Top 5 Lessons learnt from my Analytics Translation Roles
ARTICLE SUMMARY
Analytics Translation is not only a buzz word, but the bridge between data science and data-based decision making. Find out why.
A few days back, I was speaking at the alumni connect webinar series for my alma mater and some students asked me about analytics translation.
I told them it’s an emerging term but a must-have skill for data science professionals when they are interacting with non-data science stakeholders. But that question got me thinking, and I decided to pen down my top 5 lessons learned as an analytics translator.
Let me first define what I mean by analytics translation. Analytics translation is a responsibility to communicate the findings/insight, data limitations, machine learning algorithms, analytical solutions/framework, and underlying assumptions to someone who does not do data science on a day-to-day basis. Such stakeholders can be anyone from C-level executives to IT teams, to business users, from Operations teams to Sales and marketing teams. Now, I knowingly avoided using the term ‘non-data’ person. Even if corporate professionals do not do data science day in and day out most of them deal with data every day and hence no one is a non-data professional.
So, analytics translation is a bridge between data and business. Now let’s delve into what I learned from doing these roles in different capacities over the years.
Here are my top 5 lessons.
1. Build up the tempo starting from a simple example.
It’s a very common blunder to start explaining algorithms/concepts by taking a full-blown real-world example. If I start telling my CEO how regression works starting with n dimensions which correspond to different variables and how they should visually interpret the sum of squares of residuals and how we want to minimize the error; it might be a useless effort. Instead, If I start with one dependent variable and one independent variable and explain in a graph it shall be much more appreciated and understood. And then I could build up from a simple example to a multi-variate regression problem.
2. Prepare to explain peripheral concepts.
While preparing for an analytics translation meeting, we should always be ready to explain the peripheral concepts as well. For example, if I must explain overfitting in a model, it shall be a good idea to explain bias and variance trade-off or be prepared to move away from the current topic and explain the topics that are co-related or at least should make provisions for the peripheral concepts to be covered. Leaving small gaps can lead to unsuccessful communication.
An example showing low bias and high variance in an over-fitted model. The blue dot denotes training data and the green dot denotes test data. Fig Source (977) StatQuest with Josh Starmer – YouTube.
3. Summarising the concepts at the end.
It’s very important to summarise the concepts at the end and connect the two dots. It can be overwhelming for anyone to listen and grasp so many concepts in one meeting. Wherever feasible, analytics translation meetings should be done in phases to give time to digest the concepts and internalise the learnings. Hence, it would be a good idea to summarise everything at the end to make an effective conclusion. The picture below explains the concepts of precision and recall. Imagine, you are explaining the statistics of the classification model to your business stakeholders, and at the end, you leave them with this image, chances are that it shall get imprinted on their minds easily.
4. Excessive usage of visual media.
Well! this point is self-explanatory. The more graphical description of the concepts, the better it is and easier it is for everyone. The power of making a good presentation and using whiteboards extensively during these meetings are highly recommended. Below mentioned are my attempts to explain the concept of machine learning to a newbie.
5. Staying humble.
Last but not least, staying humble during these meetings is half the battle won. It shall not help anyone if a data scientist starts to throw jargon and goes to the meeting with a preconceived notion that the audience knows a set of concepts. It is imperative that we do these analytics translation sessions without any hint of superiority, and this shall go a long way towards keeping the meeting true to its original theme. Staying humble is a self-check towards avoiding any jargon without giving the proper definition and context. It helps the data science professionals truly understand what the audience expects and prepare the content accordingly.
Analytics Translation is a hot topic for the data community now and as much as it’s a data scientist’s or data analyst’s responsibility to explain analytics to all stakeholders; it’s also the responsibility of the person sitting at the receiver end to keep the session interactive and keep asking relevant questions from time to time to make the delivery person understand how well it’s going and where should the discussion be heading.
Analytics translation can be a game changer for decision-making if used properly and can adversely affect the data science team if used without full knowledge and if used without the right intentions on both ends.
Abhigya Chetna is a data science professional. Over the years, she worked to provide analytical solutions in a wide range of industries across the globe. This diverse experience gives her an edge to crossbreed her skills and deliver impactful solutions.She enjoys being an analytical translator for non-data professionals and aims to work for data literacy.
