How do you start a business from scratch? In fact, how do you make any dream come true when you’re starting from zero? Maybe you’ve already done some homework by talking to potential customers or others that already made it. But now it’s time to get cracking and make your idea a reality.
This is really intimidating if you don’t have the means to quickly get off the ground. Whatever it is that you’re trying to launch, you might feel like you first need to:
- spend years developing advanced technology (self-driving car)
- raise an enormous amount of funding to pay the initial cost (casino)
- overcome regulatory hurdles (new medication)
- gather a sizable user base (dating app)
Back in 2017, my co-founders and I faced these classic startup problems when we started Point. Our vision was to automatically reply to emails using the latest NLP technology… but we didn’t have much experience in machine learning or any money to hire people. We weren’t even sure if a product like this would be possible to build or if it was just science fiction.
What should we have done next? Invest years into hardcore R&D, hoping that it’d result in a tech breakthrough? Pivot to a completely different business idea?
Neither sounded appealing. It’s a fatal mistake to spend too much time building something the market doesn’t need. At the same time, there’s only so much talking and strategizing you can do without actually building something. While we were racking our brains for a solution, we heard about a hot new startup called X.ai and another option emerged: simulate it.
Artificial artificial intelligence
X.ai built Amy Ingram, an AI personal assistant, to automatically set up meetings over email. Users could just cc Amy on an email thread, and “she” would handle the back-and-forth with the other person to schedule a meeting. Amy sounded a LOT like a real person, and people back in 2014 praised her “humanlike tone” and “eloquent manners.”
What an amazing piece of tech, right!? What most people didn’t realize about Amy was that behind almost every email was an actual human. X.ai hired a team of human “AI trainers” to handle much of the emailing and scheduling, and they’d already raised tens of millions of dollars to do it.
Does that make X.ai a sham? As misleading as they seemed, the “AI trainers” kinda did live up to their job titles. By responding to emails, they were creating a robust dataset that could be used to train a fully-automated machine learning system. They were basically “faking” the Amy Ingram experience until they could actually “make it.”
Our team was shocked by X.ai’s bold yet ingenious business strategy. I remember mulling it over in the shower one day, when it suddenly hit me… what if we did the same exact thing?
Fake it ’til you make it?
Normally, MIT nerds like us love to put our heads down and build bots. This time, we decided to “simulate” our business idea by faking the AI before committing to building it.
Here was our master business plan:
- Run our “simulation” on a few users by replying to their emails
- Scale up to more users and hire people to write emails for us
- (Somehow) figure out how to speed up our human emailers via automation
- Release our magical, fully-automated email replier and get rid of the human emailers
We rounded up ten users, ranging from acquaintances to business mentors. We built a system that would forward our users’ newly-received emails to our own staging area. Whenever an email arrived that needed a reply, we would manually(!) write up a reply and save it as a draft in the user’s inbox, ready for them to review and send.
So how did this go? In a lot of ways, better than I ever imagined.
Our first set of users paid us for each draft that Point “wrote” and gave us positive reviews. We suddenly had a business idea with some validation and a small (but non-zero) revenue stream.
From there, we were excited to push for our first round of funding. We told investors about our “simulation” results, and to our big surprise, they really loved it. We checked the boxes for entrepreneurial hustle and innovative thinking, and as a result, we received our first investments from a VC firm as well as Y Combinator.
With money in the bank, we followed through with the next steps. We started marketing efforts and hired people via Upwork to reply to our users’ emails. X.ai had their “AI trainers,” and we now had our “Point operators.” With an actual business in motion, pre-seed investments, and a place in the prestigious YC accelerator, it felt like we made it big and won the lottery!
Of course, not everything was so simple. It was a logistical nightmare to teach our operators how to properly send emails and assign them to work shifts around-the-clock. The YC partners hated that we were paying so much money just make our product function.
On top of that, we were in a moral gray area of being an AI company without real AI yet. It was a very uncomfortable position to be in, even though putting on appearances has been a big part of the Silicon Valley culture, especially when it comes to machine learning hype. One particularly bad example was Engineer.ai, who claimed to use AI to automatically build apps but was just using humans to do it. It used the AI hype to attract customers and over $30 million in funding! It’s one thing to openly acknowledge using humans as part of your business model, but it’s a completely different story to hide it for hype.
We ended up taking YC’s advice and let go of our operators to build a fully-automated product. We had about three months to pull it off before our Demo Day. Despite the daunting challenge, we successfully morphed our email drafting service into EasyEmail, a Chrome extension that used machine learning to autocomplete sentences in Gmail. It was a much easier task to automate than writing entire emails, yet still provided nice value to users. If that product sounds familiar, it’s because Google released their own implementation called Smart Compose just a few months after we already launched 😒
Competing with Google in AI was not a fun time. But hey, at least we were onto something there!
There’s only so much you can accomplish by strategizing and talking to people. That’s still the best first step to take, but real progress happens by doing, learning, and often times failing. Our email drafting “simulation” was incredibly important. It allowed us to test a business idea that seemed very far-fetched, establish ourselves as a real business, and prove ourselves as startup founders to investors. We fulfilled our dreams of launching a tech startup, raising a seed round, and selling a real product, even though the company didn’t work out perfectly in the end.
But be careful. At the end of the day, whatever business you start must still provide value. You might be able to get small wins at first by faking it, but reaching success at scale will require you to build an excellent product or service that fits your market’s needs. Use the simulation as an idea validator and not a replacement. Also, being honest and transparent is key if you want to build a good reputation, which tends to be great for winning long-term customers and being happy in general.
So what happened to X.ai? To their credit, X.ai worked hard on an incredibly difficult machine learning problem. But nowadays, they’re outcompeted by Calendly. Forget about fancy algorithms – Calendly wins scheduling by just letting people share a link! Instead of having to raise tens of millions of dollars, it was bootstrapped by its founder and now makes over $30 million in annual revenue.
Now THAT’S an example of a real success story: no hype, no BS, just pure value made by a normal person with a dream.