Formula Racing & Data Science: Part 2

Recap

In Part 1, we discussed how team dynamics separate a successful team from others in Formula Racing, in spite of all teams having the same set of rules defined by the FIA. Then, in Data Science, we saw how the same set of reliable tools are freely available for all organisations. But many organisations fail to concoct a successful Data Science formula, in spite of having free access to common pot . 'Just another IT project' approach to Data Science is one of the major reasons for this failure.

In this article, first we describe the anatomical similarity between Formula Racing and Data Science team structure. Later, we try to list typical team structures in Data Science that are likely to fail.

Similarity between Formula Racing and Data Science Team

Role 1 : DEFINE Target and 'Idea-to-Implement' engagement

Formula Racing - Driver defines what the car needs to meet (or exceed) their team's race objectives. The Driver also provides direct feedback on how the car drives during testing (before the race) and in the actual race.

Data Science - The Business define what the organisation needs to meet (or exceed) business targets. The Business also provide direct feedback on results of Data Science models during development & testing (before go-live) and from 'live' business operations (after ‘go-live’).

Role 2 : REFINE Target, Articulate Requirements, Provide Pragmatic Steer

Formula Racing: Team Principal articulates driver-defined needs into tangible building blocks, in the language understood by Engineers and Pit Crew. Team Principal also provides a strong steer to all team activities for converging in the direction of the best possible car.

Data Science: Data Science Lead articulates business-defined needs into granular requirements in the language understood by Maths Experts and Tech Experts. He/She is provides a strong steer to all team activities for developing a pragmatic and adaptable predictive model for the business.

Role 3: DESIGN Target Building Blocks 

Formula Racing: Engineers design and develop car components using building blocks articulated by the Team Principal. A component-design transforms 'driver-defined needs' into a blueprint to manufacture the physical car. Intricate details of the component design are explained in a simple terms for seamless communication within the team.

Data Science: Maths Experts design and develop mathematical models using business requirements articulated by the Data Science Lead. Each model transforms business-defined needs into a blueprint to develop physical code for the data science model. Intricate details of the maths models are explained in simple terms for seamless communication with the team.

Role 4: DEVELOP & DEPLOY Target Building Blocks

Formula Racing: Pit Crew assemble the component, designed and developed by Engineers, to build the physical car before the race and ensure components are replaced on the car, fast and effectively, during pit-stops in a race.

Data Science: Tech Experts assemble models, developed by Maths Experts, using techniques such as data engineering, code creation and deployment pipelines to get the predictive models working and pumping out insights.

When the above roles work synchronously, chances of success are very high whether it is Formula racing or Data Science !!!

In the next section, we will see 4 typical structures that can lead to Data Sciences teams to fail in achieving success.

Typical structures driven by ‘just another IT project' approach

1. No Data Science Lead

• 
  
• 
  
• 
  
• Someone in the Business gets excited about Data Science and strings together a bunch of Maths Experts and Tech Experts with ‘latest’ technology expertise
• ‘Thinking-aloud’ statements from the Business become gospel for requirements
• Maths Experts develop slick models using 'gospel requirements' and Tech Experts perform all the data engineering, coding and deployment efficiently for the models

Result

Data Science: If Maths and/or Tech Experts are short on domain expertise or on-the-ground business experience, output of this exercise can provide skewed results – either too good to be true or too bad to spend any more time on the exercise. 

Formula Race Equivalent

Hi-Spec Car with Dodgy Steering will launch spectacularly and have a gentle crash when the steering finally gives up

2. No Data Science Lead, No Maths Experts

• Someone in the Business reads a marketing use case on how another organisation reaped significant benefits using Data Science 'technology'
• A bunch of Tech Experts is herded in a room and the marketing use case is narrated as the requirement   
• Tech Experts develop code to exactly reflect what the ‘use case’ describes

Result

Data Science: High likelihood that results from rosy 'marketing use case' will fail to materialise from this data science exercise. Also, it may give 'Data Science' a bad name in the organisation and become a very unpopular topic with the Business for any future engagement.

Formula Race Equivalent

Prototype car with dodgy steering will manage to start the race and crash spectacularly as soon as the steering gives up

3. No Data Science Lead, No Tech Experts

• 
  
• 
  
• The CIO or senior IT managers, in an attempt to be ‘thought leaders', will over-promise benefits of Data Science to the Business
• Maths experts, with basic technical skills, will be fronted to the Business.
• The Business will use blue-sky canvas to 'define' their objectives. Often such objectives need to qualified and refined to eliminate multiple interpretations
• Maths Experts will develop a 'data science' model using their interpretation. IT execs will provide a 'positive' spin to the expected output using their own interpretation to impress the Business
• Business will then look for results based on their own interpretation

Result

Data Science: Success depends on the lucky chance that interpretations of the Business, IT Managers and Maths Experts are same !! In case of an unlucky failure :-) , IT will blame the Business for no clarity and Business will lose faith in IT. The poor Maths Experts will end up getting squashed between IT (to deliver something fast) and the Business (to deliver something real) 

Formula Race Equivalent

Hi-Spec car with dodgy steering and little fuel, moving fast in some direction, stalling suddenly when the fuel runs out. 

4. Wind Tunnel Test Car

• This is completely different !!! Here, the Business or IT senior execs will be approached by a Consulting Firm, that has ‘supposedly’ delivered massive data-science driven benefits to other similar organisations
• Senior consulting members start their schmoozing campaign with decision makers in IT and Business teams
•  While the seniors are schmoozing, middle and junior level consultants will be super busy developing the data science platform and churning out reams of documents and power points to show progress. Client teams will either be over-awed or plain exhausted by the sheer 'bulk of deliverable'
• When client budget has dried up or the billing cycle is nearing the end, there will be one grand display to present the ‘final’ Data Science deliverable

Results 

Data Science: Extremely high likelihood that there is NO tangible ready-to-use output for the Business. The 'deliverable' is probably a poorly assembled data science toolkit that worked well when the consultants were on-site and stopped working as soon as they left.

Formula Race Equivalent

Brilliant car performance. However, it is in a simulated wind tunnel and not on the race track :-)

End Notes

1. ‘Wind Tunnel Cars’ are lethal. If not controlled, big Consulting firms are potential 'dementors' from Harry Potter. Their engagement approach is capable of sucking out the soul of an organisation. The Business and IT teams become peripheral, with the consulting firms taking the centre stage. The working ethos of an organisation goes for a complete toss
2.  Avoid Suffocating Good Ideas. Managers (with zero hands-on experience) driving a data science initiative is similar to microwave-cooking of risotto in two minutes. Their tendency to 'push' for timelines is a very reliable recipe to fail. A genuinely good idea can suffocate and die purely as a consequence of superficial understanding of data science and lack of business judgement.
3. Anecdotes are for Sales and not for Delivery. Success stories from sales people (or from the internet) often exclude crucial details that are organisation-specific and cannot be generalised. So blindly trying to replicate 'success stories from other organisations' is risky. 
4. Data is beautiful Data needs to be conditioned for ANY data science exercise. Be careful not to sink in the over-hyped ‘Data Cleansing’ quicksand. A distinct lack of domain knowledge by 'data cleansing' champions often leads to resource-sucking activities such as data architecture, implementing profiling tools, defining canonical, semantics, blah, blah blah....!!

We need to be clear that existing data, even if not the cleanest, was designed to address business needs at some time in the past. And it successfully did so. Data Science and Analytics came later.

So stop the whining and move on and rather help structuring application data in for future!!!

Where to Start

• Start in an iterative mode using small non-ambiguous uses cases. Continually experiment to fail fast and fix faster
• Leverage publicly available reliable code-sets in the data science domain. The first-hand experience of using real technology for your enterprise data will significantly accelerate the data science maturity process in your organisation
• Most importantly, get a good grip on your existing enterprise data. This is your biggest hidden asset. The more you understand your data, the more you will see opportunities manifest themselves!!