Atlas AI Part 2: A genius pivot, we must make
November 2021 - December 2022
This is the second post in a series about our failed startup, Atlas AI. If you haven’t already read the previous posts, you can find them at:
Everyone’s got a Gantt Chart ‘til they get punched in the face
A genius pivot, we must make
Context: The Lockdown Menace
In October 2020, we were going door-to-door around gyms in London, pitching our AI-powered gym equipment and facing wall-to-wall rejection. Enter Bill, a tech-savvy gym manager. He too rejected our pitch, but was interested in using our AI for his home workout startup. We declined, still fixated on our initial target: gyms.
A New Hope (November 2020 - March 2021)
In November, Covid shut gyms again for what seemed likely to be the next six months. By this point, even we weren’t stubborn enough to keep sticking doggedly to the original plan. We tore up our Gantt chart and reached out to Bill to learn more about his home workout startup, ZoomFit.
ZoomFit ran live video workouts that matched a coach with up to twelve trainees. Coaches struggled to monitor everyone at once: watching one person over video was challenging, but twelve simultaneously was impossible.
Bill saw that our AI could identify which trainees needed help with their form, and highlight them to the coach. This would supercharge the coach, giving them the power to monitor all trainees as if they were one-to-one sessions.
We were slightly taken aback when Bill asked about our commercial model - no one had ever offered to pay us for anything before! We proposed to integrate a modified version of our AI into ZoomFit, in exchange for ~£7,000/month. This would only require our AI, so we could ditch the camera hardware that we tried to sell to gyms and it would be a much simpler product. We signed a four-month contract, from December to March, to repackage our AI for ZoomFit. This was broadly successful and spurred us on to continue with the new direction.
The Rise of Atlas (April - August 2021)
We’d made a strong start, but our painful experience the year before had taught us hard lessons about risk-based planning [link]. What if ZoomFit was an outlier, and no other company wanted AI-enhanced home workout classes? We decided to test whether there was demand from other home workout companies, too.
We set ourselves a goal to sign deals with three clients by the end of 2021. We hit this target by May, agreeing on contracts with two more UK-based startups. While each client was slightly different, they were all using the same AI as ZoomFit for analysing home workout classes. ZoomFit extended our contract for another four months, and even included a slide about our AI in their pitch deck to investors!
In July 2021, we were accepted to a prestigious US startup program called Techstars. This 3-month “accelerator” was like a cross between an MBA-for-founders and a $120k venture capital investment. We were riding high, having:
Signed up three clients for our proprietary AI
Made a profit for a quarter, which is rare for tech startups
More cash in the bank than when we first secured investment in 2019
We even felt grateful that Covid shut the gyms, as it pushed us into doing something that was now looking like a real business! And yet, we had a nagging feeling that something was wrong. But you always feel that way, at least a bit, when running a startup…right?
The Market Strikes Back (September - December 2021)
In September 2021, we started the Techstars accelerator. With three clients and our advanced technology, we were ready to “accelerate”. We dreamt about tripling our client base and raising millions using the Techstars network, which included thousands of tech companies and investors.
As the program began, we were matched with veteran startup mentors for weekly sessions. They began by asking some basic questions about our business:
Who’s your customer?
People running home workout start-ups!
What problem are you solving for them?
They can’t build this clever AI themselves!
How are they measuring the value to their business? Is it reducing their costs or increasing their revenues?
Erm…
Is anyone doing home workouts actually using your AI?
Errrrrrrm……..
Despite Atlas looking strong on the surface, we didn’t know why our clients were paying us. After some in-depth conversations, we learned our clients wanted AI-powered classes in the hope that investors would fund their startups. There were many competitors providing home video workouts and our clients needed something to stand out, so they saw our AI as a tool to help secure investment.
Our mentors told us this was short-term thinking, rather than the foundations of a strong business, and predicted that our clients would cancel their contracts. We ignored them. After all: our new, revenue-generating business plan was genius!
We quickly changed our minds when two of our three clients asked to cancel their contracts within a month. We were reluctant to alter our business model, again, but we realised that we couldn’t keep signing up clients for software that wasn’t providing value. Our “genius” idea was doomed.
The Techstars program ends with demo day, where startups pitch their idea to a live audience of hundreds of investors. We spent the final few weeks scrambling to find a new idea that we could take to the stage. When friends told us they’d use our AI to check their form during video workouts on YouTube, we realised we could provide our AI tool to people via a laptop app.
We ended the Techstars program with a coherent vision that we pitched on demo day. This was a great experience and we received many kudos for our ambitious plans. However, we didn’t have any users or a product. Behind the scenes, we were essentially back to where we were a year ago…square one.
You can read about our next move in part 3.
What we learned
We thought taking the solution we’d built for gyms and selling it to home workout platforms was genius, because we wouldn’t have to build a new product from scratch. However, we had become a “solution in search of a problem”. We didn’t understand our new customers well and our AI didn’t solve their main problem: securing money from investors. We will write a longer post about how to make better pivots.
A more subtle lesson: customers paying doesn’t always mean you’re providing value to them. We avoided digging too far into exactly why our customers were paying - we were just happy to finally be making some money! This is particularly risky when selling a hyped technology, like AI. Customers may pay to be seen as innovators, even if the technology adds no tangible value. Without Techstars’ help, we’d have taken much longer to recognise our error.
Special thanks to Dr. Tishtrya Mehta for her detailed review and comments, which greatly improved this post.