How our tiny team beat Google
“Did you see the post about Google on TechCrunch? That’s a win for you all, good stuff!”
Last week, I woke up to this text from a friend. He was referring to this recent post.
Google’s Medical AI product had failed in real life testing. The same one we had been working on at our startup for the last 3 years. I immediately forwarded it to Florian, my cofounder.
“Looks like we made the right decisions.”
Florian and I were building a product that used AI to analyse images of the human eye. Our AI would go on to assist in screening an eye condition known as Diabetic Retinopathy (DR) - a condition that, because of insufficient screening, is one of the leading causes of blindness globally.
We’d been working on our company since 2017, and almost 3 years later, we were seeing our early decisions pay off. While Google had failed to go to market, our product was being used in over a 100 different healthcare practices, a community health screening program, and trialled at a state government screening initiative. [More details here, and here]
We were doing what we originally set out to do. We were preventing blindness.
Google are doing the exact same thing. Why are you going to win?
Two years ago, we found ourselves answering that same question to every single investor. It wasn’t straightforward.
Google had begun working on the exact same problem 3 years before us. They were targeting our primary market, and had already attained the highest diagnosis accuracy anyone had ever achieved. And of course, they had a multi-million dollar budget and some of the smartest people working on the problem.
“They’re doing everything wrong”, we had answered.
I don’t know why we thought it would be easy to sell that argument. Sundar Pichai had just presented their product at Google I/O. They published their clinical results in the biggest medical journals. This looked like a flagship effort. They didn’t seem like they were slowing down.
Most investors weren’t convinced. In hindsight, I wasn’t sure whether that answer came off as confident, or arrogant. Perhaps a bit of both.
I would be lying if I told you Florian and I didn’t have our own doubts. We often discussed whether we were overlooking something. But everything we had learned in our journey told us otherwise. We trusted ourselves and our direction, and eventually built a product that users wanted.
While it’s difficult to detail the last 3 years of learnings in a 10 minute read, here is a summary of the key principles that helped us get to where we are today.
- Know your market.
- Have the right team.
- Go through the grind when you execute and build.
- Remain confident, and remain persistent.
- Be grateful for, and leverage your luck.
In this post, I’ll be diving into a bit more detail on point 1. In future posts, I’ll dive into points 2 through 5.
Disclaimer - There are no easy ways to build a business. Every decision you take, no matter how calculated, is still a bet. Building a company is a series of decisions that equate to a marathon, and not a sprint. While we may have outdone Google for now, it does not mean that things can’t change in the future with respect to them, or any other competitor.
Know your market
Most startups die for one primary reason. They build something people don’t want. In fact, it’s so much so that, Y Combinator’s slogan is literally ‘Make Something People Want’.
That isn’t a mistake made only by startups.
TechCrunch says on Google’s AI deployment:
The system’s high standards for image quality was at odds with those captured under the constraints of the clinic. This mismatch caused frustration and added work.
The clinics often experienced slower and less reliable connections. This slowed down the screening queue and limited the number of patients that were screened in a day. In one clinic, the internet went out for a period of two hours during eye screening, reducing the number of patients screened from 200 to only 100.
Nurses tried various workarounds but the inconsistency and other factors led some to advise patients against taking part in the study at all.
Google looked at the problem like how most engineers would - “How do we improve diagnosis accuracy?”
A better question to ask would’ve been - “Are we certain on the problems our users face, and how can our technical skills solve them?”
All the money in the world doesn’t mean much if you make the same mistakes that have been proven to not work. Google had built a highly accurate AI diagnostic solution, that their users did not want.
How we made something people want
If you want to be healthy, you need to eat your vegetables. If you want to build a great product, you need to speak to your users. Without immersing yourself in your users, it is impossible for a company to really understand the intricacies of their user’s problems.
The process we followed was rather simple, and it looked a little something like this.
You’d think I was exaggerating if I told you that we spoke to over a hundred of our potential users before we even begun building our product. I’m not.
Speaking to your users is something most good companies do.
- Airbnb’s founders personally visited their users’ homes.
- Tinder organised frat parties at college campuses to meet their users.
- Adora Cheung, CEO of the now defunct Homejoy (once valued at over 150 million USD), personally cleaned her customers’ homes.
While we hadn’t quite nailed our unique value proposition at the start, after over a 100 different conversations, Florian and I were able to articulate exactly what kind of product our users wanted.
We spent hours a day on phone calls, we invited them for coffees and dinners, and we attended medical conferences. We even visited hospitals to really understand how they ran their practice. With each conversation, we learned new things, and adjusted our beliefs.
By the end of it, we had a better understanding of their problems, their motivations, and the constraints under which our solution would need to operate. Google’s direction had started to feel obviously wrong.
Clinics we visited barely had a functioning wifi. And even when they did, their connections were too slow to upload images. With their patient crowds, the internet speed would make our product a bottleneck in their clinical workflow.
High image quality standards?
Many hospitals largely used trained technicians, instead of doctors, nurses, or other medical personnel, to capture patient images. One hospital even used high school drop outs for this role. Limiting our algorithm to picture perfect images just to obtain a better publishable accuracy only drastically limited its use case.
Be aware of your biases
There were many moments during our conversations where Florian and I had to face harsh, uncomfortable truths that made us question our original beliefs. And oftentimes, we nearly fell to our confirmation biases.
It is absolutely critical to not let this happen while speaking to your users. You need to have an open mind, and with any new information you get from those conversations, you need be willing to refine and adjust your original beliefs. If you choose to ignore all data that contradicts your beliefs, the exercise is futile.
As a friend eloquently put it - “I’ve seen so many people talk to their users, only to frame the feedback through their preconceived notions anyway.”
Do not let that happen. It’s perfectly okay to have the difficult conversations and face the harsh truths.
What if my competitors are doing lots of things right and I can’t differentiate myself?
A friend, who read a draft of this post, asked me that very question.
He gave me an analogy of a hypothetical company trying to build a search engine in 2020.
“Unlike your example, Google’s direction with search isn’t, as you put it, obviously wrong”
The easy answer to that question is: Nobody tends to be perfect and, with enough digging, you will find an answer. However, that’s much easier said than done.
In general there are two pieces of advice I’d give to someone in a similar situation.
- Find a niche
- Exploit a change in the world state
For every company like Levi’s, or GAP, you’ve got professional tailors making custom fit clothing. The truth is no matter how large a company gets, they simply cannot cater to everyone in the world. The more customers you speak with, the higher the likelihood that you will find someone who wants your version of the product over a competitor’s.
When you find that niche, and couple it with a growing change in the state of the world, you will find yourself an opportunity. The great thing about change is that change is always constant. And larger companies find it much harder to adapt.
A great example for this is the privacy-conscious search engine, DuckDuckGo.
While Google almost definitely is the winner at search engines today, DuckDuckGo was able to find itself an initial set of users who were very privacy conscious. While they were initially targeting a tiny niche, that niche coincided with a recent change in the world state - everyone is becoming more privacy conscious.
While growth was initially limited to the original niche, DuckDuckGo started to take off as the world continued to move in a privacy-conscious direction. To directly compete, Google would need to re-work their entire product and business model to cater to privacy conscious users, not an easy feat.
If one were to make the argument that the type of content on the internet has changed, that it’s no longer about indexed web pages, and instead about multimedia content on platforms like TikTok - it opens the opportunity to build a search engine that Google’s algorithms weren’t originally designed for.
Bottom Line - Know your customers as much as you can BEFORE building your product. While this won’t guarantee you outdoing your competition, it definitely prevents “progress” in the wrong direction.
“We may not be as accurate as Google claims to be, but we’re definitely accurate in the ways that it matters”, we often told potential investors.
That wraps up Part 1!
If you enjoyed reading this, hit that 👏 button 50 times, and share it with your friends!
Thanks to everyone who shared feedback on the drafts, and Florian for being there throughout the journey!
In the next few posts, I’ll address the remaining points, and answer questions like:
- “How did we build the right team?”
- “How did we approach product development to build a better product?”
- “Were we just lucky that our hunches turned out to be correct?”