Our reporting infrastructure hit a ceiling we chose a year ago. It's a small version of a pattern worth naming: funding the new while the foundations quietly degrade.
Sometimes it takes our reporting team three to four weeks to build a certain type of a new report. The constraint isn’t the team or talent therein, it’s the infrastructure underneath them. A year ago we had a choice: overhaul the data model, or buy time with patchwork, sharding and query tuning and caching. We chose patchwork. And, to be fair, it was a reasonable call at the time. The optimizations worked, right up until they didn’t, and now we’re at the ceiling. Latency still surfaces with customers. The export center we shipped last week helps at the edges, but it’s a band-aid on the same wound.
What’s been sitting with me since that conversation is how common this situation is, even at the biggest and baddest tech companies in Silicon Valley. We’re pouring investment into AI infrastructure, and we should be. But the foundations that everything else stands on, reporting, the design system, the unglamorous load-bearing systems, are quietly going unfunded. These systems don’t break loudly or get the kind of attention a revenue driving feature will get from the C-Suite. Instead, they degrade slowly and compound like paper cuts. Each quarter the patch holds, the report ships a little slower, the design system drifts a little further from the product, and nobody can point to the day it went wrong, because there wasn’t one.
The math behind these types of failures is also uncomfortable. A full reporting rebuild probably won’t get funded next year either, because the thing it competes against is always newer and more exciting.
Foundations lose that fight by default, even if they were funded once. If they appear to still work (and “still works” is a very expensive phrase in a roadmap conversation) it ends the discussion.
The single move I’m convinced helps in this scenario is when you actually do rebuild, rebuild for the architecture you’ll need in a year, not the one you’re escaping. If we ever get to redo this part of our reporting infra (and I think we will), it shouldn’t be a faster version of the old data model. It should be built AI-native from the floor up, so we’re not buying ourselves another five-year ceiling.
The teachable part
Audit what you’re treating as permanent. Every team has a foundation that got built once, still technically works, and hasn’t been funded since. The design system is the one most design leaders can name. The test isn’t whether it works today. It’s whether it’s degrading slowly enough that no single quarter ever makes the case to fix it. If that’s true, the case has to be made deliberately, by someone willing to spend capital on something that isn’t broken yet. That someone is usually you.
On the shelf
The shelf
