Modyo
Get Started
digital experience

Data Flywheel: Scaling a world-class data strategy

A data flywheel is a phenomenon by which the momentum of a product or process increases at an accelerating rate due to the strategic usage of data, and can be used to improve your customers' digital experience.

Kalle Bylin

Kalle Bylin

Share this:

Related to:

Business Leaders Frontend Architects Technology Leaders

Published 29 Jul, 2021 6 minutes of reading

Last year we shared a two-part series discussing the value of shared expertise as the first step towards a world-class data strategy. We argued that companies should start with culture and people, and let technology be a facilitator

Building upon that series, we want to share more ways in which you can extract value from your data, and one thing you can do right now is to set yourself up for fast experiments. Data has no value if we can’t act on it, and today more than ever it's important that we do it fast. 

Leaders in the digital world have been adopting agile techniques for several years to structure their work. These techniques allow teams to respond more quickly to new information and consequently reduce waste. Still, an often overlooked effect of these techniques is the potential to learn in real-time and continuously improve results. 

Once you start building more relevant customer segments, you have to be able to quickly customize your digital channels to each user. Heavy and slow systems will only hold you back. 

While your competitors hoard huge amounts of data to plan changes for the next year, you can start using the data you currently have and continuously improve the digital experience through feedback you receive today.  

To do this effectively, we need to understand the concept of a data flywheel.

What is a data flywheel?

A data flywheel is a phenomenon by which the momentum of a product or process increases at an accelerating rate due to the strategic usage of data. 

Let’s break this down with an example. One of the most common use cases for data flywheels are recommendation systems. Services like Netflix are famous for including “Recommended for you” sections where a series of specially selected movies or tv shows are shown to the user. In the beginning, a very simple heuristic or rule-based algorithm is used (e.g. show 10 most popular movies) and the recommendations are usually the same for all the users.

Once the company has collected enough data on what users like to watch, they can use this data to train a machine learning model (we will call this a prediction engine). The first version of any recommendation engine tends to be overly simplistic but the recommendations are now fully personalized for each user. This improves the user experience which results in higher retention and positive word-of-mouth that attracts new users to the service. A higher number of users produces more data that the company can then use to improve its prediction engine and this becomes a virtuous cycle:

circle that says more data, improve prediction engine, better user experience, and more users

The most interesting part of this process is that momentum tends to increase at an accelerating rate. This is one of the main reasons it is so difficult for competitors to catch up. We see this over and over again. As Google’s search engine becomes more powerful it attracts more users, which in turn allows it to make its search engine even better. In the meantime, other search engines like Bing, Yahoo or Yandex struggle to keep up. 

A data flywheel has the potential to generate incredible rewards for the innovative companies that leverage it first among its competitors. It is also one of the two main ways of producing a positive ROI on data or AI initiatives.

Data ROI 

The last decade brought much hype around data, machine learning and artificial intelligence, but many companies and industries have failed to produce results that live up to the initial expectations. Unfortunately, many companies jumped blindly into data and AI initiatives without a clear understanding on how they would actually generate value. There are two main ways for companies to achieve a return on investment with data and AI:

  • Critical capabilities: Companies that focus on building the skills, resources and culture necessary to leverage data in the long term will be more successful at converting the crude oil of data into tangible business value.
  • Data dominance: This corresponds to the example shared earlier where a company is successful in activating a virtuous cycle of data within its industry. The constant flow of data drives faster improvements to the product day by day and makes it more difficult for other companies to compete.

These two concepts usually work together. Companies that have been able to achieve data dominance have usually done so by cultivating their critical capabilities.

Going back to the topic of this article, data dominance often leads people to think about the algorithmic data flywheel shared above. Data is used to train a machine learning model that becomes more powerful when trained with more data. 

This is only one of many ways in which companies can leverage the power of a data flywheel and is unfortunately usually very limited in its scope. Imagine a scenario where the prediction engine in the diagram above is replaced by faster learning throughout the company in general.

circle that says more data, faster learning, better product/ business functions/ company processes, and more customers/lower costsThis means that you don’t have to be the best or the most powerful today, as long as you are faster than your competitors at learning and incorporating new knowledge into improving your product, services, and internal processes. In the long-term you will be better positioned to outrun your competition.

The power of experimentation

“The real measure of success is the number of experiments that can be crowded into twenty-four hours”.

The quote above has been attributed to Thomas A. Edison, but has been backed by senior leadership at companies like Google, AirBnb, Netflix and Amazon. Jeff Bezos has famously stated: “Our success at Amazon is a function of how many experiments we do per year, per month, per week, per day.”

Former Amazon director and cofounder of Intuit, Scott Cook, has shared that former Yahoo executives attribute Google’s success over Yahoo to their native ability to run experiments at scale:

“[Google] just outran us. We tried management, all the stuff that management did, but we didn’t have that experimentation engine”.

Scott Cook started to focus on experimentation after noticing that many of the decisions made by him and his team were really not good. He decided to spend some time studying Toyota in order to understand how a loom maker could decide to enter the car business at a time when there were already multiple well-established companies producing cars as their core business and still become better than them.

In collaboration with several Harvard professors, they discovered that Toyota “runs itself as a massive series of experiments”. This allowed the company to learn faster than its competitors and move from underdog to leader in the car industry.

The power of experimentation lies in its ability to counteract our poor decision-making skills. Experimentation experts at Microsoft, Google, Netflix, Slack and others, have found that most teams see experiment success rates of 10-33%, with most of them being on the lower end. In other words, we can expect 70-90% of our work to be thrown away due to poor or even negative results.

"Most who have run controlled experiments in customer-facing websites and applications have experienced this humbling reality: we are poor at assessing the value of ideas".

person looking through microscope

Conclusion

Contrary to what we like to think, we are not very good at assessing the value of new ideas. Fortunately, we can use experimentation to quickly learn from both success and failure. Doing this effectively and at scale allows companies to generate a powerful data flywheel at its core where learning happens quickly and naturally.

You should also be smart about what tools you choose to work with so that you can extract value from your data as fast as possible.Try to find tools that allow you to integrate the data you already generate and then make it easy for everyone in the company to run experiments and act on that data. 

Customers are people with real needs, personalities, pains and passions. It is up to us to figure out who they are and how our work fits into their lives. Numbers are just numbers, it’s what we do with those numbers that really matters. Context and the ability to act are critical as well. Data alone is simply not enough to make better decisions. Done right, you can outrun your competitors by learning faster.

At Modyo, we're here to help you build better digital products and build better experiences for your customers. Through our platform, we can help you add more value to your products and services and achieve better business results. If using data to build better digital products is something you're interested in, come talk to us.

Photo by Przemyslaw Marczynski on Unsplash.

By

Kalle Bylin

Kalle Bylin

Other posts
For you to explore, read, and learn.