Businesses that want effective data science strategies must first build effective data science roadmaps. Kerstin Frailey, formerly of Metis and now Numerator, uses her expertise in for-profit, non-profit, and government data science to outline strategies for building that essential piece of applied data science in her ODSC East 2019 talk “Building an Effective Data Science Portfolio for Your Business.” Check it out on ODSC Thinkific.
The Overall Framework
Frailey’s overall structure involves five distinctive pieces:
- identify challenges and opportunities
Her talk focuses heavily on the middle three, but all pieces of this framework are essential to developing a comprehensive and effective data science portfolio for business use.
The first step is looking for ways to implement data science within the business itself. This input should come from a variety of shareholders and should encompass problems that the company may be able to solve with machine learning.
One vital piece of this first step is not coming with a solution first. Solution-first thinking can limit the scope of your organization’s brainstorming and could stymy the process before you even get started. Instead, focus on areas where there is a problem or shortcoming that data science might be able to alleviate.
The ideas should all be impact-focused, Frailey clarifies, not data science-focused. These are potential areas for practical implementation. There’s an obstacle in the way, and if that obstacle is solved, things can get done.
Preliminary scoping is the first piece of pitching. When someone has a potential idea, they must:
- identify the business impact of the idea—“If we can identify churn customers, we can save X dollars in advertising revenue.” Clearly state this as soon as possible to drive impact.
- develop the hypothesis
- identify major obstacles/risks
The second piece is the real pitching. This happens in a large, open meeting full of people with ideas that data science may help solve. The purpose is to get buy-in from all team members, so the objective is to remain optimistic at first. Everything is an excellent idea in order to build momentum. Skepticism is held until the next part.
Winnowing happens at the end of this process. Frailey calls this “the rose ceremony” of the pitching process in which consensus is reached about which ideas are worth keeping. These ideas should cover a broad scope, including “boring” or low hanging fruit ideas. They should have large scale buy-in from the team. These could potentially have a tremendous impact.
Frailey admits that scope is an overlooked part of this process. We get excited as soon as something sounds cool, but this causes us to miss real obstacles that could have been avoided. “An ounce of scoping is worth a pound of fixing all the stuff you broke when you were moving fast,” she jokes, but the principle stands.
Scope involves three parts:
- technical—technical team members assess infrastructure, data, and layout potential solution paths, whether data science or not. Decide if something is “good enough” and performs reasonably well and define that beforehand.
- non-technical—often overlooked, but this is an essential piece. Non-technical experts, such as managers, shareholders, or subject matter experts identify the impact, including positive and negative impact for both success and failure, and define what success means.
- in-between—this area addresses the intersection of data science and business knowledge. It’s a collaborative effort between technical and non-technical personnel to define the business impact driven by reasonable data science performance. It also helps identify potential plan b’s and c’s.
Once this happens, documentation can help with future iterations. Good documentation can help streamline scoping for the future and provide precedent.
The next step is planning. Plotting helps in finding a path and then select the most reasonable path. Plotting includes cost, risk, and benefit. What is the level of unknownness? What is the real estimate? What is our accuracy? This also includes identifying interdependencies between projects and the potential impact.
Here is where the project portfolio begins to take shape. You’re identifying potential roadmaps between projects and defining single projects that are their own portfolio. Paths have cumulative benefits and cumulative costs. Remember the dependencies between projects from the Scoping step? These dependencies may help reduce overall costs.
You must balance the reduced cost of doing parallel projects with the risk of five projects put at risk if something goes haywire with a fundamental dependency. It also increases the labor required for monitor and maintenance. This is an essential step in planning.
More documentation happens at this stage as well to make future iterations easier. These roadmaps should have areas to pivot and definitions for understanding how far you’ve come.
The future is going to happen. Following these steps helps you build in plans for when that day comes. It’s the final part of your portfolio in collaboration with data science teams and stakeholders. These steps happening every quarter can help you build a defined data science strategy for practical impact.
ODSC East 2020 promises to have even more impactful talks and workshops for building effective data science strategies and applying data science to business applications for maximum impact. Check it out to learn more about how to can apply the latest tools in data science and AI to your organization today.