Know Your Customer and Their Problem First
My ClearBrain co-founder Bilal Mahmood and I made customer-led development the bedrock of how we built the ClearBrain predictive analytics platform from day one. Most other tech companies, big and small, put the customer as the last step in their development process, but we put them first. Other companies start with a hypothesis, build, and then, as a final step, get customer input.
At ClearBrain, we started with customer conversations to help us identify problems and prioritize them based on their business strategy. Then, we developed a hypothesis, and only after we centered around the customer needs we built it. Then, we implemented a continual process of iteration, always focusing on the customer needs at the center of our strategy and decisions.
We went into our initial customer conversations with a roster of well-considered questions covering the space we were exploring. Our goal was reaching alignment with the customer:
- Is there a pressing problem they consider strategically essential to solve?
- Are they experiencing significant cost challenges or roadblocks with current solutions?
We would proceed in formulating a solution only if these lines of questioning checked out.
Our first qualified customer The Skimm, an upstart media company, had a highly technical growth analytics and data leader who was hands-on with cultivating a rapidly growing subscribership and acutely aware of the technical challenges. He had already applied customer segmentation analytics (e.g., RFM analysis) to intelligently target customers for paid subscription messaging, promotions, etc. Approaching customer qualification strategically was critical to his company’s success.
The importance of customer segmentation was highlighted by active ongoing work, support from executive leaders, existing unsatisfactory manual solutions, and significant expenditure of effort and capital to improve these solutions. In fact, their team was considering implementing their own applied Machine Learning (ML) solution to scale customer engagement.
Building their solution after we qualified their need allowed us to naturally evolve the startup across two other dimensions: team size and investment.
- Team size—By securing an upfront income stream, we gained confidence that our company could support the employment of a small team and grow it organically over time.
- Investment—By securing engaged customers and building a team before accepting an investment, we could easily convince investors that the idea and team were worth investing in.
This best-practice approach for fundraising is contrasted with struggles I experienced firsthand at other startups where we built the product first and had little user engagement to show during fundraising. Or where we got investment up front, then built a team before the product, only to find our founding team couldn’t work together effectively.
At ClearBrain, we conducted customer discovery to find a real problem we could solve, and then we built our offering for those customers and got them to pay for our software. We did this all before we raised any venture capital. The best time to raise is usually when you’re getting outside attention.
For example, the first TechCrunch interview we received coincided with a new product launch that included a self-service sign-up flow, new causal analytics functionality, and a nearly completed major interface re-design. These two events (a TechCrunch interview and product launch) made for a high-profile fundraising campaign for our company.
Times like these require thinking outside the box to maximize the opportunity. For our team, this meant rushing development leading up to the TechCrunch article. Our team accelerated the development of self-service signup. It instrumented our app with session recording, so we could track users' real-time progression from the article to the website to the app and watch the new onboarding flows to correct issues that arose quickly—and arise they did!
As a founder, when you speak to investors, stress how you are constantly looking at how users engage with your product historically and currently so you better understand what to focus on for tomorrow and beyond.
Start with key insights, build outward
The first version of our product was analytically sound, but the infrastructure was extremely nascent, and the user interface was barely interactive. Fast forward seven years to today, we now run thousands of ML pipelines and tens of millions of customer predictions daily on this same platform at Amplitude.
The pipelines are robust, scalable, well-monitored, and provide tailored, cost-effective models for each individual customer. It didn’t start that way: Our first iteration of the product felt like a barely animated slide deck.
The team we initially assembled at ClearBrain skewed heavily toward deep technical talent. As an engineering-heavy team, we fought our tendency to invest heavily in a solid infrastructure to support a pre-supposed platform. Instead, we focused on the critical insight that our customer needed to solve the problem at hand: A single numeric score per customer on their likelihood to become a future paid subscriber.
We eschewed pipeline automation to run pipelines on demand for each campaign manually. For the first run, we manually cut and pasted results into a database to drive a simple, read-only UI that foreshadowed what we knew one day would be an interactive experience. The goal was to maximize the speed of iteration and immediately show value and responsiveness to our initial customer group.
That barely automated slide deck we started with evolved rapidly as we held calls with customer development partners. The complexity of the needed solution grew organically as each investment of effort was supported by the knowledge that our customer base would use the feature.
Weeks later, we built an automated pipeline, and it was a large engineering effort; however, by delaying the build until we had customers on the system, we actually knew what to build because we already had answers to many existential questions:
- How often are campaigns run?
- Which cloud provider are we ingesting data from?
- Did the ML model we initially built run scalably with this dataset, or do we need to update it?
- What customer data edge cases are causing data pipeline failures?
Months later, we built integrations to Facebook and Google, and automated campaign audience refreshes with only those users likely to open the email. Thus, before building these features, we had answered:
- Which integrations will our customer base need?
- Which of the many APIs should we use to integrate with their existing tools?
Months after that, we built a raw event ingestion subsystem to support a publicly traded travel and lodging customer who was gathering website interactions to augment customer behavior already captured in their mobile app. Back in 2017, advanced neural network modeling was a costly project, so it came only later when we had a more extensive set of customers, and the benefit across them all clearly justified the investment (also, costs came down in the meantime).
Building from the vital insight outward ensures an efficient allocation of limited resources and allows you to prove to your customers why they were right to choose you as a partner. It allows you to listen and rapidly respond to customer needs and hopes.
Of course, this enthusiastic customer-led trajectory will be naturally moderated by your judgment of the value to other customers on the platform and a strong hypothesis that you will evolve over time.
Join the Amplitude for Startups community today!
If you enjoyed this blog article and would like to learn more from startup experts like Eric, join the #startups channel in Cohort, our Amplitude Slack community. Anyone interested in accelerating their product journey can gain free premium access to Amplitude for one year by applying for our startup scholarship program.
