It was a very interesting event to discuss implementation of Data Science into organizations. Data Science brings huge opportunities, but making it work in established organizations is far from obvious.
I like to share some thoughts on implementing a data science strategy. Three elements are critical in making data science work: the culture, the strategy and the talent. Talent is the most important but is hard to get without having the right culture, no matter how much you pay. But culture is even harder to change. How to break this causality loop?
In my opinion, the CEO should hire a very senior data scientist that will be pivotal for the transformation. I call it the TDS – Transformational Data Scientist. Instead of designing big plans, start with small projects that could have, what I call, “transformational impact”. They don’t need to be complex since they have to prove is what can be done. I call them, the tutorial projects – modifying a culture is like (re)educating a person.
This person, however, has to be a communicator as he will be responsible to engage other departments and create evidence that will help the CEO to test and find support for further incursions. Give him as much autonomy as possible and let him find his way.
Second step, a proper budget should be allocated for the TDS to create a data science team. If the culture is not there, don’t try to apply data science to all departments. The outcome should be tested and make sure actionable insights come out of it.
Now comes the really hard and critical part: deploy the projec. Lots of barriers will be issued as the TDS try to put the model in production. The support from the CEO is key. The TDS may not coordinate the implementation, but he should follow it close. Some KPIs has to be defined and a time horizon set to evaluate the effectiveness of the project. Make sure you choose the right KPIs: profit, costs or savings may not be the best metrics (or short-term ones). Look at metrics that gauge the customer satisfaction, quality of service, customer engagement (no need for NPS or surveys – use indirect ones).
After this period, the first step will be up to the CEO to decide, based on evidences, what measures should be taken: restructure the team, fire some workers, hire new ones, outsource, change infrastructure. In all these decisions the TDS should be involved. Inability to take action is not an option and gives the wrong message to the DS team – that nothing could be changed and only “cosmetic data science”, not transformative data science, is in place.
This will motivate the DS team and will change the culture of the organization. Sometimes small measures are sufficient, others shutdown of entire departments may be required. Regarding data, it’s important that an open attitude is in place. Recent solutions are capable to deal with security and privacy, like Hadoop and Spark and other cloud based ML services and storage.
Some other thoughts
Data. Data are coming in any shapes and sizes. Don’t rely only on traditional RDBMS systems. Prefer flexible data lakes easily integrated with NoSQL formats than complex and expensive data warehouses. Any user interaction should be an opportunity to collect new data points. You never know what they can be used for in the future, so flexibility is key.
From exact to probabilistic thinking. Everything is becoming programmatic and we are evolving towards cognitive computation – which basically means, machines are not operating under fixed rules, but every evolving and learning.
From rules to cognition Better to have an approximative answer in milliseconds than an exact one in 1h – or none. Companies should embrace this new paradigm We are experiencing a paradigm shift, from a purely Deductive to an Inductive scheme
From all speakers it was clear that data science is transformative force and great times lies ahead, but it’s important to be aware of the challenges.