Executive Summary
You've spent a good deal of money on a modern data stack (Snowflake, dbt, Looker), but your business users still queue up for custom reports and export everything into a spreadsheet.
The return on that expensive stack isn't showing up. Your engineers are acting as a service desk, getting bogged down in repetitive work, while the business makes slow decisions with data nobody quite trusts.
The problem isn't usually the tools or the users. It's a lack of trust. The fix is often structural: putting a governed semantic layer in place and carefully curating your reports to restore confidence in the numbers.
Why a new BI tool doesn't automatically lead to self-service
It's a common story. You've done everything by the book. You moved from an older system to Snowflake. You hired good analytics engineers to build clean dbt models. You bought a top-tier visualisation tool, whether that's Looker, Tableau, or Power BI. And yet, the true self-serve analytics you were hoping for hasn't quite appeared.
Instead, what you often have is the old chaos, just in a new system. You've moved the same problems to the cloud, which just means you can now generate confusing, contradictory reports more quickly. The commercial team has its number for 'Active Customers', and Finance has a different one. The result is that important meetings often start with a twenty-minute discussion about whose data is correct.
The difficulty is that technology on its own doesn't tend to solve human problems. The issue isn't really technical. It's usually a problem of confidence.
The problem is usually about trust, not technology
Your colleagues in the business are sensible people. They go back to what they know and trust. When they choose a spreadsheet over your expensive BI platform, it's not out of a deep love for VLOOKUPs. It's because they don't believe the numbers in the dashboard. And if we're honest, they may have a point.
I've seen this pattern in most of the scale-ups I've worked with. The data team builds technically perfect dbt models that nobody trusts. Why? Because the core business logic is inconsistent. The definition of 'churn' or 'net revenue' hasn't been agreed across the business and then set in stone. Without a solid Data Governance strategy, your dashboard is just a very convincing, but still inaccurate, picture.
A common sign of this is when people keep reverting to Excel. Every time someone exports to a spreadsheet, it's a small vote of no confidence in your setup. It points to a breakdown in Data Trust.
I was recently looking at the Looker setup for a fintech scale-up. They had over 2,000 dashboards. A quick check showed that 60% of them hadn't been looked at in over six months. At the same time, the data team's backlog was full of requests for even more dashboards. This is a common pattern: a lot of activity that looks like progress, but which just adds to the confusion and makes it harder for people to trust any of the numbers.
The way forward: Curate reports, don't just create more
The way out of this isn't to add more, but often to take things away. The goal is to shift from being a 'dashboard factory' to being a curator of trusted information.
This process is about changing your team from a reactive service desk into a proactive function that owns the quality and usefulness of data as a product. You can stop the chaos for businesses suffering from dashboard sprawl and start to rebuild faith in the data.
This is often a political challenge, not a technical one
You should expect some resistance. When you suggest removing a report that a department head likes, it can be a sensitive conversation. They are used to their version of the truth, even if it's based on questionable data. It helps enormously to have backing from a senior leader, like the CFO or COO.
You may need to slow down for a few weeks to get this right, but it lets you move much faster later on. The process of defining metrics and curating reports can feel like a step backwards, but it's the only way I've seen to build the foundation needed for proper, scalable self-service.
The result is that the Monday morning arguments about whose numbers are right tend to stop. Your team is freed up from constant ad-hoc requests to work on more valuable projects. And your expensive, modern data stack finally starts to show a proper return.