Case study: Building SaaS - Idea to exit in 6 months in AI-SaaS

The short version:

I wanted to become a better programmer, so I built an AI software for real estate agents from scratch. It took off! And just 6 months after the idea, the source code was sold to the biggest broker in Norway.

The long version:

Building and selling a software company, in 6 months, with no experience in programming

After selling my last company, Motkraft, I set a goal for myself to become more technical. Having all these ideas and visions without the tools and skills to build them were holding me back, so I set out to become a “master” at building products. I’ve done design, marketing and product building, now was the time to add the sledgehammer to the toolbox, programming.

I think there’s only one way of truly learning something, by doing!Therefore I set out to build a full AI SaaS tool from scratch. While I am certainly not a master programmer yet, in only 6 months I went from noobie programmer to selling the company and source code for the SaaS. It’s was a wild ride with plenty of bugs, pizza and RedbBull. Let me share some experiences from the journey, and how you can build products at lightning speeds yourself.

Finding something worth building

Here’s the thing about building products, most people build things no-one wants, because they have no insights into their actual problems. Try thinking of a problem for a business that you have no experience in. Pretty hard right? Most people will therefore build something close to themselves. The problem then is that you usually end up in a saturated market, shooting too wide or too narrow.

Balancing this is both the easiest and hardest part of Entrepeneurship. Many say that “ideas grow on trees, it’s all in the execution”. While that has a lot of truth to it, my opinion is that ideas definitely grow on trees, but good ideas are more like truffles. Sometimes you are lucky and stumble upon some, but to do so consistently requires a good nose (and often help from someone with a better nose than you, say, an industry expert?).

How I found the problem worth solving

When soon to be AI-overlord baby, ChatGPT, launched in November 2022, I was as impressed as anyone. However, the launch of GPT4 showed me that this really could be something. As a lucky coincidence I had been a subscriber of OpenAI long before they launched the products, which gave me early access! This opened some new markets, but finding something to build is hard. I tried medical devices, journalism and a couple others, but to no use.

The breakthrough came when my girlfriend was sending me apartments to rent. All of those listings looked very similar. Maybe AI could do that? It was a problem that was too hard to easily automate, but not too hard for AI to do well. I had it make a listing, looked good to me!

My father works in a brokers office. I asked nicely for the phone numbers of a couple of brokers. Copying the answer from ChatGPT into a word document, I showed them what it could do. They absolutely loved it.

Bother your customers

Now comes the most valuable lesson in building products, listening to your customers. And by listening I mean peppering them with questions and listen really closely to what they are saying. I traded a bottle of wine for an hour with a new broker. Without ever revealing what I was building I had her take me through their process and how she did her job. Digging just enough to find the biggest pain points.

It turns out they spend A LOT of time opening documents, copying and pasting the information into the sales prospectus. It was tedious work and couldn’t be automated as they had to read over it, make tiny corrections and rewrite some parts from technical jargon to understandable text.

AI can do this! I did the same exercise with other brokers, the result was a clear battle plan of what the product should be.

Become a padawan to a Jedi master in programming (or whatever you are learning)

Building software is a complicated endeavour. Also, building alone is draining. If I were to learn programming, I need to learn from someone that really knows their stuff. Luckily, I was working part time in a consulting agency with tons of great people. I asked the best one if he would be down to join me building this. We would split the earnings, he would teach me programming and I would teach him business.

The product

How to absolutely not build a first iteration We started building. A full blown software with authentication, billing, customer support, custom algorithms and a large database. When building a first iteration of ANYTHING, it should be as compact as possible. We completely circumvented that to build everything at once instead. There are two pitfalls here.

1. You spend a lot of time building without getting something in front of your customers.
2. What if the thing you are building isn’t good? You would have too big of a rig to start turning it around.

As two early twenties “brogrammers”, we of course thought we could build it in one go. It was an absolute disaster. Not only did it simply not work, we had brokers call us asking when they could get access (because we promised to give them something in weeks, but spent months). It felt like we couldn’t do it. It was too hard to build and we were not getting the feedback we hoped for.

What do you do in this situation? Lie down and die? Give up? A lot of startups die this way, running out of runway and traction. We felt like we had too good of a product to just give it up

How TO build a first iteration

One weekend I said “fuck it” and started all over. New repo, fresh everything. What could I build over the weekend that would actually work. That meant no authentication, no billing, no database. Just a “upload documents here” and a “generate” button. It worked like a charm.

You might think that not securing your application or putting a website like this out online for everyone to use is sketchy. “What if we get hundres of people using it”, “what if a competitor sees it”, “how do we control who gets to use it?”.

To be frank, no one really cares. And if there is a sudden inflow of users (which wouldn’t really happen either), that’s not even a bad thing! You have to let go of the notion that everything should be up and running when going to market.

We sent the first version to the brokers. They started using it and gave us some really valuable feedback (the kind that would make us start over if we finished the “big” version of the software before showing them). We implemented the changes and continuously asked for feedback. When the product itself was good enough for them to use, only then did we add authentication, a way to store the results and billing. Yes, you could not save anything in our app. If you refreshed the site, everything would be gone. That’s just the “minimum” in minimum viable product.

Selling to customers, a crash course

Calling someone up and asking them to subscribe to your software is a really hard way of getting traction. Instead, you need to get someone on your team and have them be ambassadors. I’ve done this in several companies and it always works.

Here’s what we did. The brokers that had tested the product got a deal. They would get to use it for free if they continued to provide feedback and help us sell it to other brokers. This was a great deal for them, remember, they saved a lot of hours by using our software. The cost of them using it was nominal for us, but they gave a lot of business back. The brokers gave us an entry into the franchise leadership. Suddenly, instead of us calling to pitch them, we had someone on the inside call them up and saying “you have to talk to these guys”. Now that is a great start to a business relationship.

Block their access, if they call, you are in the money!

One day, something caused the service to go down. Emails started coming in asking what happened to the service and when it would come back online. That prompted an experiment, actually shutting down access for some users. They ALL called. That is the most effective test to check for product-market fit.

The tricky balance of selling source-code

After the intro from the broker, the biggest real estate franchise was keen on buying the software. The catch was that they also wanted to build their own. We couldn’t show all our cards (aka our source-code), but at the same time we needed to show them something. What ends up happening in these situations is that you have to establish trust and hope the other party isn’t a dick. Reveal only what you need to, and if the deal falls through, say to yourself it was for the better.

That were some lessons going from idea to exit in 6 months. Did you find this interesting? Let’s grab a coffee one day, I love meeting new people.