4 Critical Success Factors for Data Science in The Enterprise

I have taught companies about Enterprise Data Science for almost five years. This concept explains how to transition from early stages in the Machine Learning Maturity Model to advanced applications.

Vin Vashishta | Originally Published: January 30th, 2018

Book time with me for Career Coaching or sign up for my Business Strategy for Data Scientists class.

Why are some companies so successful with data science and machine learning while others struggle to produce the promised ROI? Why do some businesses seem to organically adopt data science while others build solutions that either never make it to the user or are rarely used once deployed? Why is it that data science is promised but analytics and reporting are all that is delivered?

Hiring a data scientist(s) is not the starting point in building data science capabilities. It is a complex buildout that starts with a comprehensive strategy. That is the difference between companies seeing success and those experiencing poor or inconsistent results.

Like any complex capability build out, a roadmap clarifies how the business plans to move from where they are now to delivering value using the new capability. There are 4 critical success factors for data science and machine learning to succeed in the enterprise.

C-Suite or Board of Directors Level Sponsorship

The most important realization for a company using or planning to use data science is that this will change how you do business. That means most teams and operations will be impacted by these capabilities. Data science is not something a single team or individual can make successful. The business as a whole needs to buy in and support initiatives.

A clear vision articulated by members of the C-Suite helps drive that point home. When I see a data science initiative with C-Suite visibility, I am looking at a setup to fail. Visibility is not enough to drive the kind of change needed to facilitate data science capabilities.

Data science fails in a silo. CxOs and members of the Board are the only people with the authority to make enterprise wide changes. Although execution happens at lower levels, buy in starts at the top.

Strategic Alignment

Each data science initiative must be connected with business goals and strategy. Treat data science and machine learning algorithms like any other software product the business creates whether for internal users or external customers. They are built to address business or customer needs. They are more expensive so they should be used when there is not a most cost-effective alternative.

The business case for each initiative must draw a clear line between the project and either savings/efficiencies or revenue. This connection keeps data science from drifting into the weeds and failing to deliver on its promise. That is the why behind data science and machine learning. Companies that have a real, well supported case for data science succeed.

It is much easier to get companywide buy in when the reason for data science is not, “Data science is the future.” or “We need to stay competitive.” “We expect cost savings of X based on these initiatives and new revenues of Y from these products.” That is an answer people trust and can get behind even if they do not understand the underlying technology.

Setting Up for Success

Data science is not plug and play. A data science team cannot succeed without infrastructure. If you have heard the statement that a data scientist spends 80% of their time on data wrangling, you are listening to the story of a business that didn’t set their data scientists up for success. Data quality and data security are two pieces of infrastructure that are critical to build so data scientists can get to work building solutions.

There are cultural changes needed to support development and facilitate the adoption of data science work products. Deploying, supporting, and maintaining machine learning algorithms presents unique challenges that supporting departments need to be ready for. Data science and machine learning products do not work like traditional software. Users need to be ready for those differences or they will not adopt new products.

Tools and framework pieces need to be in place as well. These pieces help to facilitate the everyday work for data scientists making them more focused on solutions than on logistics. These pieces help with data sourcing and development tasks as examples.

Staffing Strategy

Data science is a team sport and building that team requires a strategy. Much of the certainty built through the first three critical success factors starts delivering value here. When you see a data scientist job description with 3 programming languages, 7 frameworks, SQL and NoSQL database knowledge, vague requirements with regards to models used, PhD required, lengthy years of experience, etc., you’re looking at a business that doesn’t know what they need so they’re asking for everything. They will not find it.

One of the key benefits to building out the business cases and frameworks in advance comes in knowing exactly what capabilities you are looking for in data science hires. The business is also in a better position to staff the appropriate number of teams and make intelligent decisions about what to build internally versus what to outsource.

Finding a data scientist is not as hard as some would have you believe. Finding a data scientist with the right capabilities to deliver value to your business is much harder. Assembling the team that allows them to focus on high value activities is the next challenge. To tackle both, HR needs tools, clear direction, and the latitude to attract top talent. A complete staffing strategy allows the business to attract and retain top talent.


Often, businesses hire a data scientist to tell them what they need while the data scientist waits for the business to give her/him clear direction about what it expects. This results in a standoff with little value delivered. Companies can spend years in this cycle before they have learned expensive lessons from multiple attempts at data science.

With that alternative, building the roadmap up front is obviously the more cost-effective way to go. How much of this can be outsourced? 3 and 4 are viable outsource candidates when data science does not make sense as a core competency for the business. Many businesses begin with an outsource partner then move capabilities back in house after a year or two. This gives a business the time to build capabilities and the benefit of near-term results.

However, completing 1 and 2 is not something an external business should be tasked with. Advise on 1 and 2? Sure, bring in whoever is necessary to share their expertise but make sure the business owns the process, not the consultant. Bring in an outsource partner once sponsors are in place and the business case is built for at least the first initiative. That insures the outsource partner builds what the business needs, not what the partner thinks the business needs. It is a key distinction.

Finally, think long term. Data science and machine learning are still in their infancy. The field is evolving rapidly. What companies are building today needs to be engineered with an eye towards what is next. Scale, combinations of machine learning with other emerging technologies, security threats, and government regulations need to be taken into consideration. Satisfy business needs today with systems that can be updated as those needs evolve.