Getting the balance right: 3 keys to perfecting the human-A.I. combination for your business
After IBM’s Watson supercomputer beat two Jeopardy! champions in 2011, artificial intelligence seemed ready to take on the world’s greatest challenges. Indeed, following Watson’s widely publicized quiz show victory, IBM teamed up with some of America’s foremost medical institutions to use Watson’s algorithms to tackle the scourge of cancer. The hope was that Watson’s A.I. capabilities would analyze the vast amounts of cancer data the institutions had amassed, develop data-driven insights, and help care providers make more effective treatment decisions.
The initiative didn’t go as planned. Oncologists turned to A.I. for answers, but Watson couldn’t deliver for various reasons, including gaps and messiness in the data and A.I.’s inability to pick up textual cues in medical documents that were clear to doctors. Several of the cancer initiative’s projects eventually shut down.
But the oncologists and engineers in some of the projects took a different tack. Instead of blaming A.I. for not delivering results, they redesigned the respective roles of human and algorithm. They realized that A.I. could rapidly cross-reference a patient’s genetic profile against the gene mutations mentioned in thousands of digitized academic papers and identify treatments that could have been overlooked. Instead of asking Watson for a solution, they asked it for solution alternatives. In reframing the respective roles of the oncologists and A.I., both were able to play to their strengths: The A.I. shrank the time and effort needed to identify a comprehensive list of potential treatments, while the oncologists used their experience to choose among those treatments and deliver them to patients.
As business increasingly adopts A.I., company leaders should keep the lesson of Watson in mind. To generate the optimum results from their investments in A.I., they must understand the different ways in which employees and algorithms can be combined and choose the most effective human-A.I. combination for the challenge at hand. Here are three principles for doing so.
First, understand the human-A.I. combinations available to your company. Our studies show that there are four templates leaders can follow as they combine employees and A.I., as exemplified by Australian mining giant Rio Tinto.
- A.I. as Illuminator: A.I. generates data-driven insights to expand the breadth and the depth of employee thinking. Rio Tinto uses algorithms to analyze thousands of drill-core logs and generate 3-D models of ore bodies. That allows exploration teams to better understand the structural configuration of the overall ore deposit, which opens new avenues in the search for mineral resources.
- A.I. as Recommender: A.I. offers recommendations and employees decide on which ones to act. Rio Tinto uses A.I. to make predictive maintenance recommendations that its specialists use to create equipment repair schedules.
- A.I. as Decider: A.I. makes decisions and employees execute them. Rio Tinto applies machine learning and mathematical programming to make real-time decisions for dispatching the load-haul-dump (LHD) machines at its Argyle Diamond Mine in Australia; those decisions are then carried out by human operators.
- A.I. as Automator: A.I. makes and executes decisions with employee oversight. Rio Tinto’s Pilbara mines in Western Australia have minimal on-site staff, with most of its engineers remotely overseeing its operations from Perth, 900 miles away. Fully automated, driverless trains carry 28,000 tons of iron ore from the mine to a port 200 miles away.
Second, use a decision tree to determine the best human-A.I. combination for your application. As Rio Tinto demonstrates, no single combination of employee and A.I. is inherently superior to another. Instead, executives need to make sound human-A.I. combination choices by asking themselves questions that clarify the objectives, the context, and the outcomes they expect. Together, these questions, asked in sequence, constitute a decision tree.
- Objectives. Do we want to deploy a new business model, or improve the efficiency of an existing process? If the former, consider using A.I. as Illuminator, which will help spur creativity. If it is the latter, the other combinations (Recommender, Decider, and Automator) are better choices.
- Context. Do we have data that answers the questions that an employee will ask while solving the problem, and if so, can A.I. be trained to answer those questions using the data? If the data is available and it can be used to train A.I., consider using A.I. as a Decider or an Automator. If the data is not available or can’t be used for training A.I., use A.I. as a Recommender.
- Outcomes. Can deploying A.I. deliver better outcomes than deploying employees? (Typically, this is the case with large-scale, routine processes that require rapid execution.) If the answer to the question is “yes,” consider using A.I. as Automator. If not, use A.I. only as a Decider.
In addition to these questions, there is an overarching consideration attendant to A.I. No matter which human-A.I. combination companies choose, they must also earn the social license to use A.I., by designing fair and transparent algorithms; convincing stakeholders that the benefits of A.I. outweigh costs; and proving that they can be trusted with data acquisition and be accountable for their A.I.’s decisions. Moreover, as A.I. evolves and its role shifts from Illuminator to Automator, taking over a greater portion of the decision making and implementation formerly performed by employees, companies will face ever higher hurdles in winning society’s approval.
Domino’s Pizza offers a good, after-the-fact demonstration of how the decision tree can work. When customers in Australia complained that Domino’s products didn’t “look good,” the company turned to A.I. It developed DOM the Pizza Checker, a scanner equipped with computer vision, to compare every pizza that employees make against a database of ideal pizza pictures to ensure that it was visually appealing.
Domino’s objective was to optimize the output of the existing pizza-making process, and a properly trained algorithm was able to do this by checking the appearance of the pizzas (context). Pizza-making is a fast-moving and repetitive process, in which product consistency must be ensured at each store; therefore, more time and cost could be saved by using A.I.-powered robots to execute the entire process (outcomes). But it has to be the employees who make (and fix) the product in line with the company’s mission of delivering a quality handmade pizza: Automating the entire pizza-making process would put Domino’s social license at risk. This line of reasoning led Domino’s Australia to use A.I. as a Decider.
Third, regularly review and refine your human-A.I. combinations. The ideal mix of employees and A.I. evolves along with the technology and company-specific factors, such as objectives, capabilities, and risk tolerances. Thus, executives should periodically revisit the decision tree to ascertain if their human-A.I. combination is still optimal.
Aircraft engine-manufacturer Rolls-Royce uses digital twins and A.I. to make preventive maintenance recommendations to its customers. Over time its Engine Health Monitoring (EHM) system has become more sophisticated, measuring more performance parameters and feeding richer data streams into its algorithms. As this trend continues, we expect to see the EHM system evolve from a recommendation to decision system, telling airlines when and where to service specific engine parts.
The way in which your company uses A.I. is no small matter. A recent BCG-MIT study revealed that companies that consider and choose the right human-A.I. combinations are six times as likely to realize significant financial benefits from A.I. than companies that don’t. To unlock the full potential of A.I., think about how people and technology will work together.
Read other Fortune columns by François Candelon.
François Candelon is a managing director and senior partner at Boston Consulting Group and the global director of the BCG Henderson Institute.
Bowen Ding is a project leader at BCG and an ambassador at the BCG Henderson Institute.
Matthieu Gombeaud is a principal at BCG and ambassador at the BCG Henderson Institute.
Some companies featured in this column are past or current clients of BCG.
Sign up for the Fortune Features email list so you don’t miss our biggest features, exclusive interviews, and investigations.