cft

On Collaboration Between Data Science and Product Teams

Collating some of the learnings from my experience of collaborating across functions:


user

Ved Vasu Sharma

3 years ago | 3 min read

We at SquadStack are currently focusing heavily on building Auctm (Smart Assistant for real estate brokers and agents in the US).

As a Data Scientist, it had been a great opportunity to work with an amazing Product Team and learn about end to end product development involving the deployment of many Data Science driven features and Machine Learning models.

Since the product thesis involves Data Science at its core, cross-functional interactions with various functions like Product, Design, Engineering, Operations, and Customer Success is indispensable.

It helped me gain incredible insights and comprehend the importance of each and every role in product development.

It's astonishing to see the efforts put in by all the different teams: Product team leading the overall product development, the Design team thinking deeply about user experience, the Engineering team (both frontend and backend) building products that can scale,

and the Operations team managing all the product-specific functions for smooth customer experience and also providing valuable feedback on the performance of ML models in productions.

Collating some of the learnings from my experience of collaborating across functions:

  1. Speed matters: For startups, speed matters, and it's important to go live with the models as soon as possible even at baseline performance so that critical customer feedback starts to come in early. Eventually, with a better understanding of the problem and more training data, we reach the highest possible performance of models.
  2. Plan ahead for the product requirements: Ideally, the Product team works ahead of Design which works ahead of Engineering so that each function can perform with proper visibility. But at times, the Data Science team even has to work ahead of the Product to facilitate deep explorations of DS features that may be required in the future. The efforts may or may not be utilized in the product but it definitely helps in making better decisions at the product end.
  3. Keep in mind the limitations of Data Science: Probabilistic nature of data science work differs from “normal software engineering” and hence it becomes important to set the right expectations across the board in terms of model performance, automation, and plan for future iterations. Data scientists also have to provide enough visibility to the product regarding what should be done today to make certain features feasible in the future.
  4. The process to monitor model predictions: Since the performance of the models can reduce in the production environment due to various factors like low accuracy of training data, change in the patterns over time, or due to uncertain times like COVID, etc, it's critical to have a process of internal review/ sampling of model predictions until we are absolutely sure of model performance. At SquadStack we have highly trained domain experts and analysts who have immensely contributed to improving ML models by providing their valuable feedback.
  5. Effective collaboration with Product (Consider Being Product-Minded): Discussions around the ideas and feasibility of a solution are very common in a fast-paced startup environment. Hence along with Data Science, product understanding with relevant software skills plays a critical role in deciding the feasibility of a solution and in collaboration across teams.
  6. Effective collaboration with Engineering: As Data Scientists we focus heavily on model performance in terms of accuracy, precision, and recall. While working with the Engineering team, model debugging also becomes critical. Factors like the latency of the model predictions and statistics around RAM and CPU utilization need special attention to figure out appropriate infrastructure for deployment. Even overall code quality and decoupling of the model pipeline becomes important for debugging pipelines in case of bugs.
  7. Provide solutions focusing on customer needs: There may be many approaches to solve any particular DS problem but while working with Product, we need to think from the perspective of what really creates an impact for our customers. Especially for startups, focusing on features and solutions that drive customer adoptions helps to achieve early PMF. Frequently hoping on to customer calls helps a Data Scientist to build the required understanding.
  8. Debate Adoption vs Accountability for each feature: For the early stage of product development, adoption supersedes everything (at least until PMF). If we can’t drive adoption we can’t drive accountability. As a Data Scientist, it's always important to design solutions for a feature based on customer behavior and the degree of change that feature brings in the product.
Image: Adoption vs Accountability | Behavior Change vs Product Change
Image: Adoption vs Accountability | Behavior Change vs Product Change


Upvote


user
Created by

Ved Vasu Sharma

3+ years of experience in building AI capabilities while working in Startups. Worked on Data Science and Machine Learning applications across various domains like Real Estate, B2B (Sales), and C2C (e-Commerce). Currently working at SquadStack to build Auctm - Smart Real Estate assistant for Agent


people
Post

Upvote

Downvote

Comment

Bookmark

Share


Related Articles