Essentials for AI success: how to use the data flywheel within an integrated artificial intelligence strategy

These days we hear about the disruptive potential of Artificial Intelligence (AI) everywhere. From self-driving cars and in-home robots, to disease detection and tackling climate change, breakthroughs with AI are happening across multiple industries. At the same time executives are learning that to make the necessary transformation to create these impacts at scale a robust AI strategy is needed that is integrated within their organisation's overall strategic approach.

There are numerous cautionary tales in the market: organisations that have invested heavily in AI and data science only to be left with multiple proof of concepts (PoCs) that fail to make it to ‘production’. There may be a 'data lake' floating somewhere, but proper value or ROI shows no signs of being delivered anytime soon. The once prevalent motivation and ambition for data science and AI is evaporating rapidly.

Then there are the success stories: those companies that have integrated AI as a core component of their strategy and have been able to transform business models, unlock new sources of revenue through additional market creation, and build core AI skills in employees across the organisation that will likely provide a pioneering position for many years to come.

In this article we'll look at how a concept called the data flywheel can be used to build an integrated AI strategy. We'll explore examples of how this flywheel can be put to use in organisations where the focus is on delivering products or services at scale and I’ll highlight how we’ve used the approach successfully at Quby. We'll end with three essential ingredients for AI success: acquiring labelled data, using multidisciplinary teams and the role of the CDO for AI leadership.

Introducing the data flywheel 

Quby is an Amsterdam based tech company offering home services technologies with a focus on reducing energy wastage. Before starting our journey with AI we were widely known for creating Toon, a white labelled in-home display and smart thermostat sold by utilities across Europe. In the last four years we have transitioned from a predominately hardware based business to one with data and AI at the core. To make this transformation we have used the data flywheel concept.

The data flywheel, sometimes known as the virtuous cycle of AI.

The idea behind the flywheel is simple. An organisation begins with an existing product that has a number of users. These users generate data. This could be, for example, IoT data from in home hardware or maybe interaction data from a mobile app. The data can be used to begin to train some algorithms. In the first instance these algorithms will likely not be great, but at some point they will allow for an improvement in the product experience by enabling a new feature or enhancing an existing one.

The feature will work better for existing users and they will start to use the product more. Alternatively the feature could appeal to a new user segment entirely and this group will be encouraged to use the product. Either way, this increased usage will generate more unique data, allowing the business to develop better quality algorithms, improving the product further and drawing in new users. As the cycle continues the flywheel builds momentum.

The data flywheel underpins the success of companies across multiple sectors

This concept is no big secret. In fact such a flywheel underpins the business model of many product focussed tech companies. For example Netflix uses algorithms built on user data to make recommendations to individuals for what to watch, but also to determine exactly which series and movies they should commission in the first place. 

Google uses the data flywheel throughout it’s business. When moving into the in-home voice assistant domain with Google Home, it has been able to harness voice data gathered through the Android mobile OS to give it a massive head start. This has been particularly valuable with internationalisation. Google Home currently offers voice recognition in 19 languages, in comparison to Amazon Alexa’s seven.

Is your company currently, or is seeking to, offer AI-powered products or services? Then the data flywheel concept is applicable to your business.

Let's explore how this worked at Quby.

Getting the flywheel spinning for early AI success

At Quby, back in 2016 we had invented and successfully distributed our Toon hardware product to hundreds of thousands of homes across Europe. As we started our AI journey we were able to use this rich source of Internet-of-things (IoT) data to train algorithms to detect usage behaviour of in-home appliances. In 2017 our first AI powered service was launched. With our ‘Waste Checker’ we were able to detect inefficient usage of appliances, from the washing machine and dishwasher to the central heating system and shower. Once potential waste was detected we offered personalised advice on which actions to take. With a few loops around the flywheel we were able to offer unique insights with market leading accuracy (to learn more see the article: ‘Empowering households to reduce energy waste by making data science end-user friendly’).

Using the data flywheel to enter new markets

From our early beginnings we have naturally continued to use AI to improve the experience for Toon users. A recent example is personalised thermostat programming that helps households ensure that their home is only heated when necessary.

The flywheel approach also allowed Quby to make a much larger step: accessing an untapped market segment. By building upon the algorithms originally developed for the Waste Checker, we can now offer services to homes with a smart meter. By 2023 more than 200 million European homes will have a smart meter. We will no longer require the installation of any additional consumer hardware such as a smart thermostat.

Whilst already prevalent in many homes, smart meters offer very limited data, typically only a data point every 15 minutes to an hour. In comparison our smart thermostat accesses high resolution data with consumption measurements on a second by second basis. We have pioneered novel 'super-resolution' techniques to overcome the limitations of smart meter data and offer best-in-class performance. Our patent pending algorithms are trained and the results enhanced by using high resolution 'labelled' data coming from our thermostat users.

Quby's bill breakdown service for smart meter customers. Utilising our large database of high-resolution data, we apply advanced techniques to offer a more personalised, more dynamic and more accurate bill breakdown service.

Since the end of 2019 we have offered energy bill breakdown and alerting services. In the Netherlands these services are available for free to over 1 million customers of the utility Eneco. Entering this market would not have been possible for Quby without consciously building around a flywheel concept and integrating this with the overall business strategy. Along the way, we’ve found three ingredients to be essential for AI success.

Three essential ingredients for AI success

1. Make no compromises in acquiring labelled data

Labelled data, which is used to train machine learning algorithms, is the key to unlocking product improvements and new opportunities. However it is often costly for organisations to acquire and companies need to be prepared to invest, particularly in the early stages of getting a flywheel up and running.

Labelled data is so valuable that sometimes it is necessary to create features which, whilst offering only limited user value, allow for labelled data to be collected at scale. At Quby a now legendary pilot project called 'De Nachtploeg' (The Nightshift) was conducted in which smart plugs were distributed to 150 volunteers. These smart plugs were connected to the three major white good appliances in the volunteer’s home: the washing machine, dishwasher and dryer. We encouraged participants to engage in manual demand response, changing the timing of their appliance usage in response to various price signals. The smart plugs were in place to confirm whether the action had taken place or not. One output of the pilot was that we found that it was not commercially interesting for us to introduce such a demand response product in the market at that time. The other, and ultimately far more important, output was that the data collected from these homes told us exactly when individual appliances were used. We could use this to label our results and test the accuracy of our appliance detection algorithms which used the home's collective metering data as input. The data from these intrepid innovators allowed Quby to make its first steps in AI and for our flywheel to really start spinning.

2. Create multidisciplinary teams with a focus on the end-user

As AI becomes more mature within a business, the organisational setup will need to evolve. At Quby we’ve moved from individuals that could manage single parts of the flywheel into teams that together can encompass all four components: data, algorithms, experience and users.

We’ve built multidisciplinary development teams with an end user focussed mindset called Value Streams. These are formed from T-shaped individuals covering data science and engineering, UX, development and marketing. This requires discipline and a product mindset throughout. For example, a data scientist should not think their job is complete when they’ve created an algorithm with high accuracy. They need to be engaged until the end user of the AI product is satisfied. 

3. Invest in AI leadership, starting at the board level

Two decades ago, at the dawn of the internet age, knowledge and understanding of technology became essential. Initially, the chief technology officer (CTO) led this initiative in the boardroom. Today, every executive needs to be tech savvy.

Artificial intelligence will have a similar transformational effect in the coming years. Executives from all backgrounds will need to become comfortable not only with technology, but also with data and AI. They need to understand how AI will transform their industry and what the state of play is, right now. A chief data officer (CDO) or chief AI officer can lead this transformation in the boardroom today. The CDO should create and implement a robust AI strategy that integrates with the company's strategic approach, empowering company-wide success in the era of Artificial Intelligence

-----

So, will your organisation become a cautionary tale or a success story in the next decade of AI? That may depend on how you make use of the power of the data flywheel. Choose wisely…

About the author:

Dr. Stephen Galsworthy is a data driven executive and advisor who loves to build teams and scale disruptive products which address significant challenges. With an analytical background, including a Master’s degree and Ph.D. in Mathematics from Oxford University, he has been leading product and data teams since 2011. 

Currently Stephen is Chief Data & Product Strategy Officer at Quby, a leading company offering data driven home services technology and known for creating the in-home display and smart thermostat Toon. In this role, he is responsible for the creation of value from data and Quby’s overall product strategy to enable commodity suppliers such as utilities, banks and insurance companies to play a dominant role in the home services domain.