BI wasteland

June 24th, 2020

Photo by Chester Ho

Yet another BI tool acquired by database.

2019: Cloudera bought Arcadia Data, Snowflake bought Numeracy, Sisense bought Periscope. Salesforce bought Tableau.

2020: BigQuery bought Looker, Databricks bought Redash.

At this point, every database company has bought a BI tool (I don’t count Microsoft and Amazon as database companies, though funnily, I do think of Salesforce as a hosted, ‘nocode’ database at this point). There’s some serious consolidation going on here, but it’s vertical instead of horizontal. BI tools aren’t buying each other up, databases buying up BI tools. The rapid success is probably from a Domino effect of founders and investors wanting to exit the market. While we once believed that a strong BI tool might come to dominate, much like Salesforce for CRM, it’s become obvious that this market is nearly impossible to ‘win’ because it lacks moats. The moat was in the database, not the dashboards.

To give you some idea, when I joined, Numeracy had 9 customers. Six months later, we had 27 paid customers. Why did we still sell if we were able to triple customers in six months? Our market had no win condition.

Win conditions are a gaming term, especially in games with more than two players. They are a target you set to win the game. Most games have several ways of achieving victory, and a key part of winning isn’t trying to do everything, but focusing all your resoures towards a specific goal that you can hopefully do more efficiently than anyone else. It’s rarely about beating down others, because conflict is expensive and wasteful.

That’s why in startups, it’s rarely about copying off or stealing customers from your competitors, but focusing on parts of the market until you are unbeatable in that segment and then ceaselessly growing that segment.

There are two common win conditions: exponential user or sales growth. We pushed for exponential user growth with marketing and sales motions but in the end, the market and product did not support exponential user growth. We weren’t building up a network effects, nor did we have such cutting edge dashboard technology (akin to Google Search or iPhone) that word-of-mouth would bring us exponential growth.1

By the time we realized this, it was too late to switch to sales growth. Building sales teams isn’t dissimilar to building products — there’s ton of infrastructure, trial-and-error, personnel, and takes years to get right. On top of that, you’ll need to fundamentally rebuild your product to support enterprise workflows, auditing, and security needs. Having spent more than a year at Snowflake integrating Numeracy, I can attest to its complexity. Just like in a game, switching win conditions halfway through almost never works.2

Most startups in this space have gone through the same conflict. Some have pivoted to the sales model and fundraised enough capital to make it through to the other side. Even then, their journeys are far from over. I would not be suprised if Mode, Metabase, Sigma or Chartio weren’t also acquired before year’s end.3 At this point, the bottom of the market is too crowded with cheap/free alternatives and the top of market with products with a decade of entrenchment. At this point, breaking into this market will require 1. a radical technological breakthrough and/or 2. untapped distribution model, neither of which I see today.

  1. Lots of people have tried to invent “deeper” technologies in this segment. LookML, Chartio or Sigma’s ‘nocode’ query builders are good examples. These features are differentiators, but they aren’t moats. For customers, they’re still “nice to haves” because while they might allow the SQL-illiterate to do analyses, the buyers in this segment are almost always SQL-literate and the SQL illiterate users often prefer Salesforce over any other tool for doing analyses. 

  2. After a year and a half inside Snowflake, we absolutely could not have turned, on a dime, to enterprise sales. To beat out incumbents in enterprise sales deals, you can’t just walk in, having filled out all the same checkmarks. It might work for your initial network, but eventually you need to overcome the “Who are these guys?” question. To do this, you either have to be radically cheaper than your competition (while still checking all the boxes) or you need a 2 year+ advantage.

    Figma built such an advantage. They leveraged the new technologies of WebGL and CDRTs to build a collaborative, web-based design tool. From that, they’ve gained a foothold into tens of thousands of customers and can now build out workflow tooling at their leisure. They obviously also get a foothold in companies at the seed stage and hold much less sensitive data. Years later, I don’t think anyone can quite match Figma.

    Snowflake built such an advantage. They leveraged the new platforms of S3 and EC2 to transparently separate compute and storage not only at a product level, but in pricing. This advantage has given them years to strip customers away from Cloudera, Teradata, Redshift. In fact, Snowflake is now building BI tools and data catalogs. Except for BigQuery (which no one on AWS could economically or practically use), we’re only seeing meaningful duplication this year (with BigQuery Omni and Redshift RA3), 5 years since Snowflake went GA.

    Numeracy didn’t have that. Worse yet, we had no guesses on what that might be. We had a better product, but the procurement process, your users have just one seat at the table and they never have veto power. That power rests with the security, finance, or executive teams. So without a bang-on-the-table level feature, you are constantly fighting an uphill battle. 

  3. I wrote this pre-COVID. The radical valuations this year totally invalidate my assumption here. My guess, instead, is at least one of those companies will be acquired in 2021. Edit Jan 13, 2022: turns out it was