If you're new here: this newsletter is about what it actually takes to make AI work at the infrastructure layer. Not trends. Not model comparisons. The unglamorous, foundational work that separates AI that changes how a company operates from AI that lives in a slide deck.
A while back, I sat down with the person running our BI function and asked what the team actually spent their time on.
She pulled up a backlog. One hundred fifty-four open requests. Dashboards, reports, one-off data pulls. The newest request was four hours old. The oldest was almost nine months old.
The team had five people, all talented. They were working as fast as they could. The backlog was growing anyway.
That's not a capacity problem. That's an org design problem. And it shows up inside almost every company I've seen trying to get serious about AI.
Most companies have a BI team when they need a data engineering team. The distinction matters.
A BI team takes requests from the business and produces dashboards. They're reactive, ordered by what the business asks for this week. The team's job is to close tickets. They're measured by turnaround time.
A data engineering team builds the platform underneath. They model data so it's trustworthy at the source. They build pipelines that don't need someone in the middle cleaning the output. Instead of three teams with three parallel spreadsheets that diverge by Friday, they build one source of record three teams can trust.
The first team makes existing decisions marginally faster. The second team changes what decisions are possible.
The problem gets worse under AI pressure.
When a company starts an AI initiative, they go to the data team: we need data for the model. The BI team gets assigned. They run the request queue. Someone asks for training data, they pull what exists, export it, hand it over.
The model trains on whatever got captured, in whatever format it was captured in, by whatever process happened to exist before anyone asked the question.
That's not a foundation. That's archaeology.
The same pattern shows up in almost every company I've been inside. The decisions that actually run the operation live in parallel spreadsheets — three teams, three files, updated by different people on different schedules. By Monday morning, nobody agrees which version is current. Operational events generate no record: a phone call, maybe a note, maybe not.
The BI team can build a dashboard from any of that. The dashboard will be real data, well-visualized, meaningfully wrong.
The Foundation-First Sequence starts before the technology question. The first step is Audit: map the decisions that actually drive the operation, trace backward to find what data each one requires, and check whether that data exists in a form you can trust.
Most of what surfaces in that audit isn't something the BI team can close with a new dashboard. It's structural. A capture point that doesn't exist. A pipeline that runs through a person. A platform that requires expert interpretation to be useful. Those gaps are data engineering work, not report work.
The org design question follows from what the audit finds. A backlog growing faster than the team can close it looks like a capacity problem. But when the underlying data doesn't land in a system, when decisions happen off the record, when nobody trusts the platform — that's an org design problem. Solving one with the answer to the other doesn't work.
One question to ask this week: if your data team closed every open ticket in the backlog, would you trust what's underneath well enough to train a model on it?
This is where the work actually lives. The Foundation Letter is about that work. If it's useful, subscribe — it's free.
The Foundation Letter publishes monthly. One idea from inside real transformations, not from a conference stage.
P.S. If you want to run the full audit yourself, the Foundation-First Playbook is the structured version: 41 questions across all eight domains, a scoring rubric, domain-by-domain gap analysis, and a 90-day roadmap builder. Founding member price is $147 through the first 30 copies (full price $247). foundationfirsthq.com/the-playbook.html