Like many investors, before I was an investor, I was a founder. Like many founders, I failed — spectacularly. Flag Analytics is the story of a lot of very early startups — poor founder-market (and founder-founder) fit, poor product-market fit, poor problem-market fit. We were not set up to be a unicorn success, from the start.
Unlike most investors and founders however, I was building a new type of early intervention system for police departments — software to help departments identify individual police officers at risk of committing just the sort of atrocities committed by Derek Chauvin in Minneapolis, Jason van Dyke in Chicago, and countless of other non-filmed incidents across our country for centuries.
The problem we set out to solve
is one that you see plastered across your screens today, and much of the last several years, and if you’re far too rare in the venture capital and startup space, that you’ve experienced in your own community since childhood. However, and you probably know this too — most officers are not at any risk of killing anyone. Take this analysis from 538 — over a 5 year period, 7,758 police officers in Chicago had at least one complaint against them. Chicago has about 12,000 officers, and if you assume a 14% attrition rate, that means there were about 19,000 officers employed in Chicago in that period of time. That means 60% of officers didn’t have a single complaint filed against them. Fewer than 9% of all officers in Chicago had 4+ complaints filed against them over 5 years. Derek Chauvin had 17 since 2001. Jason Van Dyke had 20 complaints over 17 years. The private data I had from two large urban police departments backed this up. All departments have to do is identify these officers, and intervene early.
Where we failed at product-market fit
was that we tried to use machine learning and other forms of advanced analytics to find these officers. I get pitches all the time from companies with this same fatal flaw — don’t use a neural network when a Tableau dashboard will do. It is not at all hard to figure out that an officer who has a lot of complaints in the past will probably get more complaints in the future. They will probably have the same kind of complaint — it’s not at all surprising that Derek Chauvin has two previous use of force lawsuits on his record, and you don’t need AI to tell you that he would be much more likely than the average cop to have a third problem. However — we learned during customer discovery that most police departments didn’t even have an easy-to-view dashboard where they could quickly and easily see which officers did have more complaints than their peers (normalized to things like activity level and community), so the Chief of Police couldn’t easily evaluate, even at a surface level, who she needed to intervene with. One central office partner we worked with was blown away by just getting a simple dashboard — something most startup folks would take for granted. Machine learning — expensive, difficult, hard to hire for. SQL and Tableau? Cheap, easy, lots of talent.
But there was still a problem-market fit problem.
Great — we could now, pretty easily, tell police departments who their problem officers were. Unfortunately, they then had to want to do something about it — and all of the evidence shows that at best, police departments to this day are only willing to pay lip service to reform. We were working with two departments whose leadership said they genuinely wanted to do the right thing — that’s why they were working with us. They kept getting in their own way though when it came to actually solving the problem, which showed me as a founder and a citizen that they weren’t ready to change yet. We sent a list of problem officers to one of the departments for review, and we quickly got a call back from our contact that yes — there was some accuracy here. One of the most problematic officers was well known to leadership — he thought he was Batman, and the department internally called him Batman. He was still very much employed. They knew very well he would be trouble — so why did they need us? Other officers we flagged they objected to — oh, they’re just active officers, lots of arrests — never mind that there are plenty of other active officers that have many fewer complaints. The departments — even the most well-intentioned ones — still weren’t ready to actually solve one of the biggest problems they had. Your customers have to think what you’re doing fixes a big problem of theirs, in a way that works for them.
Of course, I’ve never been a police officer
and neither had any of my co-founders. Nor, for that matter, had we been victims of the police, or had any substantial prior exposure to police. No one on our team was black. We relied on other aspects of our credibility, which worked when speaking to some of the executive leadership at departments, but made it harder to justify our objections to their pushback. It made it even harder when we talked to rank-and-file officers about what we were building — why should they trust our algorithms would understand their day-to-day, and not penalize them unfairly. Forget about convincing police unions to support any sort of new HR system — let alone one they viewed as a black box. I feel like exhibit A for why I love it now when entrepreneurs have deep industry experience.
What happened in Minneapolis and Louisville and Chicago and countless other cities across the United States is unconscionable (and to be clear, racist). But what makes me the maddest is how easy this problem is to solve, just by the police deciding it’s a thing they want to solve. Look — policing is hard. I met an officer who puts their hand on the trunk of every car they pull over, just in case something goes wrong, so there are fingerprints. You get a call and have very limited information about what you’re showing up to — which is often a person on the worst day of their life, well before you came on the scene, and it can get chaotic. But most officers are empathetic, and well-trained, and would never use any force on anyone selling loose cigarettes — and so they, more than anyone, should want to rid their workplace of anyone who would do that — and more often than not, they say they do.
As a venture capitalist, I get pitched sometimes by people solving genuinely hard problems — hugely difficult technology, major coordination problems, and more. Often the most interesting companies are ones that are just bringing online processes that are still being done via pen-and-paper — the Tableau dashboards, not the neural nets. Venture capitalists and tech entrepreneurs are incentivized to believe software can solve any problem. But your customers need to want it to be solved first.
 Complaint counts across departments should not be compared — each department has different polices for filing a complaint, which makes it much easier in some places, and much harder in others (including Chicago), and will skew counts.
 I filed a complaint myself, through this experience — an officer who I rode along with pointedly ignored a citizen obviously in need of help and trying to get our attention, and told me they should just call the police. The complaint was sustained and is permanently on their record. The other officer I rode along with carried stickers to give out to kids along their beat.