.svg)
Stop P-hacking Your Data Strategy with Eric Gonzalez
Most data leaders get the order wrong when building their data strategy. They buy the technology first, patch the people problems second, and finally address the broken processes that got them there in the first place. It's a pattern that plagues enterprise data management across industries.
Eric Gonzalez, founder of Omnificity and veteran fractional data leader, has seen this pattern play out across enterprises from financial services to healthcare. On this episode of Crunching Data, Eric shares why data strategy starts with people, not platforms, how neuroscience principles apply to building teams, and why 90% of companies don't actually need AI yet.
If you're leading a data team or trying to drive change in a stubborn organization, this conversation will reframe how you think about everything.
Here are some areas we dive into:
The scientific method is missing from most data strategies
Eric's neuroscience background taught him something most business leaders forget: you can't p-hack your way to good strategy.
"A lot of organizations today try to do p-hacking. They try to identify the hypothesis before making it and then having a statistically significant result because you basically already predicted the ending."
Instead of cherry-picking data to support predetermined conclusions, Eric advocates for true hypothesis testing. Gather observations, form a hypothesis, test it rigorously, and be willing to pivot 180 degrees if the data says you're wrong. That intellectual honesty is what separates strategy from theater.
Leadership is about listening, not plugging holes
When building data teams, most leaders promote their best developer into management. Eric calls this a critical mistake.
"If you just put your best developer as the data leader, you've probably taken 50 to 75% of their time and now dedicated it to more administrative work. You may instead have somebody who maybe is your second, third, or fourth individual on the team in terms of coding proficiency, but really has a knack for leadership and a desire to be a leader."
The fix? Have actual conversations with your team about what they want. Match people's motivations to the work that needs doing. Some engineers want to go deep on code. Others crave leadership. Neither is wrong, but forcing square pegs into round holes kills retention and performance. This approach to data team leadership is what separates teams that retain talent from those that lose it.
Self-serve analytics is overrated. Self-fed analytics is the future.
Eric has a controversial take: self-serve analytics doesn't work.
"A lot of people over index on, well, if you build it, they will come. You're telling those people that if you just build this semantic layer that somebody can access, that they're just gonna magically know how to work in Tableau or Power BI."
His alternative? Self-fed analytics. Instead of dumping tools on business users and hoping they figure it out, proactively send them alerts, subscriptions, and automated insights. When something important happens, notify them. Then they'll actually engage with your dashboards because they understand the context.
Data observability and cataloging are massively underrated
Eric has watched organizations spend six months building data lineage in Excel, only to have it outdated within an hour of publication. It's cartography for a constantly shifting landscape.
"People are spending two to three FTEs worth of time because they don't want to spend 10K a year, 50K a year, 100K a year on a tool. They're like, no, that's too expensive. So we're going to have a full-time person maintaining this, which means that it's always out of date."
Modern data observability tools can automate this. Yet companies balk at the price while hemorrhaging 2-3 full salaries on manual maintenance. The math doesn't math.
90% of organizations don't actually need AI
Eric's most controversial opinion? Most companies chasing AI aren't ready for it.
"If you actually look at what people are requesting, they want some simple linear regression model or they want some simple automation for their operations. They don't really need AI."
Before you can leverage AI at an enterprise level, you need integrated data, resolved process gaps, and solid fundamentals. You can run narrow AI use cases in parallel, but real transformation requires fixing the foundation first. Otherwise, you're just applying expensive technology to broken systems.
Numbers don't lie, but you need to tell the story
When Eric needed a data visualization tool early in his career and got told no, he didn't give up. He found Power BI's free version, built the reports anyway, and showed his VP the math: 30 minutes of manual work reduced to 11 seconds of automation.
"My favorite thing about numbers is numbers don't lie. So people can lie with numbers, but using the data to create or to tell your story is how you get buy-in."
Most leaders say no out of fear or ego. Your job is to make the unknown known, to show rather than tell, and to frame innovations as amplifying what already works rather than replacing what someone built.
Watch the full episode here.
Watch more Crunching Data
Here are some other episodes you might be interested in
