How to win with machine learning

How can you use machine learning to create something competitors can’t easily imitate?

The most successful users of artificial intelligence and machine learning develop high-quality predictions.

The past decade has brought advances in an exciting dimension of artificial intelligence: machine learning. These developments have enabled tech giants such as Apple, Facebook and Google to dramatically improve their products. They have also spurred startups to launch new ones.

Many companies are already working with AI and are aware of the practical steps for integrating it into their operations. But how can they take advantage of machine learning to create a defensible moat around the business — something that competitors can’t easily imitate?

Businesses use machine learning to recognise patterns and then make predictions — about what will appeal to customers, improve operations or help make a product better. Before you can build a strategy around such predictions, however, you must understand the inputs necessary for the prediction process, the challenges involved in getting those inputs and the role of feedback in enabling make better predictions over time.

A prediction, in the context of machine learning, is an information output that comes from entering some data and running an algorithm. For example, when your mobile navigation app serves up a prediction about the best route between two points, it uses input data on traffic conditions, speed limits and other factors. An algorithm is then employed to predict the fastest way.

The key challenge is that training data — the inputs you need to start getting reasonable outcomes — has to be either created (by, say, hiring experts to classify things) or procured from existing sources (say, health records). Consumers may also willingly supply data if they perceive a benefit from doing so. Fitbit and Apple Watch users, for example, allow the companies to gather metrics about their exercise level, calorie intake and so forth through devices that users wear to manage their health.

Obtaining data to enable predictions can be difficult, however, if it requires the co-operation of individuals who do not directly benefit from providing it. A navigation app can collect data about traffic by tracking users and getting reports from them. This allows the app to identify traffic jams and alert other drivers who are heading toward them. But drivers caught in the snarls get little direct pay-off from participating, and they may be troubled by the idea that the app knows where they are at any moment.

Another challenge may be the need to periodically update training data to reflect changes in the environment. With navigational apps, for instance, new roads and renamed streets will render the app’s predictions less accurate over time unless the maps are updated.

In many situations, algorithms can be continuously improved through the use of feedback data, which is obtained by mapping actual outcomes to the input data that generated predictions of those outcomes. For instance, when your phone uses an image of you for security, you will have initially trained the phone to recognise you. But your face can change: You may be wearing glasses or may have gotten a new hairstyle. The prediction that you are you would become less reliable if the phone relied solely on the initial data. Instead, the phone updates its algorithm using all the images you provide each time you unlock it.

It can also be dangerously easy to introduce biases into machine learning. Suppose a lender uses AI to assess the credit risk of loan applicants, considering their income level, demographic characteristics and so forth. If the data for the algorithm discriminates against a certain group — say, people of colour — the feedback loop will perpetuate, making it likely for applicants of colour to be rejected.

Building a sustainable business in machine learning is much like building a sustainable business in any industry. You have to come in with a sellable product and make it harder for anyone to come in behind you. Whether you can do that depends on your answers to three questions:

1. DO YOU HAVE ENOUGH TRAINING DATA? A prediction machine needs to generate predictions that are good enough to be commercially viable. The definition of “good enough” might be set by regulation (for example, an AI for making medical diagnoses must meet government standards), usability (a chatbot has to work smoothly enough for callers to respond to the machine rather than wait to speak to a human in the call centre) or competition (a company seeking to enter the internet search market needs a certain level of predictive accuracy to compete with Google). One barrier to entry, therefore, is the amount of time and effort involved in accessing sufficient data to make good predictions.

2. HOW FAST ARE YOUR FEEDBACK LOOPS? Prediction machines exploit what has traditionally been the human advantage — they learn. If they can incorporate feedback data, they can improve the quality of the next prediction.

When Microsoft launched the Bing search engine in 2009, it had the company’s full backing. Microsoft invested billions of dollars in it. Yet more than a decade later, Bing’s market share remains far below Google’s. By the time Bing entered the market, Google had already been operating an AI-based search engine for years. Every time a user made a query, Google provided its prediction of the most relevant links, and then the user selected the best of those links, enabling Google to update its prediction model. With so much training data based on so many users, Google could identify new events and new trends more quickly than Bing could. The fast feedback loop has meant that Bing has always lagged.

3. HOW GOOD ARE YOUR PREDICTIONS? The success of any product ultimately depends on what you get for what you pay. If consumers are offered two similar products at the same price, they will generally choose the one they perceive to be of higher quality.

Prediction quality is often easy to assess. Companies can often design AIs with a clear, single metric for quality: accuracy. As in other cases, the highest-quality products benefit from higher demand. AI-based products are different, however, because for most other products, better quality costs more; sellers of inferior goods survive by using cheaper materials and charging lower prices. This strategy isn’t as feasible in the context of AI, where a low-quality prediction is as expensive to produce as a high-quality one, making discount pricing unrealistic.

The potential of prediction machines is immense. The key to competing successfully in industries powered by intelligent machines lies in a question that only a human can answer: What is it that you want to predict?

The challenge

As more companies deploy machine learning for AI-enabled products and services, they face the challenge of carving out a defensible market position, especially if they are latecomers.

HOW TO GET AHEAD: The most successful AI users capture a good pool of training data early and then exploit feedback data to open up a value gap — in terms of prediction quality — between themselves and later movers.

HOW TO CATCH UP: Latecomers can still secure a foothold if they can find sources of superior training data or feedback data, or if they tailor their predictions to a specific niche.

Harvard Business Review