Atlas AI Part 1: Everyone’s got a Gantt Chart ‘til they get punched in the face
June 2019 - November 2020
This post is the first in a 4-part series. You can find the other posts at:
Everyone’s got a Gantt Chart ‘til they get punched in the face
Why blog about a failed startup?
We worked on our startup, Atlas AI, for four years. While we ultimately failed to achieve commercial success, we learned a lot about startups. We learned even more about ourselves. We feel grateful for this opportunity, so the aim of these posts is to:
Document our journey, for those who know us;
Consolidate our learnings, for ourselves; and
If we’re lucky, help others with thinking about startups
Let’s get cracking!
Before the Gantt Chart (June - November 2019)
Our team have a background in AI and teaching people to lift weights. Two of us were working on AI PhDs, and we were also competing in powerlifting. Friends sent us videos of themselves lifting weights over WhatsApp and we gave advice on their form like “when squatting, your feet should be wider”. We devised a startup idea to replace our video coaching with an AI personal trainer, called Atlas. Atlas would analyse people’s videos and teach them for a fraction of the cost of a human trainer.
Our first idea was a phone app where people could record themselves lifting. They would put their phone on the ground, then lift in front of their camera. However, we knew setting up your phone in a crowded gym would be challenging. Instead, we designed some hardware to convert existing gym equipment into AI trainers. We fixed cameras to the equipment and bolted an iPad onto it so people could view the feedback from the AI coach. Users would then just lift and get instant feedback afterwards, without having to set anything up themselves as the cameras would already be part of the gym. Genius!
Gyms would pay a monthly subscription for the product, because it would improve their member experience and reduce cancellations. To check this assumption, we interviewed 25 gym members. We showed them a prototype of Atlas and then asked how much they would pay for it each month. Answers ranged from £5/month to £30/month - almost no one said they wouldn’t pay. We also spoke to three gym managers, who each confirmed they’d pay somewhere from £500/month to £5,000/month.1
Full of excitement, we made a business plan showing how Atlas would make millions from a single contract with a big gym chain. While we hadn’t proved much yet about the business plan, we had strong backgrounds and a compelling story. Along with our positive customer feedback, we convinced ex-colleagues to invest some money for us to go full-time in building Atlas. What a great start!
The Gantt Chart (November 2019)
We spent our first day of full-time work making a Gantt Chart which mapped out the process to a multi-million pound contract from a big gym chain.
Outcome: get a big gym chain to sign a contract to roll out Atlas by November 2020.
…which requires the big chain to start trialling Atlas by September 2020
…which requires 5 smaller gyms to be paying for Atlas for 3 months by June 2020
…which requires 1 smaller gym to be paying for Atlas for 3 months by April 2020
…which requires 3 smaller gyms to trial Atlas for a month by March 2020
…which requires 1 smaller gym to trial Atlas for a month by January 2020
With the Gantt chart finished, all we had to do now was execute the year-long plan!
The first punch in the face (November 2019 - February 2020)
From November 2019 to January 2020, we aimed to build a product that gyms would trial with us. After several months of furious coding and taping together camera parts from Amazon, we were ready to launch!
There was only one problem: gyms didn’t want it. They wouldn’t even take it for a free trial. It wouldn’t clearly make money for them, as their members at risk of cancelling didn’t usually care about their deadlift form. Even worse, they may actually lose money if Atlas’ presence angered their personal trainers, who provided revenue for the gym. Several months in, we were failing to meet step 1 of our plan - this felt like a major punch in the face!
Regardless, we knew that startups are meant to be hard and we pushed on. We had some friends who were gym managers at a large chain. While they didn’t have any power to buy Atlas - purchasing decisions were taken higher up the chain - they agreed to trial with us as a favour. We started trialling in their gyms in February 2020. While there was some initial enthusiasm for our novel technology, very few members used Atlas more than once to check their lifting form. We certainly didn’t have the queues of raving fans that we were hoping for.
The knockdown blow (March - November 2020)
In March 2020, Covid shut all gyms in the UK. We shifted the Gantt chart timelines back by a few months, now aiming to sign the major contract with a gym chain by February 2021. We kept the rest of the plan intact, spending the next few months improving our code as we waited for gyms to reopen. While home workouts were becoming popular, we thought they were a fad that would quickly blow over. As powerlifters, we didn’t start this business to critique people’s starjump form!
Gyms reopened in August 2020 and we leapt at the chance to restart our prior free trials. We also went door-to-door around many other gyms in London. However, we couldn’t convince them to trial with us. We tried to pitch Atlas as a tool to convert potential customers who were on a tour of the gym floor into paying subscribers. This didn’t work either - most people were choosing their gym based on proximity to their home or work, rather than the range of equipment.
Gyms shut again in November 2020, which we expected to last the next six months. November 2020 also marked the end of our original Gantt Chart. By this point, we’d planned to have many happy gyms already paying for Atlas and signed our first contract with a big chain. Maybe a feature in Time magazine, too. Instead, we had zero paying customers and no backup plan - we were pitching to an industry that was completely shut down! Fortunately, we were thrown a lifeline from an unexpected source…
You can read the next part of our story here.
What we learned
Our first year illustrates the problem with Gantt Chart style planning for startups. We failed at the first step of our one-year plan, which left us with two bad options:
Give up after two months and return the money to our investors; or
Try even harder to execute a doomed plan
We chose the latter and, unsurprisingly, it didn’t work. Instead, it would have been better to acknowledge that our plan was risky. As we tried to execute, we could have prepared for things to go wrong and changed direction (“pivot”) as we learned more about the gym industry. We write about how startups should plan based on risks here.
Another lesson was the importance of interviewing potential customers correctly. By showing them a demo of Atlas before asking any questions, we hugely biased their responses. We also asked terrible questions, like “How much would you pay for this?” If you’re curious about why this was a bad question, we recommend a short book called The Mom Test. After reading it, you’ll be able to collect unbiased customer feedback from anyone - even your mom!
Special thanks to Dr. Tishtrya Mehta for her detailed review and comments, which greatly improved this post.
We spoke to a fourth gym manager, who was extremely negative. He said he wouldn't even take it for free. He gave several reasons and, in retrospect, some were worth listening to. However, he was also an unhinged racist so we dismissed it all. This experience demonstrates the value of being able to listen to people you dislike.