Using machine learning to shape the customer experience
As digital technology continues to evolve, it might be easier for companies to adopt processes that improve their own efficiency and work for their systems, rather than apply a truly customer-centric approach to product or service enhancements.
The world’s most innovative companies, like Google, Facebook and Apple, however didn’t give in to this temptation. They’ve built their success on putting their customers first - by listening to what those customers want. And they’ve employed machine learning to do that.
Machine learning is a form of artificial intelligence that provides computers with the ability to learn without being explicitly programmed.
Normal programming tells a computer how to operate in a given series of conditions, whereas machine learning is similar to our human thinking processes. It constantly responds to feedback and repeatedly tries to create better solutions.
Many companies, including Australia Post, have adopted machine learning to drive meaningful improvements to customers across products and services.
- Building a product on the foundation of customer feedback
- Human learning from machine learning
- Machine learning and the agile methodology
“Many companies are using machine learning today,” says Stephen Bellchester, machine learning specialist and senior developer at Australia Post. “It lets us create new products for the customer that we wouldn’t have the opportunity to do otherwise.”
“For instance, our customer research around bill payment shows that people want to keep, manage and pay all their bills in one place. Providers are also keen for customers to switch from paper bills to electronic billing. So we’ve used machine learning to develop a financial tool that helps customers manage their bills.”
The tool, which is currently in beta trial, connects to any Google or Microsoft web-based email address to extract bills from legitimate providers. The bills are stored in one location and notifications are sent to your smartphone when each one is due.
“Machine learning gives us the opportunity to remove ourselves from the process and let the continuous customer feedback weave its magic,” Bellchester says. “It means that the end product will be a pure reflection of what our customers really want and are able to use.”
The underlying advantage of machine learning is that product developers undergo their own form of informal education. When asked what lessons his team gleaned from the development of the bill management tool, Bellchester reels off a list that is applicable to all machine learning initiatives.
1. Getting access to a high volume of data
Bellchester points out that machine learning feeds off data and getting enough data to train it can be a big challenge.
“We needed hundreds of PDF bills from various billers and in various forms like an overdue bill, a direct debit bill and so on. Getting that first test set was a huge task.”
2. Understanding how to use that data
The only way to reach the right answers is to understand how to prepare and input the right data into the machine learning algorithms.
“The process of creating something with machine learning is more of an art than a science. It’s not the end if you run into a dead end. It’s a matter of stepping back, regrouping, thinking about the problem again and approaching it from a different angle.”
3. Having the right expertise on the team
Bellchester’s team was fortunate enough to include a member who had just completed his Master’s degree in machine learning. Unsurprisingly, he was instrumental in the set up and tapping into the full capabilities of machine learning.
4. Securing customer trust
The very first hypothesis Bellchester’s team tested was whether consumers would allow them the tool to connect to their emails. The answer turned out to be yes.
"We needed our customers to grant us access to their email inbox so the machine learning could search for different types of electronic bills from multiple providers in Australia. And because privacy are absolutely key to what we’re doing, we’re very careful to only take emails from recognised billers.”
The infinite capabilities of machine learning, coupled with the agile methodology, allows Australia Post to reiteratively refine a product or service, at speed and using relevant data that’s being fed back by the end user.
“If we’d been trying to do this in a waterfall way, I honestly don’t know what we would have built, because we didn’t know what we wanted to build at the start.” Bellchester says. “Trying to solve a problem without having to follow a rigid long-term plan meant that we achieved a much better outcome.”
“By talking to customers about what we’d achieved so far, getting their feedback and constantly iterating, we managed to get to where we are today so much faster.”
Machine learning isn’t a magic bullet solution for all of our future problems, but when used with intent by experts, it can help humans quickly create new products and services that our customers actually want to use.