How to get started with your data science strategy

Updated: Jun 23, 2019

Every business needs a plan to keep it organized and constantly on target and, in our opinion, every data science endeavor needs the same. Without one things get nebulous at best and chaotic at worst (kind of like a business without a business plan).

A data science strategy is like a business plan, a living document you continuously refine to help you organize your interests, mitigate investment risks, test hypothesis against, etc. In this post we share with you how to make a first draft of this document by gathering all your stakeholders' ideas together in one place, and rank-order them in terms of ROI. Here’s how we would draft the first version of this document.

First, gather all the relevant leadership stakeholders who have a vested interest in developing your data science capacity, and get all of their ideas out on the table. This doesn’t need to be done in person, depending on your situation you can do it digitally and asynchronously using a Google / Dropbox sheet.

Have them list out all the pain-points (e.g., costly internal bottlenecks) and revenue-driving ideas (e.g., improving CLTV) they think data science might help them solve. Then try to estimate a corresponding “Value” for each idea as best they can (using whatever reasonable metrics and time-scales make sense to your organization). Value here might not strictly be in terms of revenue or cost savings, but in terms of growth, positioning, or some other qualitative metric. Also have your stakeholders list out any challenges or roadblocks for each idea. To keep organized, try to condense these into succinct statements or questions.

Next, organize this information in a table like the toy example below. This one’s pretty basic, so use however many columns / jargon that make sense in your world (e.g., you might add a column designating the department each problem originates from, the corresponding stakeholder, etc.) - just make sure you include something like the “Opportunity” and “Value Estimate” columns. Honesty rules here - if a Value can’t be estimated for a particular idea, just leave as a TBI (To-Be-Investigated).

When you’re done populating this table, sort the proposed tasks based on Value from highest to lowest (TBIs are assigned to the low-end until proper estimates can be made). Next, add a column called “Cost Estimate” to this table and set your tech team loose on it. Let them come up with what they believe are reasonable cost estimates for each item. If they have any additional ideas (commonly issues involving internal bottlenecks in our experience) have them add those to the table. If they think some items on the table aren’t feasible, talk it over with them, and if necessary remove them. At the very least have them add what they see as technical challenges for each idea to the “Challenges” column. Adding this column to the toy example above would look like this.  

Again, you can add additional columns to get more fine-grained (e.g., separate columns for development time, estimated development cost, and maintenance costs) or to get at organizational goals - this is the simple, condensed version.

Once the techies take their pass you can re-order the items on this table based on ROI, resulting in an ROI-ranked list of your best tasks. This is the first draft of your data science strategy - a living document you’ll iterate on continuously as you go along. It'll help you

  • erase the nebulous nervousness everyone feels when facing down the unknown

  • organize your stakeholder’s broad set of initial ideas into a simple ordered list, getting everyone on the same page about top objectives

  • focus your team and get everyone energized for next steps

  • provide a basis for continuous improvement of your data science endeavors