Hi there. Let me tell you the story of our AI startup. We made a few mistakes along the way, and I think you may learn something from hearing our story.

In past articles I have talked about why we bootstrapped our company, how we feel about our place in the AI consulting marketplace, what the clients tend to look like in our space, how we think about our pricing, and a bunch more on the machine learning landscape and dynamics. This article should fill in the missing link, and explain how I got into this Startup, and where the company is going.
In terms of life planning, it was pretty stupid to get married and have kids at the same time as working full-time, and at the same time doing courses towards a masters, and then a PhD. It was… challenging. I couldn’t just sprout a startup of my own, because I couldn’t afford to do the Ramen Profitable thing on my own.

My wife worked full-time as Mom, Guidance Counsellor for the school board, and generally managing the whole house while I programmed like a monkey at a typewriter.
Life was too busy. I couldn’t start my own company. Instead I worked in other people’s startups, and had a great time doing it. I got to look behind the curtain in the startup world thanks to my dad and his friend’s investment in high-tech, and my father-in-law’s business distributing chemical products. I learned a lot. My failures were more informative than my successes.

I was working on a PhD in 2016, and could use a side job. I had quit my full-time consulting job to focus on school, and lived off some savings we had built up. I didn’t want to fully walk away from industry, and I had a good history of consulting in the cloud/IoT space, and so I joined with my friend Mathieu Lemay to start a consulting firm. We knew we would work well together because we had worked together in a previous company, and got a lot of work done in a short amount of time. We started off doing it as a side thing, and it quickly started gaining traction. Since I was part-time and Matt was full-time, we called the company Lemay Solutions Consulting Inc (originally it was just Lemay Solutions). Matt was using LemaySolutions.com for his private server since 2014–09–13, and so the name stuck. In the office, we call it LSCI for short. The official founding date for the company was June 22, 2016.
LSCI started as literally a startup in a garage. My garage to be precise.




The founding concept of the company was AI+IoT+cloud. Our overhead was basically zero. Initially, we did any job that paid. We had a Chinese menu of everything we were good at, and basically looked for projects that fit those broad capabilities. There’s a range of high-end things we can do well. That includes designing and making circuits, firmware, grant writing, and more. But our dream was to specialize. I have 3 kids and had to pay the bills. Slowly, over time, we narrowed the jobs we would take to those that fit our expertise. We specialized.
Marketing Channels
We started off our marketing strategy with a funnel of contacts from my email, LinkedIn, and facebook. Rather than using an ERP or CRM, I used google sheets with sort and filter on columns to prioritize leads, and contact individual people with personalized emails. I also tracked lead status in a dedicated column. It worked. We never really used traditional online marketing tactics. At least not effectively. Running through my contacts, I marked high value targets in a dedicated column, and reached out into my network to establish business relationships.

I added some smart features to the spreadsheet to save me time. First I added deep email and domain name validation (as described here). And later I added filters by domain name to make sure I was not contacting several people from the same small firm.
I also met clients at startup conferences, and later on through online job boards like indeed.com and upwork.com. Some channels did not pan out. For example, we tried to get freelancer.com to work for us, but the rates paid for gigs on that site are not worth the effort involved. It feels more like a mechanical turk than a professional services contracting site. We have not yet tried out toptal and a few others. So really we are at capacity, just doing the things that are working for us now. There is surely room for improvement as we grow.
Through the winter of 2016, the GPUs warmed the garage like an extra heater. We wore winter coats indoors, even as all the heaters were running. It turns out we wired the heaters wrong… It was not an optimal setup. The only bathroom was inside the house, and so there was the option to pee in the bushes at 3am, or head down to the basement. We were basically working 12 hours a day, 6 days a week. With the PhD going on in parallel it was pandemonium.
Keep in mind that I’m finishing a PhD full of machine learning and wrangling kids (wrangling data?) while this whole consulting thing is starting to take off. Here’s a picture of me winning a grad poster award during that same time period:
Compétition d’affiches des études supérieures Grad Poster Competition
As spring arrived, we had a fat hedgehog move in underneath the foundation of the garage. We called him "el-chubbo". Flies invaded, and the air conditioner was acting up. It all became too much to handle. We decided to move to a real office. On the bright side, we could finally afford to commit to the cost of having our own space. We kept the servers in place in the garage, and moved the "staff" and computers. We kept the soldering and other electronic equipment in the garage. It’s now the hardware office and server farm.
As founders, we have always had this vision of building recurring revenue; Passive income on top of our hourly work. A CEO recently gave me this great "write once; sell many" quote. That’s what we were hoping for. We always told ourselves a story about the pivot to product. We would make enough money from consulting to stop consulting, and just focus on product development. But even though we had many nifty product ideas, we had a really strong hesitation to actually launch the things we built ourselves. We developed AVRA (based on my dissertation), clockrr, genrush, and a few others that are still in stealth, but never built up enough courage or effort to do a real launch. We have always had some consulting fire to put out. I have come to believe that we like it that way.
Lessons Learned on Cash Flow
My view on the consulting world is that work takes 2 months to pay out. If you invoice on the first of the month (e.g. Jan 1), and start a new client on the first (same day), then you will send out the first invoice to that client on the first of the following month (Feb 1). With net 30 terms, you expect the payment at the beginning of the following month (Mar 1). So the money from the first hour of effort reaches you after ~2 months of delay. Critically, not all channels work this way. For example, on UpWork, we get paid a lot faster. The downside is that they take a cut, but the upside is that they "pay" faster.
We had a cash flow situation several months ago when a client unexpectedly paid late, and we were short $50K on accounts receivable. The late payment was offset by new income from new clients, but it was a real wakeup call that even after the 2 month delay, you may be in for even more waiting. Another quote I got from a CEO on this situation was this: "Sometimes you become an involuntary investor."
Another lesson we learned on cash flow was irregular expenses. It is all great to look at the world as EBITDA (Earnings Before Interest, Tax, Depreciation and Amortization), but in real life you have to pay taxes. We were on an annual tax plan with our corporation, and taking out 100% of the money to pay ourselves. The tax bill came due and we literally had no money left in the bank to pay the taxes and also pay ourselves at the same time. We basically had enough for payroll and taxes. So… We basically skipped over a month of pay to get caught up on taxes. We are now on a monthly tax remittance plan, but that was a real eye opener of a lesson.
Now let us pull back and remember that we are not just a generic consulting firm. We are an AI+cloud+IoT magic shop. What is the difference for us compared to a car wash startup? Well, we have the ability to scale in fancy ways that most consulting firms don’t. Let’s look at that angle a bit.
First, we are global. Even though LSCI is tiny, we have had clients in the UK, Israel, US, Canada, etc. I actually prospected POCs in Africa and India. This Geographic reach gives us a lot more ability to fish for contracts. It also means we can say no, and not worry that we walked away from eating breakfast tomorrow.
Second, we can charge in USD. With the Canadian dollar where it is, that’s good for us AND good for our clients. We are way way cheaper than our competition (huge consulting firms), and it’s easy for us to say "ask around for quotes." Those prospects who want gold for the price of ice we just don’t chase. I can tell really quickly now which prospects get the value prop, and which just don’t get it.
Third, there is huge demand. Now that we have a reputation, we can raise prices (have done this 3 times). We shifted to more fixed-price contracts, where possible. And, as we built up some confidence, we started asking clients for deposits, to make sure they really do have skin in the game. Shocking fact: it made no observable difference in the client’s desire to commit. It just chases away the clients that are not too serious. You get crazy credibility just for swimming in the deep learning pond. We do the deep learning stuff that, as I found out the hard way, other people pretend they do. Some consulting firms post on their website that they do deep learning, semantic embedding, knowledge graphs, etc. BUT, when you ask them to do something, you find out they are kind of experimenting, and/or subbing the work to grad students. As I discussed in previous articles, this space is so hot that the talent available to work for midsize firms is really scarce. For larger (huge) enterprises, the pricing is way higher than our pricing, but there is a different set of firms, the IBM/Accenture/PwC/Deloitte/Fujitsu/etc big fish, that deliver for that level of client.
Now, in addition to the advantages we get from specializing, there are some serious disadvantages too.
First, high-quality staff is super hard to find. We have to hire high-end developers to subcontract or to do full time. For example, besides for the founders, we have a PhD and a Masters in CS on board. We interviewed a bunch of masters grads and plan to add one in the coming week or two. We first tried to use undergrads to do the work, but it was a total mess. It just didn’t work. We need grownups on the team who can do the complicated machine learning stuff AND do the client contact part of the job. So, we had to hire smart, and we did. I am 100% positive we will face more scalability challenges as we staff up over the next 12 months.
Second, we are not at a big enough scale to do big time-sucking RFPs, which is where the really big money is hiding. One successful RFP with a fortune 1000 company could be bigger than everything we have done to date, but given the 6 months of effort required to bid on these things, and the fact that we have business coming in already, we have not made the leap to speculating on building RFPs. We have subcontracted on contracts where we are NOT the prime, and that’s as much as we can handle right now. We did tip our toe in the water by talking to a few interested parties, but our eyes glazed over at the timelines. We heard phrases like "This project will be fully specified in 12 months." UGH.
Third, a lot of the things companies want are either trivial or impossible. Unlike traditional programming, the clients often don’t understand what the technology can and cannot do. This barrier to getting a closed and signed off statement of work can hold us back in any project.
Fourth, we can’t grow the business with hourly work alone. Our model is to mark up none of the expenses. We sell our time, and configure everything else to be billed directly from the vendor (e.g. AWS) to the client. We don’t scale as human beings selling our work hours. In an up market as hyped as machine learning, we should be growing like crazy. The gig economy gave us the scale we needed to pop into existence in year 1. But now, well into year 2, I feel like we are competing with a bunch of grad students looking for spare work. We are a small fish in a school of small fish, all biting at the same huge food supply. Way back when we started this thing, I modeled the company on the TV show Suits (but with less fraud). Basically we would be the smart fix-it engineers, selling out our billable hours as the rescue firemen who go diffuse bombs and jump onto the next contract. But, no plan survives first contact with the enemy.


I thought that our differentiator would be our hardware expertise combined with machine learning, and revenue growth would come from retainer clients. But it turns out our killer app is in multi-model solutions for small to medium sized enterprise clients. Very few people know how to mix, say, a deep learning model with a word embedding model and then again plug in a knowledge graph. Doing it on big data is even more rare a skill to find. I don’t know if we will jump onto the RFP bandwagon. Instead, I think we will address our growth strategy with new sources of (finally) recurring revenue through licensing deals (1st deal closed, announcements coming soon) and growing the staff with grant backing (1st deal closed, announcements coming soon). These deals took many months to come through, and some other leads drank up a lot of time and led to no return. C’est la vie.
Travel = Bad
We also learned a lot of lessons about how to travel as a consulting firm. Travel eats up a lot of billable hours. Especially if we all hop on a plane together. It is much better to avoid travel altogether wherever possible. Or to send one person only. Both [[Mathieu Lemay](None)](None) and I traveled more early on than we do now. The trips to Florida, California, New York, Montreal, Toronto, etc. It all wears you down. And you lose even more time on jet lag and insomnia. I mean, travel is not avoidable entirely, but less is more. Mathieu Lemay drives from Ottawa to Montreal to go to a "better" gym, so it’s not like we don’t travel. One big difference is that Mathieu Lemay can work on the plane, whereas I don’t like that. At best I watch in-flight movies. Ghost in the Shell, Moana: I enjoyed both at 30,000 feet. I so strongly prefer Porter over any other airline, that I should be buying their stock. Oh, they’re private. Too bad.
Conclusion?
My goal is not some huge exit, end to live in a resort on a beach. My goal is to keep doing what I’m doing, and to do it better and better. The way Mathieu Lemay and I have looked this all along, we put a coin in an arcade game called "AI startup", and we are still in the game. As we get to higher and higher levels, the bad guys change tactics, and the missions generate more coins. We are just hanging on to the rocket ship, hoping to get our names on the high score board by the time we reach the last level.
To our valued customers who make this whole story possible: Thank You!
The same goes to my wife Liora, and my work wife Mathieu Lemay. You both bet the farm on this startup. Thanks for believing.
I wrote this article in anticipation of reaching 1K followers on Medium.com Right now I’m at just over 900. If you enjoyed this article then please try out the new clap tool at the bottom or left or right of your screen. Wherever they put it in the template. Go for it. I’m also happy to hear your feedback in the comments. What do you think?
Happy Coding!
-Daniel [email protected] ← Say hi. Lemay.ai 1(855)LEMAY-AI
Other articles you may enjoy: