Data science, artificial intelligence, and business intelligence are all hot topics in business today, but what are they, and how do you use them to drive business value?
This resource will explore data science, how to set up your first AI initiative, what to avoid, and how to get the most out of your investment.
What is data science and how does it relate to business intelligence and artificial intelligence?
Data science is the field of study that combines domain expertise, programming skills, and knowledge of mathematics and statistics to extract meaningful insights from data. Data science practitioners apply machine learning algorithms to numbers, text, images, video, audio, and more to produce artificial intelligence (AI) systems to perform tasks that ordinarily require human intelligence. In turn, these systems generate insights that analysts and business users can translate into tangible business value.
Business intelligence is hindsight. In essence, what happened? Data science is traditionally focused on foresight. What will happen or how do I make something specific happen? This foresight allows business leaders to make decisions for the future and has the highest level of value to organizations.
Download to learn more
How do I convince my team that data science is a worthy investment?
There is immense business value that can come from strategic AI initiatives, but if you can’t get your team on board, it’s hard to see the real benefit. Take the following steps to strategically develop your first project so you can get buy-in from your team.
Stop the static.
Everyone in the industry is stating they have some sort of machine learning tool or AI algorithm. Start by taking the noise out of AI and data science. An easy way to do this is to walk through the 4 levels of analytics capabilities and explore the benefits of progressing to more advanced levels.
In order, the 4 levels of analytics capabilities can be defined as descriptive, diagnostic, predictive, and prescriptive. Each level provides a more advanced degree of analytical capabilities, helping your organization ask questions that are focused on hindsight to those that are focused on foresight. Descriptive analytics can sometimes be just as valuable as predictive—it’s all about how you use them.
Make it accessible.
Make the process of building analytics understandable and consumable by your less data-savvy team members (not just coders or data scientists). Making your project accessible is an excellent way to empower your employees and allow them to create additional business value.
Data science tooling helps. Making the process visual allows beginners to easily comprehend the action that is taken at each stage of the analytical process. It’s a great alternative to diving right into Python code that has a higher barrier of entry for non-technical people you need to endorse the initiative.
It’s important, however, that this tool goes end-to-end in the data science process in an approachable and valuable way. It needs to address the coders and clickers. If it doesn’t address both, you run the danger of creating walls around your project instead of breaking them down.
You need to be transparent throughout the process of data science and AI so that your team understands that there’s actually a science behind it. Everyone needs to see and understand how a prediction or answer was arrived at. It’s important to talk about the terminology and tools in a way your team understands.
Demonstrate what you’re going to do to get the answer they’re looking for. A demonstration with the right tool makes it very approachable for your team to see what’s happening here. It isn’t magic—there’s actually science behind it.
A good data science tool allows for exploratory analysis, data prep, data cleansing, and Auto-ML for the Citizen Data Scientist (tech-savvy employee with a business background, contributing to AI activities). It also should allow your talented data scientists to build custom models, enabling your team to collaborate in the same space.
The correct tool will also make deployment of the model easy for IT to use (think APIs or curated data sets exposed to Tableau).
While you’re selecting your tool, think through these questions:
- Is it easy enough that business analysts can understand it?
- Is it elaborate enough that it can handle data scientists working inside it?
Identify the return on investment (ROI).
The use case you select for beginning an AI initiative needs to have ROI attached to it. It shouldn’t just be exploratory—there needs to be a reason. Is there value in confirming a gut feeling through data science? What about disproving a gut feeling?
One exploratory statement we like to use when determining ROI with our customers is: “What is it worth to your business unit to know or be able to predict X?” More often than not, the answer is typically, “A lot!”
The bottom line is that your AI use case should lead to cost savings and/or revenue generation.
As you’re developing your first AI initiative, you need to make AI more accessible to the rest of your team in order to get buy-in. Cut out the noise, be transparent, and choose a use case with demonstrable ROI. You can make your initiative even more impactful when you encourage your team to get involved by empowering them and allowing them to add value. Get a quick win and the next AI initiative you pitch will be even easier.
How do I prioritize the use case I should tackle for my next data science project?
A data innovation session is perfect for prioritizing your next data science projects. We focus on a quadrant exercise. This 90-minute is meant to uncover and prioritize questions that your business wants to answer.
In this meeting, the goal is to find the best value business case to use. We explore what questions can be answered through data science and organize them in a quadrant, as seen here.
Who should be involved?
The meeting is best kept at 10 or fewer total people (we bring 3). There are 3 roles of people that we need. (A person can cross more than 1 role.)
- A Leader: One that sees the value in AI in general and is targeting that next effort
- A Business or Subject Matter Expert: Someone that knows the business process or knows where there is value to be gained
- A Data Junky: Someone that knows both the data and the data sources. This is often a Business Analyst that is really great at Excel or a Data Analyst that does a lot of SQL.
In these discussions, we like to know the data sources and enough business knowledge in the room to know where there is value.
Out of this exercise, your team gets a prioritized list of “Data Science” solutions ranked by business value and data accessibility. The top of the list is how we select the first AI project.
I don't have a team of data scientists, does this mean I can't do data science?
A recent study by IBM suggested that we will need a further 28% more data scientists worldwide by 2020 to cope with an increased need. Combined with an average annual salary of $150k for data scientists, this creates a large problem many organizations are facing today.
Data scientists spend around 80% of their time preparing data for analysis, even though 76% of data scientists view data preparation as the least enjoyable part of their work. This poses the question, “Why do we have data scientists doing work lower-paid employees could handle?”
Enter: the citizen data scientist.
A citizen data scientist is “a business person who aspires to use data science techniques such as machine learning to discover new insights and create predictive models to improve business outcomes.” This was identified as a job function that will grow at five times the rate of the traditional data scientist in the next two years by Simplilearn.
How do I find a citizen data scientist?
By reskilling or upskilling employees to take on these enhanced data roles, organizations can make better use of data which leads to both cost savings through greater efficiency and competitiveness through access and use of that data.
Surprise! You may already have one!
The perfect citizen data scientist might already be on your team. These may be business analysts, data analysts, system analysts, business consultants, or reporting analysts. Look for individuals who are skilled in Excel, SQL, statistics, and data visualization tools. Additionally, this person should have the following traits:
- A contextual vision of the organization
- Proven application of analytic techniques to business problems
- Appetite for what matters relative to business priorities
- Been around the block and has connections
- A unique perspective of an individual business area
- Able to go to bat to justify business value
- Involved (hands-on) in multiple analytic areas and activities
How can citizen data scientists provide value today?
Your citizen data scientists can be powerful assets for your team. They can add business value by empowering management and officers to make better decisions, challenging the company to adopt best practices and focus on issues that matter, making and testing decisions with quantifiable data-driven evidence, identifying opportunities, and identifying and refining target audiences.
How do I find a best-fit data scientist?
The ROI of data science is evident. But, finding and retaining skilled workers to do the job is increasingly difficult. The overall labor quit rate in the US is at 2.3% and climbing. That percentage is much worse in a high-demand role like data science with some polls predicting that more than half of data scientists will be leaving their jobs next year.
If you want to take advantage of better business decisions and outcomes, you have to figure out how to close the skills gap.
What to Look for When Hiring a Data Scientist
Start by educating yourself on what a day in the life of a data scientist looks like. That way you can understand what they’re providing to your organization and whether it will be an effective investment.
In business, you should be targeting outcomes, not insights. As a result, your data scientists should have more than just academic experience; business experience is also mission-critical. When hiring, we recommend specifically asking potential candidates about what business outcomes they have driven.
Another key quality is the ability to effectively communicate a narrative. When asking a candidate about their experience, can they communicate a narrative end-to-end? They should be able to explain what the business problem was, the insight that was found, how they operationalized that insight, and the business outcome it drove.
Finally, the ability to collaborate is a must-have. Presently, many data scientists are hired but then are essentially placed in a closet. If you are managing a data scientist, be prepared to protect their time but don’t quarantine them from the business; data scientists that collaborate are the best, as it’s important that they want the team to improve. When interviewing, ask if potential candidates have been part of a team before and how they worked within a team.
How to Strategically Fill Skills Gaps
If you have five open roles for data scientists right now, you should revisit your actual needs before making any hiring decisions. Do you really need five highly skilled individuals (that are increasingly expensive)? Instead, you could utilize a handful of citizen data scientists to do more of the upfront data collection and preparation so your data scientists have more time to spend on expert-level activities.
Automation can also take a lot of the work off your team’s back. By automating model retraining and deployment, your data scientists can focus on the next project instead of monitoring past ones.
Finally, you can increase your team’s capacity through consulting. Ensure your consulting partner has the correct perspective and supports data science as a competency within your organization. Avoid consulting groups that don’t embrace your organizational success in data science.
How do I retain a data scientist?
As stated initially, more than half of data scientists plan on leaving their roles this year. The following key factors that play into talent turnover. Consider these to retain data scientists on your team this year.
- Effectively Train and Onboard - Standardize onboarding procedures to set accurate expectations upfront and get your new data scientist up to speed quickly. This will set your team up for immediate success, making it easier to collaborate with the rest of the team, move forward, and find greater wins with future projects.
- Minimize Bureaucracy and Hinderance - Enterprises are struggling to select where analytics groups exist in organizational structures, which leads to bureaucracy. Figure out if your advanced analytics team is a COE that runs distributed projects, exists in your IT organization, or exists in your business.
- Ensure Support from Managers and Leaders - A lot of leaders don’t understand what data scientists do or what it takes to be good at data science as an enterprise. Be sure to train your leaders to be supportive, knowledgeable, and effective.
- Encourage Collaboration - Don’t put your data scientists in a closet. Get them working with the business and collaborating with your team to drive outcomes.
- Provide Access to Tools - A lack of access to tools is another huge reason data scientists leave. Utilize an end-to-end tool for the operationalization of machine learning models.
What can I do to ensure success for my data science team?
Your data science team is the driving force behind all of your analytics efforts. When your team works together effectively, you’ll see more impactful business results from your analytics initiatives.
The following five strategies will help your teams work together better, so you can ensure successful AI initiatives that drive value for your business.
Collaboration is king.
Get more out of your data science team’s effort by ensuring effective collaboration across departments. Do this by being intentional about collaboration and scheduling quick-hit meetings. These are fast, 30-minute recurring meetings that happen weekly to bi-weekly to guarantee your team is cooperating and ensure there’s a feedback loop. These will only be successful when business, IT, and analytics teams are all involved.
As part of these meetings, everyone should be checking if the numbers look right. It’s crucial to discuss how the analytics were created and if they have business value. The bottom line is: Does it make logical sense?
In one example, a manufacturing company was working with a data set that showed when one operator was working on a machine, it had the most failures. If we had moved forward with that information alone, the operator would have been identified as the variable. However, after having a quick-hit meeting with the rest of the team, we were able to identify that the operator in question was doing twice as much work and should not be considered in predictive failures without accounting for that volume disparity.
All in all, collaboration is mission-critical. You’re doing something wrong if the business team sees the project once at the beginning and once at the end.
Focus on upskilling.
Instead of putting an over-qualified resource to work on data preparation and cleaning tasks, citizen data scientists can collaborate with your data scientists on data feature creation, helping you to get the most out of your resources.
When you empower your Excel gurus to take their capabilities to the next level and upskill them as a citizen data scientist, you not only are able to fill labor gaps, but you see an added benefit of having additional viewpoints and experiences weighing in on a project.
Re-use and share data sets as feasible.
This is an incredibly underrated way to improve team output and performance. Your team will typically spend a lot of time on a data set. Why use it only once?
Once a data set is enriched, very often it can be used for more than one solution. Explore opportunities to re-use data sets wherever possible to make the most out of your team’s efforts.
Plan for growth.
Many organizations assign one project to a single data scientist for months at a time, but teams can’t scale with one data scientist to one project. When it comes time to deploy their algorithm or solution, that same data scientist is now in charge of supporting and retraining that algorithm for the foreseeable future. This is not a scalable process when you have 10+ models deployed into the business.
You need to have a plan for how to keep headcount low, collaboration high, and lean on automation for scaling, retraining, and automating future deployments.
Insights are great, but they need to be deployed into the business to have an impact, making the operationalization of models key. Then, track key performance indicators to review the overall ROI of the project.
Revisit outcomes over time to see if the ROI has changed (either increasing or decreasing). Your solution may have resulted in significant cost savings without any recognition of the impact. Ensure there is a feedback loop in place to inform the rest of the team of the impacts of your analytical efforts.
Above all else, your team needs to effectively collaborate. Past that, upskilling employees, planning for growth, and making the most use of your resources will help make your team even more successful. Finally, by tracking outcomes, you ensure valuable insights are derived from your efforts and can make a strong case for future analytics initiatives.
Last updated: January 2020