JustGiving and eBay may have a category difference when it comes to their size – the fundraising platform has helped more than £4 billion to be collected since it launched in 2001, while the auction site sold around £16 billion of stock in Q1 of 2017 alone. When it comes to their use of data and analytics, however, they have a lot in common.
From using as much data as possible to fuel artificial intelligence and advanced analytics, through to embedding a data-driven culture in the organisation, the two companies are following the same path. They even share similar objectives – to grow the world of giving and to expand the range of goods bought and sold via the online platform.
From the experiences shared by Mike Bugembe, chief analytics officer at JustGiving, and Mike de Halpert, director of European analytics at eBay, at the DataIQ Summit, it is possible to identify some shared success factors for building and sustaining a strong analytical culture.
1 – Provide leadership
As Bugembe put it, having somebody who is “religious about data” is vital, but so is where the CAO sits within the organisation and who they report to. “The key thing is that they sit close to the strategy and vision because they need to educate the whole organisation and get the language right,” he said.
The fundraising platform has the advantage that its two founders have always been strong advocates of technology as a way of gaining competitive advantage – it launched a social giving app on Facebook in 2007. “The next big thing was embracing big data – I love the term – given the number of users on the platform with 23 million people transacting and 20,000 charities,” said Bugembe.
2 – Create an analytics-oriented culture
Anybody familiar with operational research will tell you that changing the culture of an organisation is one of the hardest things to do. For de Halpert, part of the solution has been the physical proximity of practitioners to the customers they serve.
“There is an inverse correlation between the quality of your analytics and the distance from the things you are analysing”
“What is crucial, regardless of how the business is set up, is that analysts need to sit next to the lines of business. There is an inverse correlation between the quality of your analytics and the distance from the things you are analysing,” he said.
Listening to what the business needs is also essential. When he joined eBay six years ago, marketing was carrying out one or two A/B tests each month, running T-tests for confidence, then rolling out. This had created an analytical bottleneck. To break it, de Halpert asked the business how many tests it would like to do – that took the volume up to 500 per month.
With the adoption of SAS and R as core tools in the analytics function, that increased further to 1,000 tests per month, which created a new bottleneck – the marketing team was only able to come up with 150 hypotheses per month. Now, eBay is applying AI to study its CRM data and generate the hypotheses. “Marketers obviously didn’t think about the threat to their job security,” he joked. The team now has four analysts who pick up those hypotheses, which up to 20 operational developers will then deploy. “It has taken on a life of its own,” he said.
While JustGiving only has two analysts, it has ensured they are embedded right at the heart of the organisation. Bugembe agreed that ensuring internal communications are clear is critical: “A lot of businesses don’t tell anyone they are trying to build a data-driven culture. That’s why it is fundamentally important to have your data leader in the organisation’s leadership team. That is one of the key signals that the business is serious about it. If the CAO is reporting to the CIO, it will not be a success.”
3 – Use lots of data
It may seem an obvious point, but the impact of using data can be significant, compared to deploying the “best” model without considering enough data. Bugembe pointed to the Netflix algorithm competition in which two teams from Stanford University submitted entries, one using a very sophisticated algorithm and the other using a simple algorithm built on all of the data it extracted from IMDB. The second proved to be the winner.
From its own analytics, Bugembe noted that, “we discovered things like ‘hero sponsors’ – people who want to be the one making a donation that gets a fundraiser over the line – or a ‘free rider’ who will look at what other people have been giving and might wait for somebody to donate £5 before they give £5 where the previous average had been £20.”
4 – Use the latest tools
As well as using AI to develop marketing hypotheses, eBay has also been running AI against customer feedback and chat. “Before, we used to examine email as a silo, now AI looks at every way a customer interacts with the business,” said de Halpert.
He describes the first exercise as “a revelation,” since it showed that opening emails is a function of having a live listing on the site for some customers. “I see AI like the internet was in the 1990s – a lot of people are throwing their spaghetti at the wall, but they don’t know what will stick. That happens when you throw AI at an entire problem set, not at individual issues.”
Bugembe has adopted machine learning and graph analytics to understand how users behave on the platform. “Only the machine can understand all of those factors, for example, that not everybody wants to do a marathon,” he said. “So we can help them to create a page with the right type of event based on those propensities. Fundraising is not an everyday event – people don’t know what images, texts or targets they should employ. The data helps us to educate our users about what to put on their page.”
5 – Build a team and keep it fresh
Bugembe warned against seeking a “unicorn data scientist” who is capable of everything, but instead advised building a group which has the skills required across data architecture, analytics and production. “You need multiple roles and also traditional analysts alongside data scientists,” he said.
“If you don’t update your skills, in five years you will be obsolete.”
Although de Halpert sat his last maths exam aged 15, he took a Masters in Economics and has progressed his skills set from Excel and SQL through R and Python and now AI and ML. “I say to my analysts, you may be experts in SQL and Excel, but there are a million analysts with those skills in China and India – and they are cheaper. If you don’t update your skills, in five years you will be obsolete.”
He also offered a valuable tip about keeping analysts’ skills fresh: “Ask your analysts what they do in their spare time – if they are not learning by themselves, they will fall behind. One of our analysts was working on image recognition which even the US was not doing.”
