I made this its own post because it's a very good point.
Do you know who knows this? Our friends at Verkada!
I had a related post 2 years ago when the current generation of hype started - Deep Learning Surveillance Startups Deep Problem. A lot of these companies are going to fall by the wayside. The 'winners' will be disproportionate that can figure out how to turn this into a business versus incumbents, not necessarily the best technology or cutting edge research.
I paid $25k to learn this lesson in the early 2000's: there was a company that made a digital audio signal processor.. it used less power than the competition with greater db output and signal quality. The CEO was also the inventor (there's your first hint) during a time when flat panel TVs were booming and plasma's used a lot of power and generated a lot of heat.
Technology quality is important. Demonstrating how it solves the customer's problem is the MOST important. In 2008 the company (along w/ my investment) went under. I guess I should have thought more about VHS vs Beta Max.
I agree with the sentiment of this as someone at an AI start up , I can add some nuance to what I've observed from competitors - I'd like to think WE are different but I'll remain undisclosed. For a capital raising start up, the security industry is an attractive target because there is clear unit you can pitch to a VC... we charge x per channel and we need to achieve Y streams and we can grow this at Z rate. The VC (who knows nothing about security) looks at this and can't help but get excited.
Oblivious Business Models: Per stream fees from $10 - $300 per stream. Not taking into account the entire value chain of oem, dealer, integrator.
Lack of integration: If it's not RTSP they can't do it. Let's not take into account that RTSP introduces a ton of artifacts that can make AI useless. The only integration is to do basic notifications to milestone and genetec and making a huge PR campaign to announce that. The "secret" for an AI startup is to focus on all of the non sexy data science parts: the integrations / firmware/ architecture work.
No understanding of the typical security hardware stack and will ask naively to on prem hardware that hasn't been thoroughly security tested. Not to mention the lack of focus on the edge (for software based AI companies)
Unrealistic Sales/ Marketing: AI is super hard, especially computer vision it will never be 99%... ever...never... I'll die on this hill. These are statistical prediction of phenomena in the real world. The value of AI at the moment is to amplify human efforts... humans make mistakes too and there is a cost / benefit advantage to using AI even if its worst than a human... but marketing that is not very sexy.