The Road Less Traveled
Back in December 2023, the startup I was working at called Armory was acquired, and I did not join the new parent company. I reflected on the fantastic team I worked with and our efforts to find product-market fit (PMF) with a second B2B SaaS product, and how we ran out of runway before we achieved the goal.
As I considered what I wanted to do next, I saw two insights on the chessboard.
- 80% of software features are rarely or never used, costing companies $29.5B annually in wasted development. The root cause? Teams can only validate a fraction of their ideas before making build decisions.
- Recent Microsoft Research shows that AI can effectively predict human market behavior and decision patterns. This breakthrough means we can simulate and test product ideas earlier, faster, and more thoroughly than ever before.
Through my experience using product discovery frameworks, habits, and tools, I realized that most idea validation was just as much about predicting human behaviour and UX nuance as it was about competitive analysis and product strategy. The emerging evidence that generative AI could unlock new technical solutions to the psychology and UX side of the validation problem was inspiring, but it only took a couple hackathon weekends to realize that while the technology definitely unlocked many new opportunities, no one would be able to magically automate the validation process with it. Real research with customers would need to remain a key part of the equation.
I knew better than to make a solution in search of a problem, so I set out to validate the problem through dozens of good old fashioned customer interviews. Unsurprisingly, I discovered that the problem was a worthy pain to solve, but there was skepticism about whether a SaaS solution could make a meaningful difference. Most of the tools we use are more focused on the result of discovery than the hard work of insights, UX, design, and so forth. They’re brilliant communication and organization tools, but I always wished there was more they could do to help me do the job instead of just communicating what happened after I did the job.
After seeing some encouraging initial validation, I joined forces with an old partner in crime, Geoff Smith, who jumped in as co-founder and CTO! We spent months interviewing, prototyping, discussing with customers, iterating, and throwing tons of code in the garbage. It was a profoundly rapid learning experience, and the AI space was leaping forward almost daily as we went, until we arrived at what Revelica is today. Thank you to the many product, engineering, and design people who already helped us. You gave so generously of your time and creativity, and that is why I feel so proud to bring Revelica into early access now. Together, we developed a vision for the future of discovery and validation, and how it can be uniquely approached with AI. It’s going to change the game.
The Revelica Method
Let's zoom into a real team: you have ten great ideas. You have time to prototype and test maybe two before cycle planning. You pick the two that "feel right" based on experience. But what if the game-changing innovation was idea number seven? Or what if you didn't even think of the game-changing idea because you were busy with operations work?
That's why we built Revelica. Not to replace customer interviews. They're still the foundation of great product discovery. Instead, we built a product agent that runs the discovery work fast enough to keep up with coding agents on the build side. The Method is the framework the agent runs. Four phases, each operationalized by skills and playbooks:
- Strategy. Model the competitive landscape, value propositions, and customer segments to identify differentiation gaps. The strategic context the rest of the discovery work draws from.
- Customer Insights. Process interview snapshots, support tickets, and observed behavior to automatically extract opportunities and pain points. With cited evidence the agent refuses to fabricate around.
- Solution Validation. Convert ideas into user story maps to test and surface assumptions before writing code. Compare multiple solutions against each other on the evidence, not on confidence.
- Delivery. Export validated, structured specifications into formats AI coding agents can use for execution. The product agent and your coding agent hand off work without a human re-explaining the customer at every step.
This isn't about replacing human judgment with AI. It's about giving product teams superpowers to explore more possibilities while staying grounded in real customer understanding. Continuous discovery is the best practice we believe in; the Method makes it practical at the pace AI development now sets. Experience has shown that the first idea is rarely the best one when it comes to delivering outcomes.
See the four phases in detail in The Revelica Method hub, and try the product agent free at app.revelica.com.
Long Story Long
If you’re interested in the meandering road, dear reader, I will delve into the back story of our journey. If you prefer to take a shortcut and check out the product, skip to the The Product.
We interviewed dozens of product innovators across the product, engineering, design, and leadership functions, many of whom were former colleagues and asked them to tell their story of building something that didn’t meet business expectations when it launched in the market. Usually when you do a customer interview, there is an awkward moment where people are trying to think of an answer to your question. But in this case, it was different: they had so many answers immediately available that they had to take a moment to pick the story they wanted to tell most. This definitely confirmed the numbers I have always read that say 80% of software products are rarely or never used by their intended customer.
Through all these interviews, I realized that I could use my experience and judgement in product development to guess at the cause of the problem. The overall pattern I noticed is that most of the teams involved in the stories were quite intelligent, making a good effort to think of their customer problems, and delivered a creative solution. The devil was in the details, little assumptions were often made about usability or value that turned into big problems when it came time to capture value with the solution. I started to see how we might train an AI to have this product sense in a predictive way.
The assumptions were typically specific to the target customer, for example, a certain group of companies not adopting a desirable solution they asked for because it was too difficult to integrate into their current process and tool stack. Another example told the tale of a mismatch between the intended user and the technical depth of the product. Engineers may insist on API products, but other customer segments can’t use them at all. Another story dealt with the difficulty of launching a marketplace because there was no demand to entice the supply to join in the start. Teams are often naturally skilled at attracting one side of a marketplace, but struggle to do the same for the other, especially at the start when there is a chicken and egg problem to contend with. One way of looking at marketplaces is that you need to make two customers happy in unique ways that your business can solve for. Often each of the customer segments have wildly different problems, so it can feel like building two products at the same time. Validating these sorts of things in advance is time consuming and requires a good mental model of the market dynamics involved.
After telling their respective tales, the interview participants often went on to discuss what happened next. There were some trends that emerged in this part of the conversation:
- Leadership was often unwilling to pivot or iterate to address problems with the solution, instead they would abandon it for a new initiative
- There were often alternative solution ideas the team didn't choose to build that would not have met the same fate
- The risks inherent in solutions become clear after you launch them, but building MVPs is very time consuming, and you don’t get many kicks at the can
- The problems that led to failure could be traced to risky assumptions made about the customers, the company, or the technology used to create the solution
It felt like the ideal way to validate just wasn’t remotely compatible with the reality in most teams, and the failure causes were often subtle. We are after all, smart teams capable of amazing creativity! The phrase "right problem flawed solution" came up a lot. It was clear that we had to help teams create, test, and compare novel ideas 10x faster to have a big impact on this problem. Fortunately, the enabling technology has finally arrived, and we’ve experimented at great length to discover good solutions.
The Product
Geoff and I got tired of making things people don't use, and decided we have just enough passion, experience, and tenacity to lead the transformation of product discovery with Revelica, so we can all enjoy that magic feeling of shipping a winner more often.
Today, Revelica is organized around the product agent. An AI teammate that runs the Method's four phases through a library of skills (single-step actions like running a pre-mortem or extracting opportunities from an interview transcript) and playbooks (multi-step chains like bootstrapping a new workspace from a company URL). The agent works alongside you, citing evidence on every customer claim and producing structured artifacts your team and your coding agents can both work from.
If this resonates, get started with Revelica free and begin transforming your product discovery process today. We're too busy building cool stuff to bother spamming you with marketing. Want to discuss enterprise needs or share how your team tackles product discovery? Book a meeting with us.
Now, let's beat the odds and build more winners together.
See you out there,
Randall
Founder & CEO, Revelica
