Fivetran Puts the Customer Last
Stop letting venture capital politics and consumption attribution games dictate how you design data infrastructure

It has come to my attention that Fivetran now offers the destination of Amazon Web Services Simple Storage Service. S3.
The Fivetran S3 revolution was not televised.

The Fivetran S3 revolution comes not from a rival vendor poking fun at Fivetran for finally offering S3 as a destination in 2023 when most ETL/ELT data integration vendors have had this in place for years. Heck, even most of the iPaaS and app integration and ops-oriented workflow builder products out there on the market that ELT vendors like to poke at for not being “modern” enough have had S3 as a destination for many years.
The Fivetran S3 revolution was announced by Fivetran, and it is just the latest in a long string of middle fingers at customers from Fivetran.
Let’s break down how Fivetran puts Snowflake (et al.) consumption attribution and venture capitalist politics before you, the Almighty Customer, then on the back end of this post we’ll look at other solutions.
Years of False Promises
Going back to about 2017 or so, Fivetran has sold the ‘Modern Data Stack’ along with Snowflake and Looker. This is the package deal that was sold to many ecommerce and SaaS startups from about 2017 to 2020 and they did a lot of joint GTM and partnerships work.
Snowflake needs the meter to start spinning to get customers activated and start collecting revenue. In their long tail of VC-backed startup customers, often there is not a data engineer on staff. Fivetran helps gets the meter spinning by getting data into the Snowflake playground where data can be joined and queried. Looker also spins the meter for Snowflake and is a quick way to show value and get a couple of dashboards built by a solutions engineering team doing a POC or by an in-house team selling the concept internally. Looker doesn’t need Snowflake per se, but it’s a good partnership because anyone investing in Snowflake over, say, a Postgres replica probably has some money to invest in data. Snowflake wins, Fivetran wins, Looker wins. The Modern Data Stack is born. At some point around 2019, dbt Labs gets into the GTM mix. Snowflake can spin even more consumption, plus dbt is a good way to roll up the heavily normalized set of tables Fivetran delivers and start stacking CTEs on Snowflake compute.
Now, here’s where it gets sillier:
Now, Fivetran during this whole time until quite literally last week has not offered AWS S3 as a destination. If ever called out on it, it’ll be some excuse about UPSERTs or “waiting until Apache Iceberg is good” or something, but that’s a bunch of baloney when S3 as a destination is offered by every other near competitor.
If Fivetran offered S3 as a destination the past few years, customers could architect something more cost efficient. This is bad for Snowflake consumption.
Snowflake, and to a lesser extent the BigQuery and Redshift teams, would not make as much money per account and therefore they would not be recommending Fivetran as much. Fivetran would lose out on new business and forward revenue streams.
If you want some real comedy, go to the Fivetran Support Portal about S3 as a destination.

For over two years there have been public, ongoing requests that Fivetran support S3 as a destination, with dozens of people chiming in and supporting this.
Of course, right off the bat, the request is motivated by “cost performance” and “architectural requirements” and per a Fivetran PM this is “on our short-term roadmap” 2 years ago.
And the thread just keeps going and going like the Energizer Bunny.

and going…

and going…this time with a release date that came and passed a year ago…

“Any update?”


Oh look!
Snowflake is now starting to release their own connectivity, offering solutions that compete directly with Fivetran and likely go after a set of some of Fivetran’s highest revenue and highest margin connectors. AWS too has announced Zero ETL.
Now, after 2 years, the Support Portal thread can finally come to a close, as Fivetran announced that S3 is the future now that the era of cloud warehouses kicking business over to them is starting to close up.
If Fivetran wanted to serve the customer first — and there’s obviously been demand for S3 as a destination given the dozens of public forum comments, not to mention plenty of customer requests I’ve personally seen and taken part in — Fivetran could have delivered this in the past then charged customers for it. Fivetran would win with the customer, and the customer would win with Fivetran, though Snowflake would lose a bit.
But Fivetran did not do this. Doing so would have soured their relationship with Snowflake (et al.) and Snowflake has always historically come before the customer for Fivetran.
The customer eats last, and it’s time for the adult customer to take back control.
Consumption Attribution: A Primer

In the six months since I published “How Fivetran + dbt actually fail” highlighting, among other things, Fivetran’s fixed, heavy normalized schema structures that they use to get business from cloud warehouse vendors by increasing costs to roll up (denormalize) and create joins after the data is landed, a number of data engineers and analytics types reached out to ask me about consumption attribution.
I’d figured most younger people don’t know this or think about it, but I was actually surprised how much pushback I got from clueless, angry analytics engineers that I’d inadvertently red-pilled them and shattered their world by letting them know that cloud warehouse vendors charging a consumption model want you to spend as much money as possible on their platform, and that it’s only getting sillier as cloud warehouse penetration of TAM exhausts and new activation slows, meaning existing customers getting soaked on NRR will be where revenues are disproportionately accrued in the near future.
If you 25 year-olds still want to argue with me, go download Snowflake’s most recent 10-K here. The word “consumption” shows up 97 times.
Learn how the world works. Learn how your industry works. Trust me, it will be good for your career.
Passing this consumption attribution to rent-seeking partners like Fivetran and dbt Labs is also why incredibly sloppy, bad stuff like this below public dbt forum post exists.
Some poor sap used Fivetran and dbt Labs together, ran this using Fivetran’s methodology that Fivetran recommends on their website, and is now facing “extreme costs” on Snowflake.

My friend, it’s working exactly as designed.
This is batch gone mad. Now, instead of solving business problems and doing strategic work with data there is extra time, money, and human effort disproportionately going toward fiddling around and tinkering with a Goldilocks set-up of not-too-frequent of refreshes but frequent enough.
You believed these vendors. These vendors have been incentivized to increase cloud warehouse billings and now you have “extreme costs” on Snowflake. Now you have to decide whether to refactor this, whether you ignore it and eat the cost, whether you allocate your time later and eat the costs now vs. eating the costs later.
Of course they are going to soak you.
This doesn’t even make sense. It really doesn’t. Changing the language of what normalization of schemas means is just pure propaganda.
Stop Letting Venture Capital Politics Dictate Your Data Stack

One of the more ridiculous trends to emerge the past few years is what I like to call “Poltergeisting” which is when venture capitalists or paid venture capital influencers use Twitter, “community” Slack groups (which are mostly all run by venture capitalists), or other industry groups to lure unsuspecting 25 year-olds into bad architecture decisions.
These venture capitalists have the attention spans of goldfish, and that’s by design.
It’s better for them to invest immediately in something, throw millions of dollars of other peoples’ money into something, use that money to hire influencers immediately and pump, pump, pump the trend to get another venture capital firm to mark it up, and quite possibly there’s not even a product in market.
The new Andreessen Horowitz thing is now “composable OLAP”… or something. Now you can use Andreessen Horowitz-backed Fivetran with their brand new S3 destination they sat on for years that magically appeared out of nowhere, Andreessen Horowitz-backed dbt, and Andreessen Horowitz-backed MotherDuck/DuckDB to make a whole loop of Andreessen Horowitz in your business.

So far, I haven’t heard much in the way of what incremental business value a “composable OLAP” provides, what incremental insights it can unlock, or what incremental time to delivery this can save. It seems to be just Silly Toys for Silly Boys to play with and further remove themselves from business value on the company dime.
At best, the only argument seems to be “the Andreessen Horowitz composable OLAP will save you on Snowflake bills” and frankly the Snowflake bills of these SaaS and ecommerce companies are a function of how bananas they are going with dbt models and 3 BI tools.
It’s a tool chain of fools.
Andreessen Horowitz has a new play to back. They’ve got to get the growth metrics juiced on DuckDB and MotherDuck if MotherDuck ever gets a commercially available product to market, and now Fivetran can all of a sudden support the very basic S3 as a destination.
What to Do?
As a data leader or data architect or an individual contributor trying to get your career going around this stuff, let me tell you the best piece of advice you’re going to hear from anyone: you’ve got to get away from this cartel venture capital crap of 5–6 VC firms dictating how data “should” work.
All of the talking heads around this paradigm keep trying to come up with new ways for customers to spend more money through new things to sell. They make claims that say people aren’t offering solutions, when the real solution is not to lean too heavily into all the nonsense.
When your organization’s connectivity is dictated by the politics and whims of large venture capital firms and cloud warehouse vendor consumption games, you’ve lost and now you have to dig out of a hole just to get a workflow built or to put a number on a dashboard.
Here’s what I recommend: think of your data investment as an actual portfolio of investments.
Wild, I know. If you start looking at your cloud spend, human capital, and products as a portfolio of investments that generate returns, you’ll start to develop habits that lead you away from these Modern Data Stack games.
You should be hedging your portfolio of data investments as well. Wild, I know. If you’re using Fivetran, dbt Labs, Databricks, Census, and Hex, and now you’re using MotherDuck influencer suggestions to try out DuckDB, for example, you’re on the Andreessen Horowitz Modern Data Stack. Your data investments at your firm are not hedged, in fact you are over-leveraged. Your connectivity gaps and the product misses across the tool chain are leveraged to Andreessen Horowitz returns and politics, not to what is best for your business. But very few think like this. I think it’s time to start.
If you want to offload integration work, it’s actually probably better to pay more money and be an actual partner with your vendors than to put up with Fivetran’s VC and Snowflake games that keep S3 off the roadmap for years.
This way, you can hire fewer analytics engineers and Airflow (or whatever) people. It’s actually cheaper to use stronger APIs and stronger middleware vs. human APIs and human middleware to manage weaker APIs and weaker middleware. Few people understand this outside of experienced CIOs and Heads of Data who work in larger enterprises.
Keboola can be a great option, Nexla and SnapLogic too, which offer broader offerings than Fivetran. Anything that gets your business away from these games is a step in the right direction.
In fact, anything is better than Fivetran. If you’re using Fivetran for database replication like Postgres or volume-heavy marketing or event platforms to a data warehouse — why? There is no reason for it. They are not able to support real CDC latency requirements, and their pricing model is structured to soak you as it is row-based. They’re not the best at anything and they’re not the cheapest at anything. In fact, they are like the fifth-best at everything.
If Fivetran were an employee you’d fire them. Just look at this S3 thing. They dragged their feet and made excuses for years, they overpromised fake timelines for two years to their customers, then all of a sudden once they’ve gotten pressure and their investors have a new thing to push, magically the work gets done. Give me a break.
Don’t be a Muppet. There is absolutely nothing Fivetran offers that can’t be done in a similar tool that plays fewer VC and consumption attribution games and has better SLA adherence and delivery, for either lower cost and/or better delivery of services.
Fivetran shouldn’t impress you much. It’s time to grow up and move on.