This is the fourth and final 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
Back to the Barbell
The Last Roll of the Dice (June 2023)
We pick up the story nearly four years after founding Atlas. We’re licking our wounds and thinking about shutting down - how did we spend the last 18 months trying to sell a home workout app when we don’t like home workouts?! It was back to the drawing board for a last roll of the dice, but this time accounting for some lessons learned the hard way. We asked ourselves the following questions:
Q1: Who do we really care about and want to help with their fitness?
A1: People lifting weights in gyms - that’s why we started Atlas AI.
Q2: What problems do people lifting weights have that are big enough that they’d pay to solve?
A2: They aren’t making progress with building muscle. They’re already paying for gym memberships, protein powder and (sometimes) personal training.
Q3: Can we help them make progress with gaining muscle?
A3: Yes…empathically yes! We’ve spent years coaching weightlifting, so we know that people will make great progress:
Follow our lifting plan, which requires lifting 3 times a week
Recover from their sessions by sleeping and eating well
Maintain good form as their weights get heavier by videoing themselves
We came up with an idea to package these features into a phone app. Users record themselves lifting with their phone camera and the app will check their form automatically. They’d also get access to a lifting plan and diet advice. We’d charge $20/month for the app.
We knew there were big risks, particularly:
Will people pay $20/month for our system? Most people don’t currently pay for a gym app and there are many free alternatives to get a lifting program.
Will our customers record videos of themselves lifting in the gym? Most beginners are self-conscious in the gym, so asking them to video themselves could draw attention and make that worse.
We felt the vision could work, and that people would set up their camera in the gym if the incentive was strong enough. However, we agreed that if we hadn’t made sufficient progress by September then we’d consider shutting down Atlas. Sometimes, you’ve gotta be willing to take big risks…and lose everything. The die was cast!
The Power of Mission (July 2023)
We made a strong start. By the end of July, we’d signed up 10 users each paying $20/month for the phone app. We reached our early users with a scarily simple, and not in any way “genius”, approach:
Reach out to friends, asking them if they knew anyone lifting weights who’d be up for a 30-minute chat with us.
They would give one of three responses:
Sorry, nope (or relatedly, no response)
Actually, I do know someone. Let me put you in touch!
I’m lifting weights! How can I help?
On the call, we asked if they were going to the gym currently and had an active goal to build muscle. If so, we’d play a 2-minute video explaining our system and, if they liked it, we’d offer them early access for $20/month.
People were usually keen to join, even if they’d never met us before. They could tell we knew our subject matter and believed we could help them. The scary part was that we never tried this simple approach before - we’d made more progress within a month than we did for 18 months of trying to sell a home workout app! The main reason was that this time, we had a mission we believed in: helping people make progress with a simple and scientific approach to strength training.
Behind the scenes, we didn’t have a fully automated app yet that could check users’ form, as well as give them a lifting plan and diet advice. Instead, we analysed users’ videos ourselves and sent advice to them over WhatsApp. We took this approach because our biggest risks were whether customers would pay, and video themselves lifting. We didn’t need a fully automated app to test these risks.
We also wanted to keep our focus on helping customers, rather than building complex tech that no one was paying for. Full automation wasn’t needed to help ten customers make progress with building muscle. However, we knew we’d need to improve automation on the app to reach our next goal of helping 100 customers.
Ominous Signs (August 2023)
We started running into challenges in August as we tried to grow to 100 customers. Our first batch of users liked the muscle-building program and were making progress, but they weren’t enthused enough to refer their friends to Atlas too. Friends that they did speak with weren’t necessarily lifting weights themselves. Without referrals, it’s very hard to make an app work commercially when charging $20/month.
In mid-August, we faced a potentially fatal technical issue as our AI form checker wasn’t working from a crucial camera angle. If we couldn’t fix it, this would kill our app’s viability. We brainstormed possible solutions, whose testing time ranged from two days to several months.
Luckily, we fixed it in two days. But when we realised we’d fixed it, we shared the same, grim sentiment:
“Is this good news, or a longer rope to hang ourselves with?”
This was a particularly ominous sign. Rather than being filled with joy that we’d saved our app, we had mixed feelings. Maybe we were starting to feel like we didn’t want to keep working on Atlas. And if you don’t desperately want your startup to succeed, it’s not going to…
The Dice Settle (September 2023)
We came back from a brief summer holiday in September and reviewed our progress over the last few months. We’d signed up 15 customers, of which 10 were sending us lifting videos and 5 were still highly engaged. While our surface-level metrics had been better in the past, it felt like we were doing better than any previous pivot. For the first time (and slightly embarrassingly), we were delivering consistent value to 10 paying customers.
Nevertheless, we were facing major challenges. Building muscle takes more than just good form. It also involves:
Motivation to show up consistently
Choosing the right exercises and weights
Good nutrition and sleep to recover
We had a novel solution for maintaining good form, which delivered great results. However, we didn’t deliver anything novel for the other components, particularly motivation to show up. While customers were making progress, our overall “muscle-building” system wasn’t as outstanding as we’d hoped. It was going to take a lot of work, and time, to overcome these challenges.
So we decided to shut down Atlas AI.
We hadn’t proved that our vision was impossible to realise, but we were facing major challenges to get to our next goal of 100 customers. We were scientists making what felt like a very un-scientific decision: after four years of struggling, we’d frankly run out of motivation to keep going. Our last roll of the dice went okay, but we got a four when we probably needed a six.
We feel incredibly grateful for the opportunities that Atlas gave us. We learned so much in the process and met many, many fantastic people. We didn’t find commercial success, but we grew into very different (and hopefully better) versions of ourselves. We feel that this truly is a success worth celebrating.
Lessons learned
Most people should not love your startup idea. If they do, it’s very likely that your idea is doomed for some reason you can’t yet see. Most people loved our “AI personal trainer phone app” idea, so it’s likely that many founders had already tried it and failed for similar reasons to us. We will write more about this in a future post.
The books warned us, but we had to learn the hard way. We’d read about some common startup errors, like building tech for too long instead of selling. But we still made all these mistakes ourselves, sometimes repeatedly! First-time founders often have to learn about themselves before they can start to learn about their market. We will write more about this in a future post.
Special thanks to Dr. Tishtrya Mehta for her detailed review and comments, which greatly improved this post.