How To Do-Your-Own Research For Your Startup
We were on the third floor of the Rotman building. Students, teachers, and entrepreneur darted in and out of the presentation hall hosting a Creative Destruction Lab session just across from where we sat. Robbie had just presented his startup’s latest progress on data analysis from IoT devices using Quantum computing algorithms. But instead of his work, we were talking about mine. I described to him how my fitness startup’s IoT sensors could work. We wanted to put sensors on weight equipment to help gyms and members track progress. Aman, Alberto — my partners — and I wanted to put a sensor on each weight plate then track them as a swarm. We thought it the cheapest and easiest solution.
“You know if you put something on the tension line instead of each individual weight on the weight machine, you would have a much simpler solution,” Robbie said.
“Agree, but what kind of sensor will that take? We would have to cut the line and put it on there. That’s way more work,” I said.
“You could use a pancake load cell.” Robbie replied.
“True, we could put it on the bottom of the rack, then we track how much weight isn’t there, but we lose motion as the weight plates go up.” I said.
“So use one accelerometer and one pancake load and combine them.”
We went back and forth like this for a few hours, then went to a bar. I took some notes and did some fact finding. Little did I know that if I had just put a horizontal pressure sensor on the vertical line, we could do some math and get weight from a pressure sensor and motion from one accelerometer with just a crimp on the weight line. This is what Shapelog does. I would learn about Shapelog months after the conversation with Robbie. Robbie was right; we were doing it very wrong. In the intervening months between that conversation at Rotman and finding out about Shapelog, I spent my time and my money working with a team out of Romania trying to get a sensor swarm to connect to a Raspberry Pi without packet loss due to collisions from too many sensors talking at the same time.
A few weeks after meeting with Robbie, I was talking to Edgar after another Creative Destruction Lab event. He and I had done our MBA together and he had an extensive IoT background.
“Paul you don’t want to deal with hardware. You start getting into different SKUs and technologies and RMAs. If you can make it with software, add the hardware later and simplify it. Plus, you’re talking thousands of sensors per gym. Why not just use a camera?” Edgar told me.
Again I had a rebuttal. Our proposal was cheap and easy and we had already started working with a solution provider. Again, months later, I found GymCam coming out of Carnegie Mellon University’s Smash Lab. GymCam solved the hardware tracking problem with cameras. I still think it’s probably the most elegant solution.
Hrvoje, my IT architect at my day job, came from Croatia to Canada on Thursday for a business presentation we were giving on Monday. Friday, we went for drinks at Flying Monkey’s Brewery in Barrie and Saturday morning we drove to Tobermory to go wreck diving. On the drive I described to him our approach on my side hustle: no sensors, just software.
“Isn’t that exactly like Moov?” Hrvoje said. “It does like 90% of what you described.”
And yet we persevered. Every step we took, we tried to get feedback from experts like Robbie and Edgar and Hrvoje. The costliest lessons were not listening to them sooner, pivoting faster, and acting on the information they provided.
When we went looking for how to do something like get a patent or if we could use no code mobile development like Buildfire, we called it research, but really it was fact finding. Patent filing required certain steps. No code mobile development had certain limitations. Research isn’t looking up the steps required for deploying a Kubernetes container on Stack Overflow. Fact finding is critical to getting anything done, but it should not be confused with actual in-the-field research. A line needs to be drawn between the peanuts poor fact finding costs versus the fundamentals research changes. Picking the wrong vendor may lead to a costly mistake, but picking the wrong solution (or wrong problem) can tank the business. Entrepreneurs need to compartmentalize tasks and be clear about what they are doing. Talking about an idea is not marketing; working on a pitch deck is not marketing. Every company needs marketing and if no marketing is being done beyond working on the pitch deck or talking about an idea, then there is a resource allocation gap that needs to be filled. The same applies to research. Looking up code snippets is not research and shouldn’t be treated as such. Research needs to be actioned at a fundamental level to validate the hypotheses that underlie the business proposition.
Research is testing a hypothesis in the real world, then choosing a new hypothesis to build on or replace the old one. Startups need to relentlessly apply the scientific method to their efforts. Decisions need to be based on facts. For our startup, Spotfit, that sought to place a sensor on every weight in a weight stack, we should have followed Robbie’s blunt fact: many sensors is more work than one sensor. We should have quickly combined that fact with Edgar’s insight: managing hardware at the level we were describing meant we would be devoting more time to hardware than to our core business: health and wellness. Instead we pursued an IoT solution for another six months before the lesson finally sunk in. We chose to ignore the research and the feedback and not revise the hypothesis. Sometimes things need to be spelled out in black and white or in our case in dollars and cents.
After our visit to Canfitpro in August 2019, we researched the competition for the next few weeks and what we found Shapelog and its pressure sensor horizontally ona taut vertical cable. Another company, This Is Beast, put the sensor on user’s wrists and just had the user enter in the weight instead of going through the rigmarole of tracking weight plates in motion. This meant any cable machine and any free weight could be retro-fitted with its user’s sensor instead of our solution that required dozens of sensors and batteries per machine. Once we abandoned our initial hypothesis, the options opened up and we found GymCam and Beast Sensor and dozens of other companies working on the same problem.
It’s easy to blame COVID-19 for the rise of misinformation, but it is only a symptom of an underlying crisis and a very human tendency to ignore information humans do not like. Business is much less forgiving of those that choose to ignore the evidence. Snake oil salesman do exist, and do find success and smart business with proven solutions also fail. There is also a wide spectrum between an evidence-based approach that creates value and snake oil. The act of gauging the business on the scale is at least an attempt to measure, an effort to justify, and an example of doing the research necessary for a business to succeed. The alternative is blind faith — not actually a good thing — or deception.
With Spotfit, we tried to make something we didn’t have the data for using AI. We had many hypotheses, but we could not prove a fair amount of them. As we progressed on the journey and delved and prodded and tore down, we found better approaches to the same problem. Competition is indicative of a growing market. And that we had competition was good evidence we were on the right track. We just needed to identify what that track was. Answers should lead to more questions and the subsequent answers should maintain a consistent result with the previous ones, that is how math works, how science works and how a successful business should work. For us, the research was telling us we were very far from having more answers than questions and none of us we ready to work on blind faith. Stopping is okay, perseverance is better, and making it is sticking it out and being prepared.