It usually starts with a Slack message on a Tuesday morning. Your data engineer — the one who built all the pipelines, wrote all the transformation logic, and is the only person who understands how the dashboards actually work — just gave two weeks' notice.
Suddenly, every report your leadership team relies on has an expiration date. The infrastructure that took a year to build is about to become an unmaintained black box. And you're starting a recruiting process that will take months.
This scenario plays out constantly at mid-market companies. And it's almost entirely preventable.
The Concentration Problem
Most mid-market companies have one, maybe two people responsible for their entire data infrastructure. These are typically talented, well-compensated engineers who built the stack from scratch. They chose the tools, designed the architecture, wrote the transformation logic, and wired everything together.
The problem is that all of this knowledge lives in their heads. The dbt models might be in version control, but the reasoning behind hundreds of small decisions — why this metric excludes returns but that one includes them, why this source syncs hourly but that one syncs daily, why there's a manual adjustment for Q3 2023 data — is undocumented.
When that person leaves, you don't just lose an employee. You lose the operating manual for your data infrastructure.
The Scramble
Here's what typically happens next:
Weeks 1-2: Everything still runs because the pipelines are automated. Leadership doesn't notice yet.
Weeks 3-6: Something breaks. A source API changes, a schema migration fails, or a dashboard starts showing incorrect numbers. Nobody knows how to fix it. The former engineer is unresponsive or already at their new job.
Months 2-4: You're recruiting a replacement while the existing infrastructure slowly degrades. Reports that used to be trusted now come with asterisks. The finance team reverts to spreadsheets.
Months 4-6: The new hire starts. They spend their first two months trying to understand what the previous person built. They want to rebuild half of it because they'd have made different architectural choices. You're effectively starting over.
Total time without reliable data infrastructure: six months. Total cost including recruiting, onboarding, lost productivity, and degraded decision-making: significantly more than the original engineer's salary.
Why This Keeps Happening
The root cause isn't bad planning. It's a structural mismatch between the scope of the problem and the staffing model used to address it.
Running a production data stack requires at least three distinct skill sets: data engineering (pipelines, infrastructure), analytics engineering (business logic, modeling), and BI/visualization. At large companies, these are separate roles on a team with management, documentation standards, and knowledge sharing practices.
At mid-market companies, all three roles collapse into one or two people. There's no redundancy, no knowledge transfer, and no continuity plan. It works until it doesn't.
The Alternative
A managed data stack eliminates concentration risk entirely. Instead of one person who knows everything, you get a team with documented processes, shared institutional knowledge, and built-in redundancy. Nobody quits and takes your data infrastructure with them because no single person owns it.
The transformation logic is version-controlled and documented. The pipeline configurations are standardized. The monitoring and alerting is managed. When a source API changes at 2 AM, someone is already handling it before you wake up.
And critically — the knowledge of how your business metrics are calculated, why certain architectural decisions were made, and how the various systems connect is maintained as institutional knowledge within the team, not as tribal knowledge in one person's head.
Prevention, Not Recovery
If you're reading this because your data person just quit, we can help you recover. But the better conversation is the one that happens before that Tuesday morning Slack message.
The question to ask yourself: if your data person gave notice tomorrow, how long until your reporting infrastructure becomes unreliable? If the answer is anything less than “it wouldn't matter,” you have a concentration risk worth addressing.