How to Deliver Business Impact with Data
Tips for Developing Data Products
Post was originally published in January 2020 and has been updated in May, June and October 2020 for relevance.
After developing data products for 10 years, I learned to ask all of these questions during the kick-off(s) the hard way. Data Analytics/Business Intelligence/Data Science/Analytics/Machine Learning should not be done in a vacuum or for the sake of doing BI/ML/AI. Instead, it should aim to be part of a data product deliverable that helps bring value back to the customers and the business.
What is a data product?
A product is something that's "offered to a market to solve a problem, or to satisfy a want or need" [Koombea].
A data product solves your customer(s) pain-point by leveraging data.
What does a data product deliverable look like?
Ahead of any development, it helps to understand the following about the -- business and the stakeholder/customer -- ask, to help you scope out what it will take to deliver a data product:
How does the company make money? What is its core mission?
What is your customer really after? e.g. What is the underlying goal (vs. an ask to do X approach)? What is their pain point?
Why is this initiative important now (above all other priorities)?
What is the business question?
e.g. Suppose your client/company spent $200K on data + salary. How can the data product you create business impact? Generate (or save) $300K for the business? $400K+?
Has this question been answered before? If so, what was done? what didn't work? what are some lessons learned? What would they like to do differently? What is the business logic we should include?
How will the results/deliverable be used by the stakeholder, to answer the business question?
Who is the stakeholder that will use the deliverable?
Does the stakeholder have any ideas/directions they'd like you to explore? (e.g. If the deliverable is a visual, did they have a sample visual/infographic in mind?)
Are we on the same page about what the scoped-down, POC deliverable will look like?
What will it take to get the deliverable into production? What's the rollout strategy to customers? What does the input and output format look like?
Does the deliverable have an associated KPI?
What's the deadline and associated checkpoints?
Are we on the same page about next steps?
Start with descriptive analyses; only when you understand what happened in the past, should you move onto predictive analyses to forecast what may happen in the future.
It's usually hard to estimate how long something may take when it comes data (esp. sight unseen). I recommend time-boxing the work instead, which also fits better into the Agile process. Katie Bauer calls the time-boxed analysis the Minimum Viable Insight/Analysis [14 min], which is then presented to the stakeholder(s), to discuss next steps.
If you need an expert to help you scope out or develop your next data product -- or you'd like to swap horror stories, please reach out.
Keywords: Data products, business impact and value
You may also like:
Making Data relevant to Businesses, by Ravin Kumar (PyData 2019)
What's Data Science Reporting, by Amy Tzu-Yu Chen (PyData 2019)
Why You're Not Getting Value from your Data Science (Harvard Business Review)
Why organizations fail to make data-driven decisions, by George Xing
ASCCR Frame for Learning Essential Collaboration Skills, by Eric Vance and Heather Smith
Why your organization may need a Data Product Manager, by Seth Rosen