Order some 21st century management with your big data lunchMonday, Feb 27th, 2017
Why is it that every company today wants to have an Hadoop-enabled data lake, large-scale data pipelines in Spark, and artificial intelligence powered by Tensorflow? Yet, if you talk about how they are going to build and use all these state-of-the-art technologies, they come back with huge monstrosities of requirement and design documents, annual IT roadmaps & budgets and 18 month release schedules. The impact this dichotomy has on big data initiatives, is significant. Budgets spiral out of control. Deadlines fly by so fast, they create a doppler effect. And the dreaded "that's not what we need" when you finally ship to the client. That sounds a lot like the following quote:
The same companies that brought us the 21st century technologies, also adopted a completely different way of management. The famous Netflix culture document, the Spotify engineering model, and How Google Works are just a few examples. But how do we, mere mortals, adopt these principles in our organisations? Barry O'Reilly, the co-author of Lean Enterprise, organised a workshop in Berlin to help us with that. In this blog, I share some of my learnings from the workshop and I apply it to our own field of data analytics and big data. The slogan of Lean Enterprise is "Think big. Learn fast. Start now". Let's dive in.
Think big - Principle of Mission
18th century management theories tell us to command and control your employees to run an efficient factory. I've talked about that before. That principle simply does not hold in data analytics. What we're doing is too complex. Instead of trying to control and manage your smartest workers, it is your responsibility as a leader to give them a clear mission. Once the mission is clear, your next task is to step out of the way. Give them the autonomy to achieve that mission on their own.
For data analytics teams, defining a mission can be straightforward: "Reduce customer churn by analysing behavioural data". Or "extend the lifetime of a machine by optimising its maintenance plan". The team could decide to build a streaming big data analytics solution to do predictive analytics on day one, like you see with many large data analytics initiatives. But more likely, the team will start by talking to customer service, or by visualising a few machine logs to find some obvious patterns. Once the low-hanging fruit is captured, the team can move on to more advanced modeling and big data processing. After a while, it can start building reliable dataflows in the cloud. But it moves at its own pace, and it aims to get closer to the mission at each iteration.
Learn fast - Measure outcome, not output
Story time: the BI department of a large Belgian company wanted to 'adopt agile' so they religiously started measuring team velocity and they built a benchmark across teams to track teams who were 'underperforming'. "What do you mean, your team only delivered 80 story points this sprint? All the other teams delivered 100 story points." You can guess the result. Teams inflated and tweaked their story points, so they kept a consistent, high-performing, velocity across sprints. This is what Barry called a watermelon metric. Green on the outside, but red on the inside.
Who cares about team velocity and story points anyway? How are they helping the business win? Instead of measuring output, we should measure (business) outcomes. The outcome is directly linked to your mission, and the outcome metric should be defined and agreed upon before the team starts working on the mission. Eg, we define customer churn as "customers who cancel 1 or more subscriptions within the next 4 weeks and don't return in at least 6 months." If you can improve an outcome metric by sending a simple message to customer service, that is great. If you can improve a customer metric by refactoring a bunch of inefficient code, that is great as well. If you can do it by writing a streaming big data analytics platform in the cloud, that is the same level of greatness.
Start now - innovation portfolio
Where do you begin? Within a company, at any given time, there are 100 appealing data analytics ideas and you can't make a complete ROI case for every single one before you start because you would spend the next 6-12 months in planning and analysis. By which time, the opportunity might be gone again.
One super useful exercise we did in the workshop, was map out all the projects and initiatives going on in a company in an innovation portfolio. The idea is that initiatives go through several different phases: Explore, Exploit, Sustain, Retire. Each phase needs another approach to management. In the explore phase, there is still a lot of uncertainty. And you focus on doing experiments and validating initial hypotheses. In exploit, you already have a better understanding of the problem and you can start making forecasts. In sustain, you have your successful big data initiatives, either saving costs or generating revenue for the company on a daily basis. In retire, you realise that the ROI has been decreasing for a while, and you consciously pull the plug to focus the team on other missions.
Once you've mapped all initiatives in an innovation portfolio, things become a lot clearer, even for a small company such as Data Minded. And you can make strategic decisions on where to invest and which initiatives to kill. Barry advises to apply the 70/20/10 rule. 70% of your effort should go in the 'sustain' initiatives, that are generating revenues now (Horizon 1). 20% Should be in 'exploit', because some of them will start generating revenue in 12 to 36 months (Horizon 2). And finally, 10% should be invested in 'explore', for the long-term revenues (36 to 72 months).
Obviously, I'm only scratching the surface, and I can recommend everyone to dive deeper into the subject by watching talks, reading the book or joining a workshop. When I talk about these principles with clients, they often push back because "we are not netflix". I think that excuse doesn't hold anymore in 2017. We're using all the technologies of the cool kids. And most importantly, these cool kids are actively competing with Belgian companies. Netflix is eating away at Proximus and Telenet, Uber is challenging the taxi industry, Airbnb the hotel industry, Alibaba is pushing the port of Antwerp to become digital, banks are facing drastic restructuring, and our media companies like de Persgroep and Mediahuis are competing with the entire internet for our attention. The avalanche has already started. It is too late for the pebbles to vote.