Original article was published by Goda Ramkumar on Artificial Intelligence on Medium
10 dimensions of making data science work (part 4)
Collaboration: Dimension #4
Here is part 4 of the series of 10 dimensions of making data science work. Just as a refresher before moving ahead, we covered attunement to have the right Expectation in part 1. In part 2, we understood the art of getting to means from the end — Strategy and we talked about Roles in part 3 that make the data science function a whole. So those who have read all three parts kind of know what to expect, how to plan, and who to staff the team with by now. Here comes the next major challenge -the fourth dimension which determines if there is momentum or only motion. The dimension of Collaboration.
Given most startups begin as very small teams who almost know each other’s siblings’ names, collaboration often does not seem a challenge in the beginning. As and when a start-up enters the growth phase, teams grow; new people start joining who bring in newer perspectives, practices, and ideas and that is when collaboration to achieve a common goal starts seeming like a thing to do.
Collaboration is a result of clarity of goals, organization structure aligned with the goals, practices that bring people together to achieve the goal, and evaluation mechanisms that measure performance as a team
Collaboration: 2+2>4 ?
Here are 5 stages in which collaboration evolves in a startup. You need to know where you are to see where you are going.
Stage 1: Not in the picture — During the early stages of a start-up when achieving product-market-fit is the goal, the roles are themselves not separated and hence collaboration is not in the picture. Most decisions are driven intuitively and via market pulse study with little help from data function which might not exist in its full form.
Stage 2: Embedded and in-silo — This is the phase when analytics drives the data function to achieve trackability and insights. The need for speed leads to analysts embedded within product and business functions. Each function chases several what-if analysis and studies the behavior of various metrics to plan the next big move leveraging their own set of analysts. They don’t talk to each other much and work in-silo and mostly in local environments with less visibility of insights across the organization. This stage is probably the one where only the CEO knows about all efforts undertaken. It is common to be in this phase but also necessary to see the signals to make the effort to move to the next stage before this stage goes out of hand.
Stage 3: Centralization with some motion — This is the phase where analytics function evolves into a central team serving multiple business and product functions, initial versions of data-platform are built and a platform team that handles the same emerge, a few data scientists find their first problems to solve. At this stage, there is a good feel of a lot of motion but usually, it wouldn’t all be moving in a direction to achieve a goal. Hence it would be motion without momentum. It is expected to happen but should not be allowed to sustain. What helps at this stage is applying the Drivetrain model and getting a data-strategy blueprint as we spoke about in the Strategy dimension. A good example of the same is also explained in the article on tweaks that could turn the data science journey from typical to ideal. This would help clarify goals for different functions of the organization and identify parts of the picture that are dependent on each other and hence move together. A good recipe to set effective OKRs for the organization to chase.
Stage 4: Motion to Momentum — This is the stage of getting the most out of data science function and actually making 2+2>4 via collaboration. Making efforts to lay down data strategy blueprint with clarity, ensuring the teams are structured right tied to OKRs, and staffed with the right roles and talent is the beginning. What makes all this effort effective is the garnishing with practices that make these teams come together and work towards a common goal. Defining OKR only does half the job, the remaining is done by defining a pod to achieve the OKR that is staffed with all kinds of people — data engineers, data scientists, product, analytics, engineers, QA, and so on. Pods cut across boundaries of teams and make planning collaborative than in-silo. I have personally seen this working wonders in bringing a sense of mission in people. There are always architects, heads of data science, analytics, and platform functions who keep a watch, recognize patterns, and ensure there is less divergence across pods. What is crucial for people to feel part of the pods is to ensure personal aspirations and goals are aligned with that of the pod. If the assignment of people to pods is not thought-through or we continue to evaluate people’s performance based on criteria other than the success of pods, we would be taking a half-baked approach that is bound to fail.
The mantra is —A clear plan, Specific goals, Cross-functional pods and an Evaluation mechanism that ties it all.
Stage 5: Making it a norm— This is the most mature stage when collaboration becomes part of the culture as all practices of strategically prioritizing goals, setting OKRs for the goals, forming streams and pods, and performance evaluation tied to pods become the norm. Coming back to the data strategy blueprint every 6 months to make amends and plan further becomes a practice. The icing on the cake would be adopting tools that make sharing, tracking, and talking easier. After this, every new business unit and every new venture the organization steps into just needs to adapt to this faster and move with speed.
Collaboration is what makes lego blocks build something creative. Otherwise, can you imagine lego blocks with no way to connect to each other? No function in an organization is independent to achieve the goal sitting in a closed room. Definitely not data science.
Soon, we will cover the fourth dimension of Culture in part 5 of this series.