The drill has been the same for decades: Companies have shipped low-value, repetitive, or labor-intensive work to cheaper locations around the globe to save costs.
But now that cost equation is changing – and not simply because labor rates are rising around the globe. Machine learning is causing some companies to rethink their sourcing rationales. Indeed, it could remake the business process outsourcing (BPO) industry. Rote tasks once ripe for offshore centers are being taken over by machines, and outsourcing providers are being forced to come up with new value propositions for customers.
In our recent study “Making the Most of Machine Learning: 5 Lessons from Fast Learners,” conducted by SAP and the Economist Intelligence Unit, we asked companies not only about their adoption of machine learning but also how they were sourcing their business processes. We found that those companies that were seeing benefits from machine learning were spending more today on business functions sourced locally (whether performed in-house or by a third party) than in distant geographies. In fact, 58% said they spend more than half their budget for business processes locally (22% spend at least that much in low-cost regions). By comparison, 39% of those companies that have yet to see real value from machine learning spend more than half their business process budget locally, and 29% of these said they do more of their sourcing in low-cost regions.
It seems that companies that have embraced machine learning as part of their larger digital transformation strategy can increasingly make their sourcing choices based on the value it provides to their customers rather than cost alone. That makes them more likely to keep their most strategic business processes close to home. Companies that build their own machine learning capabilities will have less need for long-term outsourcing agreements or offshoring arrangements to help the business grow; AI capabilities become a force multiplier for their existing workforces.
Consider Intel, an extremely early adopter of machine learning to increase efficiency and quality in its factories, which is now applying those capabilities to its customer-facing business processes. With more than 100,000 reseller-customers, Intel’s own sales force could focus only on its largest clients. A new sales enablement system, powered by machine-learning algorithms, can identify those resellers that offer the highest probability for sales while keeping the sales process in-house. The system has already delivered more than $100 million in additional revenue, according to Intel’s chief data officer and vice president of Enterprise Data and Platforms, Aziz Safa.
Those enterprises at the leading edge of machine learning adoption are “not necessarily looking for near-term cost reductions,” says Stanton Jones, director and principal analyst with business transformation and sourcing consultancy Information Services Group (ISG). “They’re looking for things like improving productivity, compliance, or customer satisfaction.”
Outsourcing service providers, meanwhile, are increasingly incorporating their own machine learning tools into their services as intelligent automation becomes a competitive differentiator in the marketplace. “Organizations are going to be evaluating the efficacy of the technology that the provider is bringing to the table,” says Jones. “Providers can create a strategic advantage by, for example, having a machine learning algorithm that performs and learns more rapidly than a competitor’s.”
Will machine learning result in the mass repatriation of all the work currently done in offshore locales? Perhaps not. “But when it comes to newer transactional processes, decisions are more likely to lean toward keeping them onshore,” says Arjun Sethi, a partner at the consultancy A.T. Kearney. Those that do bring processes back home will be able to hire a much smaller number of higher-skilled employees – 25%-30% of the number used offshore, according to Jones – and match them up with intelligent automation.
One interesting short-term counter-trend, though. Not all low-value, repetitive work can be automated by machine learning – yet. Tasks requiring even the slightest degree of intuition or inference still baffle an algorithm – such as telling the difference between a cat and a house, for example. Machine learning can only tell the difference if it is fed millions of images of each of them.
Guess who’s doing that? “Most of the software that’s being developed to power autonomous cars is heavily using machine learning, and these machine-learning algorithms are having to take in very, very large amounts of unstructured data [in the form of] videos and photos,” Jones says. “There are large teams that are identifying what a stop sign looks like over and over and over again. In many cases, that training is taking place in lower-cost offshore delivery centers.”
Interested in learning more about the Machine Learning Fast Learners? Read the study here.