While working on his degree in management information systems, David became very curious about finance. That led him to becoming a financial analyst. But after doing this for about nine years, he realized that he missed working in tech:
"That's when I moved to San Francisco and got into analytics and data science. I think of my job as fundamentally doing the same goal, which is achieving impact for a business. But I was now working with much more fun data — lower level data in all shapes and sizes."
David currently serves as the Head of Data at Kumospace, a virtual office product.
"People are tired of having to hop onto multiple screens just to be able to talk to their colleagues. What Kumospace provides is easing the friction and fun collaboration. Think of folks who are working from home or by themselves in a different city; and our platform allows them to feel like they're back at the office."
Being the first data hire at Kumospace, David says he was fortunate enough to work for a company that already looked at data and analytics as the key aspect of the business:
"I was fortunate to join a team that had the foresight to know that they wanted to hire a data person fairly early in the process; and they invested significant resources in building out what we analysts call analytics instrumentation."
Because of this, Kumospace was already generating interesting data before they had David on board:
"When I joined earlier this year, I already had a lot of fun stuff to work with. We were able to reduce the time it took us to understand what are the most popular parts of the product, what things we can improve, and the user journey when someone wants to evaluate the product.
That's why I say I was fortunate: We knew that our top priority when having all this incredible data was to make the user experience even better. And this is where we settled on launching and data activation."
What is data activation?
"Data activation is simply enabling data that traditionally is kept in a data warehouse. We allow the data to be used in any part of the business that benefits from it."
In a traditional analytics model, the raw data is generated by the product (e.g. your website or SaaS product). A data warehouse then transforms the data to generate useful indicators, metrics, and hopefully a lot of insights. The journey ends here.
But with newer tools and modern approaches, we're now able to maximize that data by sending it to other platforms:
"For example, we may know in our data warehouse that a user tried to use a certain feature, in the case of our virtual offices such as changing the layout of their office. With data activation, we can detect if our users are struggling to set up a certain layout in their office, and then our customer success team can reach out to them based on that information."
But who decides what to do?
At Kumospace, the ownership of initiating the data activation is given to the head of product:
"We learned early on, especially for projects that directly affect user experience, that we benefit from having a centralized vision. We recently gave the ownership of this project to our head of product because there are multiple teams that want to use tools like this.
Being the head of data, my role is to support them. They are the creative masterminds of campaigns, copy, design, and I of course support them with data. But someone needs to be able to step back and plan the vision for the entire journey. This is where a head of product comes in."
The product team determines the interactions, meaning who engages with the user, how, what frequency, and what message. The other teams then work together to get that vision coming to fruition.
Technical implementation of data activation
What does the technical implementation look like? It involves two general steps.
Data transformation
David shares that a part of the reason a data person generally warms up to the idea of activation is that the logic responsible for the transformation, calculating KPIs and other metrics, lives inside the data warehouse.
Kumospace uses dbt logic for this purpose so in case they make a mistake, they can quickly roll it back and the logic stays in the data warehouse:
"In my case, if I'm the responsible person to determine how many minutes a user was present here or there, I want to make sure 200% of the data is accurate. So I do that by keeping the logic inside the data warehouse."
Curation
After checking if the data is accurate, they will then decide which data should be sent to other tools on their tech stack:
"It's not only about opening the faucet and letting all the data flow because there are cost implications to that. We have to think about efficiency because we also have constraints. So by curating, we can make sure that only the highest quality data and the most potentially impactful data is sent to run."
Segmentation for data activation
Segmentation plays a huge role at Kumospace because their users have different use cases.
"We offer the possibility of working full-time from Kumospace. Some companies use us for events, and we also have a very friendly, free offering for friends and family who just want to stay connected. Correctly identifying the use case can make a large difference in the success of any campaign."
This use case information is utilized in the data warehouse calculations:
"If I'm thinking of a campaign that I want to communicate with the sponsor or the office manager of a company wanting to set up the virtual office, I want to make sure they have the right use case, the right segment, the right persona, and that guides me when determining what data to make available."
Identifying critical lifecycle stages
"Another thing we discovered was evolving and identifying our critical moments in the life of a user. It's realizing that since we are predominantly a sync platform, we can be more agile if we talk to each other more often instead of only sending an offline message — and that's our pitch.
We also learned that when you get to know our product (early evaluation stage), it makes a big difference if you try to evaluate Kumospace on your own or invite friends, colleagues, or someone to check it out with you."
Through segmentation and data activation, they were able to come up with an effective message for users who are in the early stages of using their product:
"We can tell users: 'Hey, you're checking out, Kumospace. Invite some folks!' That way you get to experience what we call not only productivity features, but also the joy features such as listening to music together, eating a donut, or pouring yourself a drink."
David's personal approach to analytics
When it comes to data analytics, there are two schools of thought: one likes to measure everything (e.g. measuring the smallest mouse movement to clicks) and the other looks at the high-level metrics.
David believes in looking at the high-level stuff in Kumospace:
"We believe in identifying intent and success. It definitely means a big eye on reliability so we've invested in making the experience even better, and we have dedicated engineers to make the product better in every delivery cycle.
We don't do super granular tracking because we believe there's so much information to gather from intent. For example, with the features I described earlier, are you trying to work, share documents and links, or do you simply need a break? It's that healthy combination of both that, in my opinion, is the definition of success."
Business goals and product usage metrics
David shares that they're looking at the time spent on Kumospace because it's the most reliable indicator that the user will be coming back tomorrow:
"One of the ways that we're trying to fight Zoom fatigue is not having to switch between four different platforms per hour. So when you are presented with a single screen where you're able to do most of your work and interact with your colleagues, then that's a strong indicator. So it's time and it's increasing usage of the features."
Correlation and causation
While there will always be a natural battle between correlation and causation, David says the best way to settle this would be to pursue a test based on those strong inflection points.
"We do get a lot of inspiration from seeing strong inflection points in certain metrics that are associated or are directly correlated with an increase in retention. However, I always include the healthy disclaimer when sharing that because I think that when a company is small enough to move fast, there's a gain in pursuing a test based on those findings."
For example, if a user is showing a much stronger retention when the usage of a certain feature increases from X to Y. If David's team sees a strong enough improvement in retention, then they go ahead and attempt to test out their hypothesis through a campaign (such as a feature change) rather than do a prediction model.
While this is in no way saying that he's against doing models, David just prefers to do something that is more actionable.
"I would love to spend more time doing that, but that's when I think a lot of data scientists allow their heart to win just because we would like to deploy a more sophisticated tool. At the end of the day, the tool has to remain a tool.
I can do the models that I want to build on the weekend if I need to, but I feel better if I leave for the weekend having shared an actionable suggestion with my product team which includes evidence that certain feature usage has a strong inflection line at a certain point."
The cupcake approach
What do you do when product managers' research and feature requests say one thing and then the data says the other thing?
It's definitely a challenging situation and David believes that at this point you have to treat anyone's time as a portfolio:
"You have to allocate some time to more conservative projects, but you also have to allow for some percentage for more aggressive VC-type investments.
These problems are the ones that may not be backed by science, but you're able to gamble to some degree. The reality is, there are a lot of product minded people with strong gut and intuition and they may be worth it."
But gambling on these projects doesn't mean assigning people to work on something that isn't backed with preliminary data at the minimum. David advises making a counteroffer on it with the cupcake approach:
"Ask them, 'what's the smallest cupcake you could do if I let you do it?'"
Collaboration between data, product, and marketing teams
"We want folks to see these platforms as available tools, but we want the vision to be unified. Therefore, because most of our KPIs and OKRs in terms of growth, product, and account growth fall on product, we decided that it was the most fair to give that ownership to the product team because it has to come together into a unified vision of the touchpoints with a user."
While the marketing team usually has a lot of needs and ideas to interact with a user on campaigns, they have to also play nicely with the product plans:
"Yes, it brings more people to the kitchen but there's no reason why we can't get in a room and think holistically about what will happen if we launch this new campaign."
And for small companies, it also makes it easier for teams to support each other:
"So if marketing wants to propose a notification to those campaigns or the user experience, they better partner with me to gather some data that supports their point and the product team will do the same. I will happily provide them with supporting data but I will also bring my own opinion."
When it comes to product marketing, the Kumospace team believes that it lies and lives on the product. Which is why there needs to be a collaboration between teams:
"I feel that as companies and departments grow larger, there's more need to put the teams together. I believe that there's a tremendous strategic value in making sure that the product and marketing teams are living wall to wall, because the messaging has to be coherent. The product team builds the features so when we market them, the product team has to have at least direct input into that process."
Final advice
Don't strive for perfection, especially if you're in an early startup.
"Focus on speed and value instead."
Do build a beautiful partnership between your research/data science and product teams.
"We stayed aligned incredibly and intimately with what the roadmap was looking like. The more an analytics stakeholder gets embedded in the team or aware of the roadmap, the more we can anticipate."
Thanks for listening! If you found the episode useful, please spread the word on Twitter mentioning @userlist, or leave us a review on iTunes.
