How We Built a World-class Data Science Team @Airtel!
When I joined Airtel about a year back to build the data science team from scratch, I was reminded by a fact that over 75% data science projects in Fortune 500 organizations fail to produce results or are cancelled! The reasons vary, but leading cause is the inability to build Tier 1 data teams that are motivated by a strong vision and rally around it, aren’t afraid to hire the best, are keen on building a culture that promotes trust, transparency and collaboration, have a strong bias for action and coach and mentor each other to succeed.
Given failure is a common problem, I will start this article from the fundamental premise, starting with the definition of what an organization is.
The definition drives straight to the point — people in an organization do not merely “develop software”, “use Excel to analyze data” or “create PowerPoint presentations”, they organize themselves “for a purpose”.
But first, let’s introduce ourselves! Airtel is world’s third largest telco with ~400 million subscribers in 16 countries and provides mobile telecom, fixed line and WiFi broadband, voice, DTH, services to both consumers and businesses. It is building a growing portfolio of consumer internet and fintech properties; Airtel Payments Bank, Airtel TV, Wynk Music etc.
Today we generate 100’s of trillions of records annually from calls, network, apps, IoT devices, phones, GPS and more. Very few entities in India, or globally, come even within the (lower) order of magnitude of this volume of data.
When I first arrived at Airtel about a year back, the situation was very similar to what one would typically expect in an organization where data science has not taken roots. We had a small team of analysts engaged with the business units such as marketing, and customer experience and a couple of data scientists working on a few projects.
It was also a tall task, given Airtel had very little visibility within the data science community. Although we had more data than most companies in India, we were not a destination for aspiring data scientists!
So how did we do it. Here is my typical prescription that I have followed at previous instances of similar situations, so I stuck to my playbook of 6 tenets.
The 6 tenets
Tenet 1: Understand, articulate and internalize the purpose of the data group and set the vision. Vision is like the destination sign on a bus: people are unlikely to board the bus if they do not know where it is going.
Our vision was simple: “Create India’s most sophisticated mobility data team to lead digital and data products driven “Airtel 3.0” transformation”
The above sentence embodies two key purposes that our group had organized for
- To be India’s most sophisticated data driven team in mobility
- To lead data driven transformation for Airtel
This one-line statement helps us to keep a laser focus on what we do everyday!
Tenet 2: Hire the best leaders and provide them with freedom, flexibility and support
Here is a simple truth most people in the industry would avoid talking about. Too many data organizations fail for one reason; they are headed by mediocre leaders. When you hire a B-grader to lead, they soon hire C’s, who in turn hire D’s and the whole org goes into a death spiral quickly.
Recognize that just hiring a top-talent to lead the org is not sufficient; they need the following ingredients for success:
- Freedom to hire the best
- Flexibility in terms of technology, locations, budgets, positions/ titles and other operational aspects
Hiring data scientist is a painstaking and difficult, given the demand-supply mismatch and expanding number of AI startups that we continue to compete with. Strong HR, Finance and Operations Support, often overlooked and underestimated, are crucial for building a world-class data team.
I often compare data organizations to aircraft carrier battle groups; they need large number of other ships to protect (the destroyers and frigates), nourish (the refuelling ships) and nurture (supply ships) for successful missions. Likewise, creation of a strong data group is a product of support and nourishment from the rest of the organization.
Tenet 3:Build a culture that promotes trust, transparency and collaboration. This is probably the hardest part. Once top talents have been hired, giving them the freedom and flexibility to operate and achieve goals, and enabling them to work collaboratively with the rest of the organization to achieve business objectives, is crucial.
I ask a trick question while interviewing candidates, “what do we, as data scientists, deliver to our customers”? I get a varied set of answers; most common ones are, “analytics”, “recommendations”, “insights” etc. All these answers are correct to some extent. However,
The true deliverable for all analytics professionals is “trust”
Our (internal) clients are typically not deeply weeded into the intricacies of our data, and most are not familiar with the analytics, tools and techniques we use to deliver the insights and recommendation. Despite this, if we want them to fully embrace data-driven decisioning, they need to implicitly trust us.
How does one build trust? that’s next, tenet #4 — strong bias for actions and closing loops.
Tenet 4: Create a strong bias for action with focus on end-goals; this is crucial for measuring success and progress. “Closing the loop” is also important for establishing trust with stakeholders in an organization’s early days into data analytics venture.
The single biggest factor behind building trust in an organization that’s embracing data analytics can almost be broken down as a process.
Start with an easy-to-deliver project with low complexity
In the early days, achieving success in a small project is far more important than demonstrating analytical prowess in a long-term, complex project which may get delayed or cancelled, or whose recommendations may be to hard to comprehend or implement. This is highly counter-intuitive, but will likely save the data org’s reputation, sanity and maybe even, existence in the long run.
It is also absolutely important to close the loop with your stakeholders. From time to time, the insights we develop may not exactly answer the business questions. Following up on a stakeholder’s subsequent queries, while may sound like a trivial issue, is often left behind as we focus on the next big project. One common reason data projects do not get traction within large organizations is largely due to lack of follow-ups. Unlike most corporate functions, data science is iterative, involving hypothesis formation, preliminary data analysis and sharing with stakeholders, refinement (of the goal, hypothesis, methods — perhaps multiple times) leading to the final analysis and recommendations. All of these need constant communication and closing the loop.
Tenet 5:. Data analytics is still a bit like black art, in the sense that the starting point is often an intuition used during the hypothesis formation phase. The chance of success improve when all teams involved have ownership of plans, goals and decisions-made, i.e., when the decisions made are decentralized and not top-down.
Complexity of today’s business world is enormous: an average organization today has hundreds of processes, multiple data sources (internal and external) and numerous technology options to model and analyze data — each of which are evolving continuously.
It is impossible for one central figure to decide or plan for all the program details, technology choices, model / algorithm selection or monitor the outcome on a continuous basis. A decentralized structure usually offers a better chance of success by combining the best of talents and technology to solve the problem at bay.
In a decentralized structure, independent teams fulfil different functions at different timelines, and yet cannot function without each other. Making this distinction is important for keeping the focus within the group. At Airtel, we have done a decent job at this by creating separate groups for data science (algorithms and models), data engineering (platform, data analysis and visualization), data product managers, business etc. The smaller groups within the decentralized structure, besides providing functional expertise and opportunities for learning, allow for career growth opportunities for employees.
Tenet 6:. Finally, a data analytics team is like a football team — coaching, mentoring and developing people for the future builds long-term relationships and trust and produces better results.
A world class org should inspire people to take ownership and mentor them for achieving excellence — rest will organically follow.
Epilogue
Creating a world-class data organization that commands trust and respect, and has measurable impact is extremely difficult — there are less than a handful in every industry (except in the tech industry). If this article helps even one data group make small improvements, I will consider my efforts worth it.
How can you be a part of our journey?
An industry disrupted by Reliance Jio leading to a 3-player market for world’s largest subscriber base, plus the upcoming 5G networks and emergence of AI, IoT, AR/ VR will enable us to re-imagine and re-shape the world, and make significant business impacts in the next 18–24 months.
The time for getting involved in this historic journey is right now. If you love all things data and digital, ping us.
Special Notes
This article would not be complete without a special note of thanks to my HR partners in crime — Gautam Anand, Mallika Bhargava, Sagar Raina, Preetesh Saha, Diksha Singh, Sampada Gupta, Manu Kaushik, Sophiya, Mannat Kaur— to name a few. Together we succeeded in hiring the best in the world, notwithstanding the long discussions and even, midnight friendly firefights, mostly initiated by me :-)
