In September, I sent this email to about 2,000 people across the industry:
Subject: What if you had the answers to tomorrow’s test?
Hi {{firstName}},
I know you’re slammed busy at {{companyName}}, so I’ll keep it short.
I’m the Head of Quantitative Research at Alphanume and our proprietary datasets exist solely to generate alpha, and not in a gimmicky “track parked cars via satellite images” kind of way. We believe that it’s extremely unlikely that using our datasets won’t result in the kinds of PnL you entered this industry for in the first place.
You and I both know everyone says that though, so we’d love to just send you a data sample so that you can see it for yourself.
If nothing on the catalog whets your appetite though, we have the full capacity to build out any dataset you can think of and bring you back a free sample in less than 72 hours.
Just shy of 18 people responded, of which 10 resulted in a call, and 5 have become our partners.
If you’ve run any systematic strategies or anything with a quantitative blend, you’re likely aware of how important data is and will be, and maybe, just maybe, you might have even considered selling some of those datasets yourself.
This will be a break from our usual research posts, where I want to share a transparent view into what it actually looks like to run a data business in quantitative finance, drawing on what I’ve actually run into at Alphanume.
I chose to talk about this because I think you guys, specifically the readers of this publication, have the highest likelihood of engaging in a similar practice. There’s virtually no real public information out there talking about this, so the least I can do is give you some hard-earned knowledge to start with.
So, with that being said, let’s get right into it.
You Have To Be a Bit Cracked
The big-picture model is actually pretty straightforward:
Find a factor and/or predictable market effect
Gather and organize the data to make it tradeable
Serve the data to end users
It’s simple, but in order to succeed, you have to make sure the market effect you’ve found is actually real.
Most people who understand data like this and how it’s used in trading are already extremely skeptical by default (“if it works, why wouldn’t you just trade it?”), so if you put out a product that doesn’t really predict anything or hold up out of sample, you will simply won’t get paid for long.
Now, as you likely already know, no one just wakes up and finds capturable market phenomena.
But let’s say you did, and you’ve got a killer factor that tells you with high confidence that a particular basket of stocks will fall by more than 5% intraday.
This may be true, but what if the majority of funds you pitch to have a long-only mandate and couldn’t trade the data even if they wanted to?
You would “simply” have to find another profitable market effect. One for long-only funds, one for commodity futures, one for index options, one for bonds, and so on.
After all, you’re trying to run a data business as opposed to just finding one edge and going all-in personally, so you need to have something that works for more than one brand of systematic trader.
This is naturally the toughest step as it typically takes years just to find one reliable edge, but if you’re cracked enough to have at least 5 of these, things do get a bit easier.
The Economics
Once you have your book of edges that are ironclad and undeniable, then it’s time to think about the economics of actually selling it.
What’s interesting about data is that the most important data point is tomorrow’s, so neither your costs nor your revenue are “one-off and never-again”.
The costs:
Upstream data providers
Although the final dataset may be proprietary, you still need an upstream source for exchange prices, historical and live orderbook quotes, and so on.
It often isn’t worth it to go straight to the exchange or SIP for data licensing, so the best path ends up being enterprise agreements with other data providers that already have the licensing and include support.
Speaking from experience, this is going to average ~$2-4K per month for stocks and options. It’s expensive relative to a personal plan because you’re paying for the redistribution rights and commercial usage.
Also speaking from experience, even if you just use a personal plan when starting out, your usage is monitored and people will start asking questions about why you’re making 1,000,000 requests a day.
Infra, infra, infra
I’m not going to bore you with technical nit-bits here, but a large part of this is an infrastructure-optimization game.
Surprisingly, this part is not so expensive monetarily, where we’re currently averaging about $250 a month on a mix of AWS EC2 + RDS.
It is expensive in time and brainpower though, as most of this stuff is hard, especially if you choose to carry the IT load with a small team, or even by yourself.
A lot of datasets are going to have to be built walk-forward and point-in-time, so if anything happens (an outage, you accidentally delete a table, etc.), you’ll just have to start over from scratch and wait another year to accumulate it again.
So it becomes a matter of ensuring there are backups for the backups, load balancers and rate limits for if someone is spamming an endpoint, notifications for when things inevitably error, and compute cost monitoring so AWS doesn’t bankrupt you.
The only reason our infra costs are so low is that we spent months migrating from sub-optimal configurations, then re-migrating when that didn’t work, migrating again, and so on.
The little things
Beyond that are the little things like payment processors, domain hosting, email services, and so on. These go with running any business and are typically inconsequential cost-wise.
So, now that you have the costs, we can talk about the revenue.
The revenue:
Retail/Individual
Selling to independent traders is the natural starting point, as most don’t have access to things like Barra factors or Bloomberg feeds, but that’s also what creates the first challenge.
Systematic individual traders that use data feeds are surprisingly, or unsurprisingly, a minority in the ecosystem.
Most individual traders either run a fully discretionary process or primarily use chart-based strategies, so even if you have a killer dataset for a strategy they can execute today, it just doesn’t fit in their wheelhouse.
This makes sense though, as a lot of things have to be true for someone to run systematic strategies:
Access to some capital in order to allocate to multiple strategies
Domain knowledge to understand the market effect the dataset captures
Technical competence to backtest and deploy strategies in an automated fashion
Although the minority systematic cohort has these, they’re highly skeptical of any service outside of core price data, and rightfully so. Social media feeds are awash in “paid signal groups” and “vibe-coded trading tools”, so don’t expect to just say “I have gold” and have people believe you.
No group is a monolith though. A few people in each one will see your value-offer, give it a try, and if your data is as good as you say, they’ll often stick with you.
Quant funds/Enterprise
Things change significantly here as the literal business model of these firms is to use datasets to make more money.
The biggest challenge here though is establishing credibility.
Back in the alt-data craze of the 2010s, there were hundreds of startups trying to pitch hedge funds satellite parking photos, credit card transaction data, and so on, shooting for the $100k+/yr deals you might’ve heard about.
A handful from this era survived, but as a result, there’s now a big “put your money where your mouth is” attitude, where funds will want things like a 90-day trial of the data before they make a decision.
For some of the heavy-hitter funds (>$10B), they go to great measures to be sure that everything checks out 100%.
Just a few weeks ago, I mentioned that one of our datasets started its history in 2024, much to the dismay of the data sourcing team who wanted at least 20 years of history.
That email from the top of the post is an example of how we get most of these calls to begin with.
Once you get past the slow cycles and objections though, this segment will make up the largest portion of your inflows, and the churn is generally significantly lower. Because there’s so much time and labor involved here, you can often justifiably price-in a premium for the effort to make things worthwhile.
Final Thoughts
In sum, if you’re a quant thinking about stepping into the data business; go for it.
It isn’t easy, and you’ll definitely have a lot of “I AM LITERALLY GIVING YOU THE CURE, WHY DON’T YOU TAKE IT” moments, but it’s one of the last “moats” in the AI era and if you’re good, you’ll definitely be rewarded for your service.
I also wanted to send a personal thank you to those of you reading who gave our data a shot. I appreciate and value your trust, and even if you don’t plan on being here forever, I’m grateful we were able to cross paths.
As always, thanks for reading, and I’ll see you in the next one.