or business unit. This is up from 47% in 2018, which means AI adoption is becoming more mainstream, it suggested.

According to McKinsey, the top quarter of businesses – the 54 “AI high performers” that took part in the study, where AI is being used in five or more business activities – reported seeing an average revenue increase of at least 5% from AI adoption in the business units where AI is used. The study also found that these businesses saw an average cost decrease of 5% or more from AI adoption in the business units where AI is used.

McKinsey’s research found that these high performers in AI are nearly three times more likely than those from lower-performing companies to report revenue gains of more than 10%.  It also found that businesses are most likely to report revenue growth from AI when it is deployed in marketing and sales, product and service development, and supply-chain management.

Overall, almost two-thirds of the organisations that took part in the study said they saw revenue increases from AI adoption in the business units where they use AI. 

About one-third of respondents said they expect AI adoption to lead to a decrease in their workforce in the next three years, while one-fifth expect an increase. Those organisation identified by McKinsey as AI high performers tend to do more retraining.

According to McKinsey’s data, the sectors with the most high performers are high tech, telecoms and financial services. McKinsey said there were no respondents in electrical power and natural gas that met the criteria to be classified as AI high performers.

The research showed that infrastructure, professional services and the pharmaceutical sector are the sectors that generally score lower for adoption of various AI-related technologies, such as machine learning and robotic process automation.

According to Jacomo Corbo, co-founder and chief scientist at QuantumBlack, which was acquired by McKinsey in 2015, infrastructure companies are generally slower with enterprise software implementations, which means IT systems and data are more functionally siloed. “Both business-side and IT department resources generally have fewer resources adept at configuring and building new data services and workflows, including those that embed AI and machine learning,” he said.

Looking at the professional services sector, McKinsey said it, too, had similar issues to the infrastructure sector and was at a disadvantage because of the lower level of sophistication of the enterprise IT systems deployed. This leads to challenges with integration and data.

AI in professional services is further complicated by the nature of their work, which tends to be very bespoke and non-repeatable. “Many professional services firms may not have access or licence to a lot of the data they generate,” said Corbo.

Asked about the pharmaceutical sector’s lower score in AI adoption, Corbo said: “We see a lot of variation within this sector. Some companies are definitely adopting AI and machine learning at scale across functions, but on the whole there is still relatively little penetration at scale in the sector.

“Much of that is to do with a combination of factors: established practices around IT outsourcing that is now starting to reverse itself as companies build capabilities around AI; a complex data landscape including legacy systems, made more complex by regulatory oversight and different requirements by function and different regulators in different geographies.”

Significantly, across all sectors, irrespective of the level of AI maturity, McKinsey’s study showed that less than half of respondents (41%) said their organisations comprehensively identify and prioritise their AI risks. Corbo said: “This reflects that the majority of companies, including those that are adopting AI, recognise that they are not being systematic nor comprehensive about how they are identifying and mitigating risks associated with the design, development and operational deployment of AI.”

As Computer Weekly has previously reported, organisations are starting to realise that to deploy AI, they need to give careful consideration to fairness and bias in the training datasets used for machine learning and AI, and AI algorithms need to be accountable and transparent.

Corbo said enterprises should also assess data model drift and degradation, and the potential for adversarial attacks, where a hacker deliberately feeds the algorithm corrupted data.