Formula Racing & Data Science: Part 1

Background

‘What should a data science team look like to deliver consistently tangible business value?’ is a question that always evaded a direct answer for the last couple of years. While watching the Italian F1 on Sunday, suddenly the same question presented itself in my mind; and accompanying it was the answer. The laser-sharp focus, of different complementing roles in each F1 team, to ensure their driver’s success seemed to answer the question.

Wisely Experiment, Fail Fast and Fix Faster

Based on the 'formula' (and hence the term Formula racing) governed by FIA, F1 teams always have a foundation car before the start of a season.

Then, prior to every race, the process of continual experimentation starts until the car is ready for that race. Factors such as track temperature, rain forecast, tyre wear rate, down-force on bends and straights, historical results etc. are assessed and the car is fine-tuned to make it fit for the track. However, it is simply impossible to control, or even anticipate, all factors.

During the race, all planned and unplanned challenges are smoothly addressed by the well-rehearsed switch and fix capability.

After the race, continual experimentation starts again for the next race.

In general, I would call this as ‘Wisely experiment, fail fast and fix faster’ approach.

I believe this approach aligns brilliantly with Data Science as well. There is never a perfect mathematical model to predict the desired business outcome.

Prior to deploying a data science model in production, you do continual experimentation and train the model until predictions are satisfactorily close to the desired outcome.

During production run, it invariably needs minor tweaking to suit real-life data when model needs a bit of switch & fix treatment for data types, volumes, etc.

After analysis of prediction results from production, the continual experimentation starts again on the model for the next feature or change in direction from the business.

It is important to note that Data Science is very different from typical IT projects such as ERP/COTS implementation or Application Development. This was the 'golden-egg hen' that made millions (or billions?) for IT Services, Product & Consulting firms; firstly by spending years to implement ERP and develop software and secondly, by spending more years to maintain the same!! Also, I would not be a big supporter of the argument that the recent ‘Agile – DevOps’ stuff delivers IT benefits in compressed timescales and budgets!!

What is the Formula 

So, here is the thing.

In F1, all teams use the same basic ‘formula’  i.e. rules governed by FIA. However, successful teams (such as Scuderia Ferrari, Mercedes Racing with budgets c.$410 million) have evolved their team structure that give them the competitive edge. And this evolved through trial-and-error over many many years, for different roles to operate like clockwork. Newer teams (such as Torro Rosso, Haas with budgets c. $100 million) use the tried-and-tested structures of successful teams and adapt, as per their resource availability. (2021 will be interesting when the budgets are capped to $175 million !!)

Each team role - Team Principal, Engineers, Pit Crew & The Driver - has a specific function which is directly linked to the common objective of maximising driver success on the track.

The point is that team structures evolved to meet F1's specific needs and constraints rather than just cloning team structures from 'another similar motor sport'.

In Data Science field as well, all organisations have a common pot with same ingredients for everyone to use. And the ingredients are: Maths Models, Raw Algorithms, Machine Learning Platforms such as Jupyter, programming languages such as Python, R, Julia, SQL, applications and databases, cloud technology provider, Excel (J) and so on. In addition, there is a free goldmine of amazingly high-quality expertise and resources on the web.

However, many organisations have struggled to create their own successful formula using the common pot of ingredients.

In a large majority of such cases, organisations approach their data science initiatives like another ‘IT project’ and completely overlook the importance of the Team composition.

For higher changes of extracting actionable insights and tangible tangible benefits from Data Science, organisations need to start approaching Data Science as a Product and not just another IT project. This will immensely help them build teams that can truly focus on the Business objectives rather than on project objectives.

In Part 2 of this article, I shall elaborate on:

• Anatomical Similarity between F1 & Data Science Teams
• Typical 'another-IT-Project-mentality-driven' Data Science teams in many organisations today