Not getting what you expect out of your data science team?

The “upside-down approach” to creating the team and why it is doomed to fail

The word is out: 87% of data science projects never make it to production (VentureBeat, 2019). 85% of projects in AI and so-called “big data” fail (Gartner, 2017).? “Through 2022, only 20% of analytic insights will deliver business outcomes” (Gartner, 2019)

In reading that, I will bet a lot of people are surprised by how terrible those numbers are.? However, having worked in data science for more than 20 years (back before it was even called “data science!”) there is only one thing about those numbers that surprises me: that they are that good.

Many companies in their race to become data-driven have taken what I call the “upside-down approach” to creating ROI from their data.? It is unfortunate and expensive.? But it is also predictable based on 4 key pieces that are lacking from these flailing organizations: data culture, the right problem, the right data, and the right people.? In that order.??

Lack of the right data culture

Simply put, organizations with strong data culture foster collaborative environments where data is not just a byproduct, but a strategic asset.? Having prioritized data-driven decision making, they instill a sense of data literacy at all levels of the organization.? They know that the field is evolving at a near light speed pace and that in order to stay relevant they must openly promote continuous learning, investing in their employees’ futures.??

The key to the creation and sustainability of the right data culture is the unwavering support of the executives and senior leadership.? Those leaders with strong data culture view data not just as a tool for operational efficiency, but a catalyst for transformative change.? They empower teams to harness the full potential of data to provide key insights and drive strategic outcomes.

In order for these leaders to create that culture, they have to first accept some things that are hard for business-minded people to do.? A data culture does not mean that you just throw money at a problem or buy the right technology and tools and success just happens.? In fact, those are almost opposite of what those who are truly successful at creating a data culture do.? Instead, creating a data culture comes with first accepting that it takes a lot of time.? It must be done deliberately and methodically.??

Want to know how?? Keep reading!

Lack of the right problem

I have lost count at this point of how many times I have been in a meeting or a conversation with clients where people have a hard problem and assume it will be solved by data science, machine learning, AI, etc.? Worse, some leaders create data teams without there being a problem to solve at all.? This is a costly mistake that results in frustration for management as they spend a ton of money with no ROI and for the employees who struggle to find things to work on to be relevant and have meaningful careers.

In the “upside-down approach,” management hires a bunch of data scientists and expects that there will be something that they create that is good.? (See my previous point about the success rate of data projects.)? Maybe they will, maybe they won’t, but this is a reckless thing to do.? For example, how did you know what kind of data scientists to hire?? Maybe you want someone with a skill set in forecasting?? GIS?? Image analysis?? Without knowing what problem you are trying to solve, you have very likely hired the wrong person or people.??

The savvy leader starts instead with a serious look at what is the business problem that needs to be addressed and works to identify those that can only be addressed with data.? This initial step serves as the North Star guiding the entire analytical journey, ensuring alignment with organizational objectives and resource optimization.? These don’t have to be super complicated problems, but they do need to be things that can only be solved with data.? In fact, it is better when starting out for a leader to start with some quick and easy wins as this helps bolster the overall data culture of the organization.? Some of the biggest returns I have seen from a data solution have involved some really basic solutions.??

Lack of the right data

The old saying holds: “Garbage in, garbage out.”

This could not be more true than for data science.? In talks I have given about life as a data scientist, I have told the audiences that 80% of the life of a data scientist is actually a data janitor.? Those new to the field laugh in disbelief, but you can tell the weathered and experienced data scientists in the crowd based on the tortured expressions on their faces – the weary sign of agreement.

In the “upside-down approach,” managers hire people but do not have data for them to work with.? It simply does not make sense to bring in a team of data scientists if you do not have any data for them.? You might have the most pressing problem with huge business impact and the possibility of tremendous return, but data cannot be invented and none of this will be realized without the data.??

So what do you do if you have such a great problem and not the right data?? There is no one right answer to this question.? Typical options include buying data from a commercial source or trying to identify proxy data within your existing warehouse.? Sometimes data can be synthetically generated.? If your organization has a strong data-driven culture, it probably already knows what data is missing.

Lack of the right people

The “upside-down approach” is sadly incredibly common among organizations of all sizes.? One of the easiest ways to identify it is by looking at how a data team was initially hired.? In particular, the sequencing of the hires can tell you a lot.

We just talked about how you should not hire data scientists without first knowing what the business problem they will be addressing and, second, what data they will be working with.? That is the hint right here: problem, then data.? The problem typically comes from the business and does not require a deep background in data science.? But what about the data?

There are some key people that you need to hire before you worry about hiring the data scientists.? The most important of these are the data engineers.? These people play a pivotal role in the data ecosystem because without them, no data would flow.? Their expertise in designing scalable ETL (Extract, Transform, Load) processes is what is needed to get the data ready for the data scientists.? Collecting, moving, aggregating, cleaning, and storing the data properly from the start will save you a lot of time down the road.? To watch a good data engineer work is like watching Yo Yo Ma play the cello or Julia Child cook.? It is a thing of beauty!

Along with the data engineers, you need to think about access to the data.? Who is allowed to have what access to what?? I can’t tell you how many stories I have heard or personally been a part of where data scientists are sitting around not able to solve the problems they have been hired to solve because nobody really knows if they are allowed to have access to the data.? Say the term “data governance” to a room of Chief Data Officers and they will get some sour looks on their face.? However, it is vitally important to immediately establish roles, policies, standards, lineages, and processes around the handling of data.? If you do this “upside-down,” you are creating nothing but frustration and heart ache for your data team.? Do it right from the start and you again save a lot of time.

Finally, we can talk about hiring the data scientists.? Unless you are already working at a place like Meta, Google, OpenAI, or a handful of bleeding-edge companies, you probably are not looking for a researcher.? You want a data scientist who can provide solutions to business problems immediately.? This is where you return to the starting business problem.? When you are starting out creating a new team, it is generally good practice to hire generalists.? However, maybe you can afford an extra data scientist who has a subfield of expertise such as natural language processing or image recognition or graph data science.? This is where you tailor the hires to that immediate business problem you are looking to solve.?

Then return to the right culture

Once the right culture, problem, data, and team are in place, you are off to the races, right?

Wrong!

Here it is very important to return to the data culture.? Like creating that culture, creating data science solutions takes time.? Good data cultures understand this and carve out that time.? They understand when a quick POC is all that is needed versus something fully deployed to production that will take longer.? They allocate the necessary data engineers to the data scientists (a good rule of thumb is 3:1).? They have good program managers and tools in place.? They adopt methodologies such as Agile and thus have clear expectations on when a realistic delivery timeline is.? The data scientists regularly communicate their results so the organization can see the process at work, provide feedback, and ask questions.??

It is all too common in an “upside-down” organization for problems to appear early on.? I have heard countless stories about the data team not being able to solve the problems they have been assigned because they are not given access to the data they need to do their job.? Gone are the days where data science problems were typically limited to and solved by a single individual.? Collaboration across the business is key.? In a good data culture, data flows to and is used by whomever needs it (and has the appropriate access based on the data governance policies).??

Another problem that can easily happen in an “upside-down” organization is for the data team to become a victim of their own success.? They were initially brought in to solve a difficult and pressing business problem and they got it done.? They get a reputation for being magicians who can solve every problem.? Again, this is a sign that the first two needs, culture and data, are missing.? Organizations with a good data culture understand that you cannot put the problem before the data.??

Parting thoughts

There is nothing magic about anything I have described above, and yet 72% of survey respondents in NewVantage Partners’ 2021 Big Data and AI Executive Survey said that only 24% of Fortune 1000 businesses considered themselves to be data driven with 92.2% of mainstream companies reporting that they continue to struggle with cultural challenges relating to data (Harvard Business Review, 2021).? For companies of this size, change is hard.? It is not impossible, but takes significant time and effort led from the top.??

For those just starting out creating a data science function, you have a golden opportunity to get it right from the start!? Please reach out to me if you would like to talk more about how!