The industry has a habit of rushing headlong into the latest ‘vogue’ trend before considering why, and how, it makes business sense, and the same is true for data science.
While the term ‘data science’ has only recently become vogue in the advertising industry, it is not a new concept. Brands and agencies have been analysing various forms of online and offline data to refine products and target advertising for many years, to varying degrees of sophistication.
But several recent market factors — namely the tightening regulation over the collection and storage of consumer data — have forced businesses to rapidly upskill their data expertise, bringing ‘big data’ and data science into its heyday.
One only need to look to Salesforce’s eye-watering US$15.7 billion acquisition of Tableau, Google’s US$2.6 billion purchase of data analytics startup Looker and Publicis Groupe’s US$3.95 billion acquisition of Epsilon to understand the value of data analytics and management in 2019.
However, even now brands are underestimating the level of investment required — and steps that must be taken — to effectively build in-house data capabilities.
So let’s dive in.
Embed data scientists at centre of business, do not silo
Data experts unanimously agree that placing data scientists at the core of a business is more effective than having a siloed team.
“There are a lot of nuances to the business that you don't get by just looking at the data, and having a siloed team could lead data scientists to come up with conclusions that do not make sense in the context of the business,” says Appier vice-president of enterprise AI Charles Ng.
He adds that data experts can be more prevalent in certain regions (such as Japan), and that data scientists like working together — requiring businesses to strike a balance.
One local brand that has adopted this approach is Grab, which has embedded data scientists in each of its core businesses (transport, food, maps) from the outset.
“Doing this puts our data team closer to the customers and markets they serve, and allows them to get a better understanding of the problems they are meant to solve,” says Jagannadan Varadarajan, head of data science (machine learning) at Grab.
MullenLowe Group head of data science and analytics Jonathan Hart says brands often make the mistake of “throwing data scientists into the marketing mix” without first setting the business up for a new way of thinking.
“Data scientists are not another ingredient for your marketing team but a catalyst for something to change. Their skill set is built on engineering and maths, so they approach problems differently,” says Hart. “If you want the full value out of that investment you have got to be willing to change your thinking and processes, otherwise you are limiting the value they can bring.”
The challenge — and expense — of integrating data science capabilities into the fabric of the company is what most brands underestimate, Hart adds.
“Buying the technology and the talent is only the start of the journey,” he says. “There is a longer term cost to training talent and rethinking processes.”
Projects of this scale must first get “significant buy-in from the top”, he adds.
Having a centralised data infrastructure is not totally essential, but it helps
The condition of a brand’s data infrastructure prior to them hiring a data science team is cause for debate.
Many brands — especially the more traditional ones that straddle online and offline — have data sitting within different business units that has been collected in different formats, making it difficult to cross-analyse. Eventually, the business will need to clean up and centralise that data.
But while Ng believes this must be a precursor to any data science initiative, Hart doesn’t see non-centralised data as a major impediment.
“A lot of brands have gone through very expensive lengthy processes to try to tie all their data together to create a golden record for their customer. Those initiatives are not a prerequisite to good data science — it is also possible to create new data,” Hart says.
Indeed, Hart sees value in hiring a data science team and centralising data “concurrently”, so the centralised data lake can be set up by the team who are going to use it the most.
Once centralised, the data hub needs to be governed by data management experts whose job it is to ensure consistency and eliminate data that is no longer relevant, says P&G’s Peri.
“That is the key, because if you don’t govern a data lake it turns into a swamp and becomes a value deluder,” he says.
Help grow local talent, and be prepared to pay premium salaries
There’s also the more pressing concern of there not being enough talent to go around.
“More and more companies are understanding that data is a strategic advantage, but the talent pool is not growing as fast as the demand,” says Peri. “We are already experiencing a talent shortage.”
This is especially true in Southeast Asia, which has a high concentration of tech startups. In a 2018 report about Southeast Asian startups by Slush Singapore and Monk’s Hill Ventures, 90% of tech experts identified the skills gap as a “major issue” for the tech industry.
“When we first started out, hiring wasn’t easy. Many good talents had gone overseas due to the lack of opportunity in this region,” admits Grab’s Varadarajan.
To overcome this, Grab has taken a “global-local approach” to talent cultivation, setting up four R&D centres in Southeast Asia to grow the talent pool, with the remaining centres placed strategically in well-established tech hubs Seattle, Beijing and Bangalore. The ride-hailing app has also partnered with The National University of Singapore to help train the next generation of tech talent.
To be in with a chance of competing, brands should offer data scientists “meaningful, challenging” problems to solve, access to the world’s best technology, a workplace that values their skills, and of course, a competitive salary, Peri suggests.
Own the data, outsource amplification
Brands are turning to data science to help overcome two opposing challenges: not having enough first party data, and having too much data to manage.
The use cases of data science vary significantly across the two, but what everyone can agree on, is that it has never been more important for brands to own that experience.
“Data is a strategic asset for any company,” says Procter & Gamble chief data and analytics officer Guy Peri. “Just like you need to be in control of understanding your supply chain, you need to be understanding your data and how best to leverage it.”
The notion was echoed last month by Coca-Cola’s APAC marketing technologies leader Vidyarth Eluppai Srivatsan, who said at ATS Singapore it is easy for brands to outsource identity to someone else “then realise we are hooked onto that outlet”.
But there is a role for partners. Data providers can layer on third party data sets to create a richer offering, while agencies can help with attribution and tracking performance measures, Hart suggests.
“Media agencies or tech companies can be amplifiers to your in-house capabilities,” Peri adds.
Use cases vary depending on the relationship with first party data
P&G has grown its data science capabilities “significantly” over the last three years, from a team of two data scientists to a global division that includes data scientists, data engineers and data management professionals.
Data science forms a “core” part of the 182-year-old company’s strategy to stay ahead of the curve, explains Peri: “At P&G we are very focused on constructive disruption. We believe the entire industry is going to transform, and data analytics and data science will be at the core of that disruption.”
The FMCG brand uses a sophisticated data engine to glean insights from multiple third party sources.
For example, it has developed a ‘smart audiences’ capability to identify high propensity consumers that are most relevant to a particular brand. It employs this technique in China, where it analyses 3,500 different consumer attributes to triangulate which of the 1.2 billion consumers in China are most likely to buy.
“We use those smart audiences to buy our media, so we decrease any waste that our media is being spent against consumers that — no matter what we tell them, how much we advertise to them — won’t convert,” says Peri. “It is a good way to become more efficient and effective with our media spend.”
The ‘smart audience’ data is formed from 50-60 different data sources, including BAT (Baidu, Alibaba and Tencent) purchase data and P&G’s own consumer insights, Peri says.
Meanwhile, it is using machine learning algorithms to help account executives place appropriate products in stores across seven markets in Asia.
The ‘smart selling’ algorithms use a combination of geographic, shipment and demographic data to recommend the best products and promotions to offer to maximise value for the consumer and retailer. This is especially valuable in markets like India, for example, where P&G stocks 1.1 million retailers.
To reduce its reliance on third party data, it is one of a host of brand owners looking to gather first party data for the first time.
“In certain markets we have our own loyalty card programme where a consumer can buy Pampers, and scan the QR code to get points. That is a great example of first party data that is not reliant on anyone else,” Peri says. “We are doing more and more of that.”
Coca-Cola also revealed last month it is looking at ways to grow its first party data, such as gleaning mobile IDs from apps (likely payment) that interact with its vending machines.
“First-party insourcing and having that direct relationship with consumers is on the race,” said Vidyarth Eluppai Srivatsan, APAC marketing technologies leader at Coca-Cola, speaking at ATS Singapore.
On the other side of the coin, digital-first brands like Grab are sitting on treasure chest of first party data.
The super app claims it processes approximately 40 terabytes of first party data every day. In just seven years of operation it has formed a 250-strong data science team, making it one of the largest in Southeast Asia.
“Grab has a data-driven culture that’s embedded right at our very core,” says Varadarajan. “We use big data to derive insights to address customer pain points, evaluate model performance through experiments, and set up data-driven action plans.”
Data science and artificial intelligence have helped the super app decide which new products to launch (such as GrabShare in 2018), improve the user experience of its app (such as its food recommendations, travel time estimates and map geometry), as well as to personalise its marketing. It also uses machine learning to spot fraudulent activity.
For Grab data science forms the cornerstone of its product development, while for P&G it shapes how multi-millions of dollars worth of advertising is allocate. So while initial investment may be high, the pay-off is significant.