In recent years, we’ve witnessed the meteoric success of product-led growth (PLG) and bottom up selling motions in the enterprise software market. Since we strive to have a finger on the pulse of the data world, it made us wonder: Would PLG find success within the data space? To answer this question, we assembled a unique forum for our Q2 ElevateData event. We had the usual suspects: Leading Chief Data Officers (CDOs) from companies like Amazon Music, Earnest, Walmart, Frontdoor, Freedom Financial Group, AWS and Poshmark. To enrich the debate, we also hosted CEOs of emerging data startups like Monte Carlo, Streamlit, Mozart Data, Unravel, Nexla, and Acryl Data. Having this diverse group in one Zoom room led to a hearty conversation, exploring PLG in data and different go-to-market approaches. The discussion largely focused on the CDO perspective on PLG, specifically the dynamics that Vendors should be aware of within the data org and how to navigate them. 

Before we dive into the learnings, let’s unpack product-led growth. It is the concept that the product adoption by users starts the sales cycle and is the primary driver of growth. It changes  the strategies for customers acquisition, retention, and expansion. Zoom and Slack are prime examples of B2B productivity companies that were built on the foundation of PLG. Other examples include HashiCorp and Auth0, from the engineering world. These and other companies have been able to drive interest and value for the user, which in turn helps to drive value and ROI for the beneficiary/buyer of the organization. And indeed, companies that adopt a PLG approach generally supplement it with direct Enterprise sales. While PLG has worked well for collaboration and engineering, will it work for products that are critical for data infrastructure? As our own Barkha Saxena, CDO of Poshmark, pointed out, “as a data person, I look at collaboration tools very differently than tools that data flows through.” Let’s now explore what the group discussed:

Data leaders talked benefits of PLG:

• Adoption in the data org: PLG allows data practitioners to educate themselves and choose their own solutions – through trials and open-source tools. According to Swaroop Jagadish Co-Founder of Acryl Data, the modern data engineers demand consumer-grade delight from products they use. Leveraging this model, the organization can evaluate and purchase tools with a higher degree of confidence that they will be adopted!
• Champions for the vendor: Successful bottom-up companies gain zealous supporters of their product in the organization, and they spread it from there. This creates a community of champions that carries over when people switch jobs and bring their tool set with them.
  

And then the CDOs outlined their considerations for Vendors to be aware of:

• Trenches vs. 60k feet: Shwetank Kumar, CDO at Freedom Financial Network, articulated one of the dynamics that PLG creates within the data organization: “This can become very adversarial at times. You might have a reasonable product in one domain, or a great product for a specific use, but there is a larger organization strategy around data infrastructure, analytics, and tools.” Vendors that take a PLG approach should not only understand how their product can help the end-user but also how it fits into the broader data strategy and stack. One of the best ways to accomplish this is by establishing relationships across the data org. If not, it could lead to an adversarial relationship that is counterproductive to the goals of each party! To see bottom-up success, it is critical that the user, beneficiary, and buyer are as close and aligned as possible.
  
• The proliferation of data tools: At our last in-person ElevateData event, one CDO from a Global 100 company shared that there were two to three times as many seats for data tools as there were data analysts, scientists, and engineers. While this may be an extreme case, it is important for the organization to not lose sight of the management or standardization elements. More tools can create more flexibility, but at the expense of cost and maintenance.

• Trust and consistency in Data across the Org: Any data tool which can lead to data discrepancies and bottom line impact on data quality at any level will eventually become obsolete in organizations striving for high data integrity and consistency. This is why it is critical to understand organizational dynamics, key influencers and overall data strategy to ensure right partnership as PLG companies work on expanding the adoption of the product through complimentary enterprise selling.

• Infosec, compliance, and data privacy: When adopting new data tools the users need to make sure it is not creating security holes for the company’s security posture, nor does it violate any privacy / data sharing regulations. This of course becomes a critical factor in deal success in larger organizations. To mitigate that, many PLG data tools should consider an “on-premise” deployment model or minimize the type of information collected. In addition, the leaders agreed that SOC 2 is a must no matter how small or big the vendor is. It is also interesting to consider that the majority of tools are getting pushed out early on, with the Vendor not even knowing they were evaluated. So being stale on security, reliability, trust, etc. is a disadvantage.
  

The group then discussed how successful GTM could look like:

• Success lies in finding the middle ground between different stakeholders and balancing motions. For example, Barr Moses of Monte Carlo noted there might be a combination of community-led and direct sales in a motion she describes as “customer-led” which emphasizes the need for thoughtful POC. Bringing elements from both strategies to allow practitioners to choose their solutions, educate themselves, try it out with a trial or open source. This convergence of approaches may address many of the concerns outlined by our CDOs.
• It was also noted that the next wave of successful companies will need to figure out how to be good citizens of the ecosystem. It’s time to think about interoperability and weave in the concerns of data platform teams earlier on. If you’re not placing yourself as a citizen in the open ecosystem, you’ll do yourself a disservice
• Another interesting point was made by Saket Saurabh of Nexla: There’s an awareness continuum, and somewhere in that, you want to try the product. And somewhere there is a stop where you need approval. Question is whether the goal of PLG is to get you in or end-to-end.
• Our CDOs provided advice on converting users to customers: It’s key to find validation of use cases to a team (beyond the individual). Next part is what is the nature of these integrations, and making sure you can answer questions on compliance and security early on. The central data teams will likely pay if it’s something that solves 2-3 use cases well for a team over some time.

And wrapped up by talking about implications of bottom-up adoption of business facing data tools on the Data org:

• Bottom-up adoption for business teams facing tools meets a need for a business user that the data team isn’t addressing with their current capacity – there is a gap in bandwidth and the data team cannot work through all the different use cases and the pressing business desire to move fast. This gap resembles the proliferation of Shadow IT tools, which eroded the dominance of the IT org
• There are so many parts of an org that use data in some way, so you have these pockets of opportunities for bottoms-up in every part of the org which has some data competency. This will be value added to the organization as it increases ROI from data. 
• But the CDOs don’t think the data teams are at risk of being substituted anytime soon: There will always be a need for a central data team and governance which will provide the foundation for further value addition by PLG companies.

As we reflect on our discussion, it is clear that just like other sectors, enterprise sales will need to compliment PLG in data sector as well. While the path from bottom-up adoption to Enterprise sales is not straight forward and would require investment and relationship building as in typical enterprise sales, PLG would provide an advantage by having internal champions in and outside the data orgs, eliminating/substantially speeding up POC processes and enhancing widespread adoption new technologies.  

——-

Thank you to all of the ElevateData members for joining us and passionately exchanging your ideas, as well as to the inspiring founders who joined us for the event! If you’re a data leader and are interested in joining ElevateData, please reach out to one of us! 

Your ElevateData Founders,

Barkha, Oren, and Alex

Last week, we hosted an ElevateData dinner, the last 2019 gathering in the series that unites Data Leaders to delve into hop topics in the industry. Just in time for driving New Year’s resolutions, we focused on a headline-generating buzzword and a top-of-mind topic for businesses: Machine Learning (ML). Given the continued commotion around this topic, we wanted to foster a reality check on the progress and potential of Machine Learning with a group of data leaders from different industries and companies. The result was a wonderfully insightful and lively conversation. Here’s what we’ve found:

Machine Learning is still at the beginning of its journey

When asking the data leaders how they define Machine Learning, it was clear that there is still variance in how data science vs. ML vs. AI are defined. Yet, it was widely agreed that we’ve just started to scratch the surface of the potential of ML. It’s interesting to note that the group overwhelmingly agreed that not every business problem requires ML and it is not perceived as a magic solution to throw at any problem. In fact, there are so many business problems across industries where the right solutions are data-related but not ML-driven, like reporting/dashboards, forecasting, inference, experimentation, etc.

Current impediments to Machine Learning

The discussion then focused on what’s standing in the way of Machine Learning delivering on  its promise. Here are the four major obstacles that were identified:

• Lack of organizational readiness: It seems that many companies and industries haven’t yet reached the maturity level in data infrastructure and optimization/efficiency needs warranting investment in ML. Clear frameworks for measuring the impact need to be defined, tested and implemented. There is also the human factor: in many scenarios, there is  friction around adopting ML, as intelligence and automation delivered by ML are perceived by many as a black box creating loss of control, as well as driving job loss. The group agreed that leadership is critical to ML success. We’ve observed that organizations where Data Science directly reports into the CEO (like Stitch Fix) often see their data endeavors propel.

• Technology deficiencies: Technology needed to make ML ubiquitous is still in the development phase. This is true for both development and deployment technologies and is leading to inefficiencies in ML solutions production cycle. Many companies are making progress in that direction but we are still far from having a full end-to-end ML solution.

• Continued ML expertise gap: Despite an abundance of training programs, we continue to have a shortage of true ML expertise which is a combination of multiple skill sets including but not limited to programming skills, full understanding of algorithms’ inner workings, expertise in statistical concepts, and business acumen.

• Next wave of innovation not enabled: Another factor inhibiting growth is skewed perception of where ML is often considered for optimization and increasing efficiency as opposed to something with a promise to deliver true innovation.

Machine Learning in 2020 and Beyond

After discussing the obstacles standing in the way of ML, the data leaders shared their perspective on driving forces which will help scale ML to next level in 2020 and beyond. Here are the five key takeaways:

• Moving beyond the buzzword: It seems that the ML overhype is stabilizing. With growing awareness around hype vs. reality, just adding “ML” to every pitch and story is no longer creating the perception of the proposed data product being extra valuable. We are getting better at asking questions to validate the how, what and when about ML. This will help build credibility for ML application.

• Education and messaging to build real belief in ML: To help ML realize its full potential, data leaders will continue to engage in demystifying machine learning so that it is not considered to be a black box. This will require using the communication tools and strategies tailored to specific situations and org structures. By reducing uncertainty associated with ML solutions, we can truly help unlock the full potential of machine learning. Data leaders will continue to work on positioning ML-driven solutions as enablers for people to do better at their jobs to mitigate the fear of job loss.

• Focusing on broader data skills in hiring: Industry players are starting to think more holistically about data skills, with a growing acceptance that all data skills are valuable. Data leaders will have proactive conversations on this topic, reenforcing the message that analytical and statistical skills combined with business domain knowledge are  critical to building credible ML solutions.

• Technology advancement: Through continued investment in ML platform development, some players will emerge with affordable and comprehensive technologies. That would likely speed up the adoption of ML but as development takes off, we would need to remember that ML systems have deeper technical debt challenge compared to software engineering. Therefore, it will be critical to set up the right processes and shape perspectives to ensure sustainable development.

• Data ethics evolution: ML can lead to negative outcomes due to poor quality data, biases in the modeling, and more. A recent example is the Apple Credit Card controversy, with reports that the algorithm was discriminating against women. There are other scenarios where simple lack of judgment or not thinking through the broader impact of ML can lead to negative consequences. The group wondered if the role of a Chief Ethics Officer will emerge to focus on the ethical usage of data, and it expects an ethics discussion around data usage to become more prominent.

Our guests: Thank you to our guests from Unity, Snowflake, Stitch Fix, Uber, KeepTruckin, Shipbob, ZScaler, Streamlit, SAP Aruba, and Ethos for joining us for this dinner.

Thank you to all of the ElevateData members for joining us and passionately exchanging your ideas, ElevateData would not be what it is without all of you! We can’t wait for what 2020 has in store.

Let us know if you’d like to join this amazing group at the next ElevateData dinner!

Your ElevateData Founders,

Alex, Barkha and Oren