“Data is the new oil” — as the frequently misunderstood "truth" states. But it doesn’t mean that hoarding huge amount of raw data will make you a rich oil-sheik (or data-sheik). Just like oil, raw data isn’t valuable in and of itself. Data is more valuable when it is refined and turned into a data product that can support decision making and fuel intelligent automation.
In this article I cover the three major use-cases of data products using real-world examples on: 1.) data-informed decisions making supported by curated data sets; 2.) intelligent business processes automation and 3.) intelligent applications;
Before we delve into the details, let’s have a look on why data products are important.
In the past 20 years, digitalisation disrupted many industries. Think about it for a moment: if the COVID-19 pandemic hit Earth just 20 years ago, many of the things we take for granted today wouldn’t be available. We would not be able to renew our car and household insurance from our mobile phone, order groceries from the comfort of our couch, or would not be able to order medicine from one of the online pharmacies. Kids wouldn’t be able to follow classes on their iPads either.
Digitalisation of every aspect of our life was a crazy ride so far but is far from being over.
Yet, in 2021 the mobile app as a digital frontend for a business or main source of differentiation is not enough anymore.
According to Gartner, less than 0.01 percent of all consumer mobile apps had a chance to become financially successful in 2018.
So what’s the next frontier then, the next source of scale and differentiation…? — you could ask.
This time we need a few, more transformational steps and make sure that we work smarter, not harder. Smarter means: we make data-informed decisions in all corners of the organisation all the time. It also means that we upgrade our business process automation with adaptive intelligence. In other words, we start embedding more “data products” in the fabric of our business.
Hmm, data products you say?… an interesting term, but what is that, how does it look like, and how does it impact my business? How is it different from analytics or AI? Curious? Let’s have a look at the three types of data products and their impact on the business.
The first form of a data product: curated data sets, supporting data-informed decision making
Let’s first meet Rent the Runway, a high fashion renting service. We will have a look together at how this company shook up a century-old industry using data products.
Rent the Runway scaled the rental model up to become a business with yearly revenue of over $100 million and a valuation of $800 million. Jennifer Hyman — the company’s CEO — said in an interview, that one of the areas to first plow resources was data science. Rent the Runway collects a myriad of data points about the customer experience: the sizes, the body type, did she wear it, did she love it. The company knows if their customers are pregnant, planning a meeting, or going to Miami next week. They use this data to predict demand and optimize stock and shipping. Algorithms crawl customer reviews to learn which dresses women are renting for certain occasions. With every dress, it lends, Rent the Runway’s algorithms get a bit smarter when it comes to setting prices and controlling inventory.
But more interestingly, Rent the Runway is using this data to build a close partnership with the fashion brands. The very ones that they were the most concerned about, because they could have held them back from building what they call today — “a closet in the cloud”. Fashion brands could have looked at a fashion rental service as a threat to their business structured around exclusive, luxury retailers. Despite the initial concerns, today Rent the Runway exchanges curated insights with the designers, who use this data to make their future collections even more successful.
These insights, covering the customer preferences are presenting a data product. It provides answers to the question of why one item is more popular and what aspects should be changed to be on top in the next season.
It also answers the questions about the durability of the items: there’s a huge difference in the bottom line between lending the same item 8 or 30 times. It is priceless to know which zipper, which textile and which dry-cleaning technique leads to an improved lifetime of the clothing.
How is that different from the good old reporting and business analytics?
In the “good old days” Rent the Runway would have looked only at Conversion Rates, Average Order Value and Revenue Per Visitor.
That provides a limited window on the events of the past. It describes what happened with the sales numbers in the last quarter, but seldom provides an answer as to WHY it happened like that. Insights offer more depth to the KPIs. When you observe that last month’s sale trend is dipping, you should be able to ask who, what, when, where, why, and most importantly how.
Modern data products go way beyond colourful static charts and enable you to autonomously interrogate the various dimensions of the model, using interactive features of the business analytics tool of your choice. You can ask: what is the sentiment of the customers towards my product lately? What is the impact of our latest campaign? Why is the impact like it is? Where should I advertise in the future? What happened to the conversion rates? Are those changes correlated to our last feature release or pricing agreement change? How can I improve? What stage of the customer journey leaks the most?
To answer those questions, data teams need to collect and process far more data than usual and to collect it from disparate systems stored in various formats. They extract customer sentiment from support calls or emails as audio and freeform text, product ratings from feedback forms. They track the level of engagement by implementing click-stream analytics or web-log-scraping. They package all this as well-defined data models, organised around a central theme or key business process and make sure the data in the model is up to date and free from data errors such as statistical outliers, duplicates and so on.
They use automated data pipelines that continuously extract the data changes from the sources, refine it, and transport it to its data consumer.
The second form of a data product: predictive analytics fuelling intelligent automation
Staying for a moment at the Rent The Runway example: knowing a thing or two about the popularity of clothing items, what breaks, and when it breaks is a game-changer for warehousing too. It helps to predict demand and calculate the just-enough supply levels, to ensure every customer gets what she needs and when she needs it. Meanwhile, working capital is not tied up in unnecessary inventories, that generates extra handling and storing cost.
That in itself is a good motivation for businesses to look behind the second form of data products: algorithms that can predict the future… at least parts of it ;)
Inventory optimisation is a key ingredient of “operational excellence”, one that has a large contribution to the bottom line and frees up resources for scaling.
But the power of predictions doesn’t stop here: it fuels intelligent automation, leading to new business models, disrupting traditional ones. It is also not specific to industries like eCommerce. A great example of that is Komatsu — a construction and mining equipment manufacturer — who is the forerunner of predictive analytics in the heavy machinery industry.
Komatsu was founded over a hundred years ago and had a revenue of $22,5 billion in 2017. Currently, it is the world’s second-largest after Caterpillar. People who know the brand might picture a huge bulldozer or a mining haulage truck and a conservative company. Yet, Komatsu spent the last decade infusing all of their machinery with new digital capabilities. Traditionally the company was sending service technicians on fixed intervals to the mining site to carry out preventive maintenance, exchange hydraulics, and other components that were thought to be at the end of their lifecycle. In some cases the exchange of the spare part was unnecessary, in other cases, the scheduled maintenance was too late and the machine broke down. And, unplanned downtime was causing delays and loss of revenue for the construction and mining projects.
To change that, Komatsu invested in building a telemetry system into their mining equipment. This enabled not just to automatically bill the working hours of rental machines, and automatically report project progress, but opened up a new opportunity. The data remotely collected from hundreds of sensors is fed into a predictive algorithm, which orders a field service just weeks before the machine is about to break down. This automated and real-time process saves millions on spare parts and field service logistics. It also prevents loss of business due to unscheduled downtime.
Predictive analytics — a.k.a. machine learning — allows companies to build intelligent automation in every business domain: be it predictive maintenance, inventory optimisation, personalised offers, but even self-driving cars and automatic check-out in a grocery store. It enabled many innovative companies to set themselves apart from the rest of the pack. These early success reports already put machine learning in the headlines.
The third form of a data product: intelligent applications
For setting the context, we will be looking at how data products helped to turn around the fortune of the Washington Post in an era when demand for traditional newspapers is declining.
Newspapers all over the world find themselves in a difficult quest for a profitable revenue model, both offline and online. In the past 10 years, most hundred-year-old publications disappeared or became a mere shadow of their former self. The San Diego Union-Tribune, once estimated to worth roughly $1 billion, was sold for less than $50 million. “A rock-bottom price” — as it was called by The Wall Street Journal — essentially a real estate deal.
In this context, it’s understandable to wonder why Jeff Bezos, the founder of Amazon, would buy the Washington Post back in 2013.
Within three years the paper had doubled its web traffic and became profitable — an impressive feat for a media company that struggled in the wake of the financial crisis. First of all, Bezos has brought masses of technological talent to the Post and they started to build data products in the form of web components. One such component is presenting different article headlines to different readers in the same time. It will figure out which one converts better with the different audiences. Another one monitors the reader’s scroll frequency and inactivity to re-engage with them at the moment when they’re most likely to look elsewhere. Yet another component is an AI-powered reporter-bot which covers niche news for small segments. It was used for the first time in the 2016 election in the USA, to cover every single congressional district election that was happening at the same time when the presidential election happened. According to Jarrod Dicker, VP of Commercial Technology and Development at that time: “that was virtually impossible with manpower”.
Soon, software-applications, and machinery will behave differently in the hands of different users, at different locations, and in different scenarios.
Many intelligent algorithms will collect behavioural, location, and preference data and will offer actions when needed, where it is needed. Fin-tech applications will offer travel insurance when you book an airplane ticket. They will offer a daily rate and stop charging you as soon as the GPS sensor in your phone detects that you have returned to your home country. Banks will offer a mortgage when you receive your first transfer after a raise. Coffee brands will provide you a voucher when you walk nearby to one of their shops.
Google Maps provides many small examples for the intelligent application of the future: it offers to tag frequently visited destinations and places them on shortcut buttons. It sends you push-notifications with traffic information, just before your usual commute in the morning. The app delights you with surprising functionality based on your behaviour: it offers options for sightseeing when it detects that you’re scanning the map while being far from home.
In the future, we will hear more often about hyper-relevant customer experiences. Apps will soon figure out what you need even before you articulate the wish.
Summarising data products
Data products are a result of the “productisation” of data, with users who consume that data and their experience at the center stage. Sometimes data products are intelligent components embedded into other products, thus users are not humans, but systems. Nevertheless, it is the humans who will ultimately benefit from this extra intelligence.
Data products are best served by a a data product manager. They will work backwards from the business objectives and expected outcomes. Suddenly the curated data sets are becoming easy to use and to explore. The overall consumer experience is becoming a central discussion point. Product managers are not just reacting to report requests anymore, but think of a consumer persona, a life-cycle, quality attributes, business value, key resources and activities. At last but not least they provide more context and understanding about how teams can contribute to the growth of the organisation.
This article first appeared on Linkedin and it is part of a series on scaling businesses with modern data analytics.
Comments