Percepta: Lessons Looking Back

Startups are hard — a wild, yet worthwhile journey. However, after working on Percepta for 2 years, it’s finally time to say goodbye.

I am writing this mainly for myself, as a way to document what happened and the lessons learned. I also hope that anyone else who stumbles upon this can learn from my experience.

What is Percepta?

At Percepta, we were developing computer vision models that would process anonymized video footage (people were abstracted into object meshes) to analyze actions and behavior. We specifically applied this to detect and alert shoplifting incidents.

The anonymization component was important to 1) actively avoid biasing our models on demographic features such as race and gender 2) protect identity/data of people in the footage. The long term plan, initially, was to win in shoplifting, then expand into other use cases in retail (eg. planogram compliance, proactive customer service, inventory management), and expand into other verticals (eg. public & corporate security).

Percepta came to be when Philippe, my friend and co-founder, approached me with this idea, and specifically the technology behind it. The original proposal was actually bike theft detection, but we found working with public safety too difficult and we wouldn’t have been able to acquire data to train our ML models in any reasonable amount of time. While exploring other use cases, we began making some inroads in the retail space — the customer interviews we conducted were promising, so that’s what we pivoted to instead.

Lesson 1: Any money that isn’t customer money IS NOT validation!

We got some initial validation that shoplifting was a huge problem (cost billions a year), processes were largely manual (thus inefficient and ineffective) — which suggested the even this specific use case alone could hold its own as a business. Combined with our technical proof of concept on a small dataset and a solid pitch deck, we were able to nab some early wins with regards to fundraising & publicity.

Winning pitch competitions and getting money definitely feels awesome. However, I certainly conflated early investor money with market validation, whereas ultimately, the market is market validation. Most ideas don’t work anyways, so fundraising is more of a reflection of one’s ability to sell/storytell and investors taking a bet on you.

Lesson 2: Product Market Fit and nothing else…

We ultimately did not achieve product market fit either. At least in the United States, shoplifting is largely decriminalized, and more importantly, store employees are discouraged from intervening due to liability. So despite customers saying that they loved what we were doing, they wouldn’t actually use Percepta to stop shoplifting (the main value proposition) because their operations wouldn’t support it.

We also should have acted on this conclusion earlier. We were preoccupied with product development — getting models to work, sourcing data — and should’ve spent more energy ensuring we were building something customers needed and would buy. There is no room for wishful thinking in enterprise sales. If it’s not in that department’s top 3 priorities, and if that department doesn’t have the budget or isn’t somehow vital to core business, it’s almost not worth it. Selling anything is already hard enough, and if what you do doesn’t literally provide 10x ROI, don’t bother because customers will not buy.

What we should’ve done was target assumptions that had the largest risks upfront and early, conducted customer/product discovery on a continuous basis (eg presenting mockups, work with smallest sample size of customers that’d provide enough signal), and actually try to perform the customer’s job (retail security) to get on the ground insight. Focus on fast iterations and fast learning, predetermine your best guess/benchmark for failure/success. For further reading on the topic, I highly suggest Talking to Humans.

Lesson 3: …except perhaps Founder Market Fit

Sounds obvious, but you should actually care about what you’re building. In retrospect, we had little business operating in retail/shoplifting. We were excited about the opportunity, but weren’t ultimately that invested in the use case.

You could enter as complete outsiders and make it work by sheer will, but at the very least, having someone involved with the minimum effective dose of insider knowledge can make a whole lot of difference in terms of time/money/energy saved.

When things aren’t going well, which is most of the time, you need that intrinsic motivation to keep pushing. Don’t do something hard that you don’t feel bought into or aren’t excited about.

A hodgepodge of takeaways on company building:

Beyond creating a successful product, building a company/organization is hard too. Here’s what I learned/messed up on.

• Have a really clear sense of what matters and what’s going to move the needle. In the beginning, it’s derisking your biggest assumption. Then it’s probably building product, sales, and hiring.
• Who you work with is so important and one of the few things you can control. Be very deliberate here.
• Fire fast — we hired a promising engineer that ended up not working out. We saw the problems about a month in, but were too delicate and kept giving chances for another 3 months. Stop wasting precious time & cash, trust your gut. Front loading more diligence in hiring could’ve help avoid this scenario as well.
• Write things down/keep documentation — summary emails after calls with external partners, meeting notes/jira tickets that outline next steps. Have minimum viable process, not no process
• As a founder you do hold authority, whether you like it or not. The reframe for me was that authority was about being responsible — to customers, your team etc. So use authority because you are ultimately responsible, especially for anything bad that happens
• Be selective with mentors/advisors/other accelerator programs. You are racing against time to escape being default dead, which you most certainly are in the beginning. So be crystal clear and hyper specific about what you get out of these engagements and how they move the needle.

Despite everything, we still had some awesome highs — our first pilots, hiring great team members, investment from a huge strategic etc. I’d like to thank our awesome team, customers, mentors and advisors, for coming along the journey with us.

The early team!

At a high level, I think we were onto something with the private/ethical AI trend. Society and regulators are starting to pay more attention to AI ethics, particularly with facial recognition and in areas such as policing. In particular, increased regulation may force certain aspects of privacy-preservation and ethical usage to be table stakes. On the flip side, we may also be too early. Explainable AI, a bit more of a middle ground/stepping stone, is gaining traction, and is perhaps a more attractive avenue for pursuit given its utility in both AI deployment & adoption as well as model training.

One alternative direction that we never fully explored was building an API company. Think “privacy-as-a-service”, where we would provide tools that helped with anonymization and building these privacy-first models. Whole host of other problems there to solve, but reach out if you’d like to jam on it!

Startups are hard, but at the same time, nothing else beats it. It’s not for everyone, but I don’t think I’ve ever learned and grown so much. If you made it here I hope you found this useful (and don’t make the same mistakes!), and please share with anyone who would find this useful as well. If I can help in any way, reach out on twitter @itsjonnymak