Skip to content

6 Questions To Scope Business Requirements Before Starting a Data Science Project To Improve Success

Published:

When there is a data science problem to solve, my instincts are to start coding and build something impactful with the latest technologies. However, I’ve learnt the importance of slowing down and taking the time to ask “Why?” to understand the root business problem from the stakeholders’ perspectives.

The Problem: Stakeholders Don’t Always Know Exactly What They Want

We assume that business stakeholders know what they are asking for. However, they may not always have the technical expertise to understand the nuances and trade-offs in data science. It is important to clarify assumptions, otherwise you may be working on incorrect assumptions and waste resources solving the wrong problem.

Here are 6 questions to ask stakeholders and scope business requirements:

  1. What is the current business process for decision making? What are the friction points in the current process?
  2. Who are the key stakeholders and subject matter experts (SME) with domain and data knowledge?
  3. What is the problem we’re trying to solve or the opportunity to be gained? What is the business impact and how will the solution be used?
  4. What data is available for this project? Who owns it? How timely is it?
  5. What are the current benchmarks and acceptable success metrics?
  6. What constraints are there? What are the timelines and resources available for a feasibility analysis, creating a proof of concept (POC), and productionising?

Another framework you can use is “The 5 Whys”, where you continually ask “Why?” to get to the underlying problem.

I recommend writing these requirements in a shared document so stakeholders can review and provide feedback. This ensures everyone is aligned and has a single source of truth.

Asking these questions and documenting answers will provide you with better context on how to develop your solution and save headaches down the line.


Previous Post
What Are Embeddings and Why They’re Incredibly Useful
Next Post
Traditional Natural Language Processing (NLP) Tasks: Understanding What Large Language Models (LLMs) Can Do For You
-->