Daniil Bratchenko is the VP of Business Operations and Analytics at DataRobot, a machine learning platform for data scientists to quickly build and deploy accurate predictive models. Read Daniil’s Medium and follow him on Twitter.
Can you tell us about your journey from software engineer to heading up Business Operations and Analytics?
For most of my career I’ve worked at small companies or startups where I’ve had to wear many hats. I learned to focus on adding value in any possible way, rather than sticking to any one specific role.
I joined DataRobot as a front-end engineer because, at the time, the company needed a front-end engineer. The initial role lasted about two weeks. After that, I worked in various roles across the R&D part of the company. It was the only part of the company back then.
When we were satisfied with the product and switched more focus to Marketing and Sales, we wanted to run the business side of the company the same way we ran engineering: rely on well-defined data-driven workflows that would allow our flat organization to grow without sinking into chaos.
My experience with building workflows made me a good candidate to help make this vision a reality.
Business Operations might not be the best term to describe what I do, but we use it for the lack of a better one. My role is to build what we call machines, or systems that connect people, applications and data into effective and scalable workflows that help the company achieve its goals.
Can you tell us about how data is represented and organized within the DataRobot organization?
At DataRobot, we focus on data that helps operate the company day-to-day and quarter-to-quarter.
Data can be collected and represented in many ways. Only some of them are effective. When deciding what data to use we look at goals and workflows. Before investing time into gathering data or building a dashboard, we answer a few questions:
- Who will be looking at the data?
- When? What will trigger them look at data?
- How will they use data to make decisions?
In most cases, we use the “dashboard+algorithm” approach. This means that we develop both a dashboard and a script to review the dashboard at the same time. Then we iterate on both the dashboard and the algorithm until we are happy with the result.
We generally have two types of dashboards or reports:
- Operational: used by a team that does the work to make day-to-day decisions
- Monitoring: used to review the performance of a given team or workflow and make adjustments
You’ve written about how metrics can change an organization’s behavior, what are some of your success stories?
There are many ways data can add value:
- The right data at the right time increases the quality of decisions dramatically
- Mere exposure to relevant data changes people behavior
- Data helps learn and evolve
- Shared performance indicators is a great way to align people for achieving a common goal
Besides the case you referring to, a few other stories stand out.
Our CEO’s meeting room looks a bit like a command and control center. It has a few huge screens with dashboards. He refers to the dashboards often during meetings to ask for details or point out facts that do not add up. This works so well that one of the first things he asks people when assigning them a mission is to build a dashboard that would allows them to monitor results.
This data-driven approach percolates throughout the company. Some teams adapt it better than others. For example, many members of our Sales Development team build custom dashboards to augment their conversations with the Sales Representatives they work with.
The biggest indicator of success for me is when people who join the company say that they never saw so much of the company’s operations being measured and analyzed this early in a company’s life. They mostly mean it in a positive way.
Do you have advice on how to get people to rely on data and dashboards?
You need to show people the value. There are two types of value from data:
- Visibility and control
Feedback helps you understand how you’re doing and you can adjust behavior. Its value increases if you have clear measurable goals, you can track achievement of the goals and make decisions based on intermediate results.
Visibility and control are mainly for people who are responsible for a part of the organization: managers and leaders.
Then there are people who prefer data and people who prefer anecdotes. You need the former if your organization does not have data in its DNA.
Find early adopters that meet the following criteria:
- They like data
- They need either feedback or visibility/control that data can provide
- You know where to get data they need
Something to watch out for: be very clear with exactly how the data will be used in people’s workflows. Often, people think they need data, but when they have it, they don’t use it because it does not fit into their day-to-day operations. Ask “how exactly are you going to use this dashboard to make decisions?” and be critical to the answers.
When you have successful cases with early adopters, recruiting others into your data sect is easy.
How do you see the future of analytics?
Demand for extracting value from data far exceeds the supply of people who can do the job. This is good news for data analysts, data scientists and other people who possess the magic of turning data into actionable insights. It does not look like the trend will reverse anytime soon.
This means the world will try to adjust by:
- Training more people to work with data. This includes both people who solely focus on working with data and people in other roles who get some data analysis skills.
- Automating data science (this is what DataRobot does, by the way).
- Creating more products that perform data analysis on a limited set of repeating problems.
Whatever profession you have, you would be better off understanding data.