As someone who spends a lot of time in this space, their only "killer app" is automated workload / data distribution management. Which is cool, and hard to get right, but clearly something the cloud vendors and other data players are have taken steps towards / offer more or less the same outcomes.

And in contrast their Silicon Valley roots means a lot of their tooling/UX/data capabilities are ... undercooked. Their Web IDE feels like a throwback to 2003 Hadoop, their ETL capabilities are a joke, they don't support joins in views ...

And they've also squandered some opportunities to actually offer a differentiating "all in one" data processing experience for ad hoc/exploratory, BI/aggregated, and Big Data/AI/ML model crunching. For example, here's their garbage blog post on Spark SQL - https://www.snowflake.com/blog/snowflake-spark-part-2-pushin...

tl;dr when someone writes a Spark job that includes a filter against data in Snowflake, it's more efficient to let Snowflake filter the data before shipping it off to the (much more performant) Spark engine to do the actual analytical pieces of the query plan, instead of just shipping all the data over and letting Spark do the filtering.

Like ... wow, predicate pushdown is your answer?

Contrast with Azure Synapse providing Spark and SQL Server compute in the same environment; Databricks adding Delta Lake capabilities to be more schema-on-write friendly; Dremio building AI into their caching, and Starburst into their workload management ...

Anyway, I don't see any secret sauce, which means it's still just traditional enterprise sales cycles...