The Economics of Data Businesses - by Abraham Thomas

A handful of business models dominate tech today: SaaS, marketplaces, e-commerce, on-demand, social networks and so on. Most of these business models have been studied widely, both their execution and their underlying dynamics.

But there’s one notable exception: data businesses. Despite the fact that many of the largest and most dominant tech firms in the world are data businesses, there are not many resources on the what, how and why of this business model.

This essay is an attempt to change that. Read on!

Every company uses data, but not every company is a data business. A company is a data business if, and only if, data is its core product. Data is central to the activity of the company; without the data, there is no company 1.

Google, Bloomberg, Yelp, and ZoomInfo are all data businesses. They acquire their data in different ways, and they generate revenue from that data in different ways. But for all these companies, data is the fundamental unit of value creation.

THE FIRST fundamental truth of data business models is this: it’s all about the data.

Successful data businesses are all built around a unique or proprietary data asset. There are a few ways to build such an asset:

Note that these methods are not mutually exclusive. If anything, they tend to be mutually reinforcing.

THE SECOND fundamental truth of data business models is this: whoever controls the data, captures the value. Intermediaries get squeezed.

A common failure mode is to build a business on top of somebody else’s data. If you depend on a single upstream source for your data inputs, they can simply raise prices until they capture all of the economics of your product. That’s a losing proposition.

So you should try to build your own primary data asset, or work with multiple upstream providers such that you’re not at the mercy of any single one.

You should also try to add proprietary value of your own, lest either your suppliers or your customers encroach and disintermediate you. A sufficiently large transformation of your source data is tantamount to creating a new data product of your own.

These tactics interact. Sometimes the very act of merging multiple datasets adds substantial value 2. Joining data correctly is hard! Other non-glamorous ways to add value include quality control, labelling and mapping, deduping, provenancing, and imposing data hygiene 3 4.

Some companies, discovering that they can neither control their data assets nor add intermediary value, pivot to picks-and-shovels instead. Tools to support data businesses — everything from monitoring to pipelines to governance — can be lucrative in their own right.

The gold-rush metaphor may be over-used, but it’s still valid. Prospecting is a lottery; picks-and-shovels has the best risk-reward; jewellers make a decent living; and a handful of gold-mine owners become fabulously rich.

THE THIRD fundamental truth of data businesses is this: they start slow.

You’ll that none of the above data acquisition methods are ‘easy’. They need upfront investment or a certain amount of scale to work. Absent either of those, building a data asset is a process of slow bootstrapping.

Adding to the problem is the fact that almost all data products have a ‘minimum viable corpus’ — a size below which the data simply isn't useful. This parallels the concept of a minimum viable product in software, but an MVC is usually much harder to build than an MVP.

The analogy with software doesn’t end there. Almost every apart of the software business stack has an equivalent in the data business stack. Where software firms invest in devops, QA and product, data firms have to invest in data ops, data QA, and data product. These tend to be just as complex, with the extra hurdle that third-party providers are rare, hence you often have to build this infra in-house. All of this is expensive.

As a result, delivering one's first data product requires significant time and resources.

And this is a good thing! Remember the classic wisdom: my capex is your barrier to entry. The effort required to go from zero-to-one in data businesses is one reason they are so formidably defensible. It's also why ‘brute force’ remains one of the most popular strategies used by players in this game. A data product that can be built easily is a data product that can be replicated easily.

But even after you build your data asset, you’re not home free. For reasons we’ll get into later, most data products require category creation. You have to educate your ecosystem, evangelize your product, nurture your customers over time. Early sales cycles are long, and win rates are low. But it gets better — a lot better.

One would think that a business model that requires substantial upfront investment but pays off in buckets later on would be a perfect fit for venture financing. That may have been true in earlier eras, but not today. Tech investing in recent years has indexed heavily on growth rates; as a result, data businesses — with their slow early growth — often find it difficult to raise venture capital.

(It’s also the case that compelling opportunities in data have historically been rarer than opportunities in software, even if they're more lucrative. VCs are familiar with outlier math, but their lack of reps evaluating data businesses tells against them.)

THE FOURTH fundamental truth of data businesses is this: they accelerate.

Everything starts slower on the data side. Building a valuable data asset takes time. Building the supporting infrastructure to actually deliver that data takes time. Sales cycles take time.

The classic mistake people make is to see this and jump to the conclusion that early-stage data businesses don't work and will never work.

But that's a category error, and it’s due to a fundamental difference in dynamics. Software business economics tend to degrade; data business economics tend to improve.

Why so? Here’s how it works:

Now, many of these effects taper off eventually. Marginal data costs go back up once you start hitting the long tail; price flattens out once the marginal data point no longer adds insight; most real-world entities have finite (even if large) cardinality. These curves are sigmoid, not unbounded.

But you can get a very long way before that happens. At-scale data businesses are huge, and many of them are still growing fast.

The holy grail is when all these scale and network effects combine such that you can be both the lowest-cost acquirer and the highest-paying buyer of your data inputs —while still offering your data outputs to customers at the lowest price in the market. If you get this far, you’re unstoppable.

ZoomInfo (NASDAQ:ZI) is a data business whose core asset is a directory — names, titles, contact information — of corporate employees. If you're a salesperson and you want to sell product X to company Y, ZoomInfo will help you identify and contact the exact person Z who you should be talking to.

Here are two fascinating interviews with Henry Schuck, the founder and CEO, conducted a decade apart.

The first one, from 2012, is all about the brute force early stage:

The second one, from 2021, is all about scale effects:

I want to break down all the different loops and motions in play here:

And all of this pays off:

Note that ZoomInfo did not raise venture financing; they were bootstrapped until a private equity round in 2014.

THE FIFTH fundamental truth of data businesses is this: they are virtually impossible to displace.

I’ve talked about this before 6:

The conventional wisdom in tech is that customers will switch to a new product if it’s 10x better than what they're currently using.

But what does ‘better’ mean for data?

If your data product is substantially similar to existing products, then one way to be ‘better’ is to offer a dramatically lower price. But this doesn’t work for data! The concept of minimum viable corpus means you need to be above a certain size and quality, else your data is almost worthless. And the very particular economics of data acquisition (brute force, economies of scale, data collection loops) mean that it's hard to build an MVC that’s dramatically cheaper than the incumbent offering. So ‘disruption from below’ rarely works.

Another option for being ‘better’ is to go the other way — offer dramatically higher quality. Typically, you need to be an order of magnitude faster, or more accurate, or more comprehensive — sometimes all three! — before you start poaching customers from the incumbents.

And even that might not suffice; the value of data lies in what can be done with it, and often, increasing speed or accuracy or comprehensiveness doesn’t really increase utility by enough. If you're competing with a table-stakes data product, heck, even increasing the utility by a huge amount won't work.

One way to think about this is in a ‘jobs-to-be-done’ framework. If a particular piece of software does a particular job, it's relatively easy to imagine a next-gen product doing that same job 10x faster or better or cheaper. But it's actually quite hard to envisage what an improved version of an existing data product can do that is 10x better than the existing data product.

In fact, typically, a successful new data product is not a variation on the existing data; it’s a brand new data asset from a completely different source, exploiting a completely different set of loops.

Meanwhile the original data asset continues to sell. Churn rates for mature data products are minuscule. So just as a matter of usage, it’s quite difficult to disrupt established data businesses.

Then there are all the ‘classic’ sources of defensibility mentioned previously: high barriers to entry; economies of scale in data acquisition and infrastructure; positive feedback loops / network effects in data and customer acquisition.

And finally, incumbents can use certain tactics to bolster their structural advantages:

‘Commoditize the complement’ is the best known — either with software, or with third-party data 7.

‘Creating data standards’ is another common one — every large data business has done this, from PCI to CUSIP.

‘If you can’t beat them, buy them’ — mature data businesses are extremely acquisitive. If they can buy data that is either additive or adjacent to their core product, they almost always will.

Put it all together, and mature data business have a set of flywheels and compounding structural advantages that are almost impossible to displace:

But data businesses aren’t quite winner-takes-all. Instead, the biggest ones tend to form duopolies — think Visa and Mastercard, Nasdaq and NYSE, Google and Facebook, Moody’s and S&P, Experian and Equifax, Bloomberg and Refinitiv. I suspect this has more to do with de facto government policy than with actual business dynamics.

Abraham Lincoln, Ulysses S. Grant, Grover Cleveland and William McKinley all worked for the same data company: Dun & Bradstreet.

Dun & Bradstreet (NYSE:DNB) is 180 years old. No business survives that long (two world wars and a civil war, recessions and depressions, market booms and busts, multiple technological revolutions, good and bad management, you name it) without an amazing structural moat. That’s the power of data businesses.

And DNB is far from the only centenarian data business in the world! Equifax was founded in 1899. Reuters was founded in 1851. Various Lloyd’s entities (the List, the Register, and the insurance mutual) were founded between 1686 and 1760. Standard & Poor’s dates back to 1860. All data businesses, all chugging along today, and none of them particularly close to being disrupted.

In SaaS a particular generation of software can become obsolete in a few years; products built in the early 2010s look decidedly old in the tooth today 8. But data businesses can last for decades, and as the above examples show, even centuries. Invest in building a data asset once, keep customers forever: the ratio of LTV to product creation costs is off the charts.

THE SIXTH fundamental truth of data businesses is this: they are rare.

You might think that starting, buying or investing in data businesses is a slam dunk. Unfortunately, it’s not — and that’s because successful data businesses are rare.

This shouldn’t come as a surprise, because valuable data is also rare. There's a lot of data in the world, but most of it is junk. (Sturgeon’s law: 99% of everything is junk). The list of criteria that a data asset must meet in order to be the core of a successful data business is long and extremely restrictive.

And of course, many ‘obviously valuable’ data assets (information about companies, say, or people) have already been captured. That leaves non-obvious data assets9 — which in turn implies category creation. Category creation is hard!

But (speculation alert!) I wonder if we might see a ‘deployment era’ for data just like we’re currently seeing in software. Instead of massive horizontal platforms covering entire information categories like people, companies, or events, I wonder if more targeted, bespoke, vertical- or application-focused data businesses will emerge — data for very specific niches. We shall see!

Toronto, 29 Jan 2022.

Kevin Kwok has a great essay on data content loops. This was a strategy we used to great effect at Quandl, the data business I co-founded. Kevin describes how Rich Barton used it thrice, at Expedia, Glassdoor and Zillow.

Auren Hoffman has a comprehensive Data-as-a-service Bible. I also recommend his World of DaaS podcast, especially his interviews with other data business founders.

I wish there were an essay with a truly in-depth treatment of Google as a data company (which it so clearly is!), not a search or advertising or content or AI company. If you come across such an essay, I'd love to read it.

This is my very first Substack post; thank you sincerely for reading this far! If you have any comments or suggestions, please send them in.

If you liked this post, please do 3 things straight away:

People sometimes use ‘data business’ and ‘data-as-a-service’ interchangeably, but strictly speaking DaaS is just one flavour of data business; there are many more.

Note that merely ‘repackaging’ data is rarely sufficient to generate strong economics. This is another common failure mode: thinking that dashboards and visualizations and interfaces are the key to a data business. Data consumption UX is a nice-to-have; data itself is the must-have.

Google, creator of some of the most powerful software in existence, is happy to open-source large chunks of their code, but they keep a tight rein on their data. I find this revealing.

Some of the most moat-y SaaS companies are the ones that have created a data asset: Salesforce's role as the system of record for most of its clients is perhaps the best example.

Pivotal is a newsletter about data, investing, and startups. It has three underlying themes: the data explosion, markets in everything, and software eating the world.