One job, described two different ways.

For five years my job has been to sit between a messy dataset and somebody who has to make a decision by Friday. Banking, retail, supply chain: transaction tables, deposit funnels, regional sales, warehouse migrations. The work was always the same shape underneath. Find the signal, prove it isn't noise, and hand it over in a form somebody can act on. The dashboards and the SQL were never the point. The decisions were.

What changed is what I can hand over. The instincts that make a good analyst, being suspicious of a number until it survives a test and caring where the data came from, turn out to be exactly what production AI systems are missing. So I build them defensively. Retrieval that takes an as-of timestamp and has no default for it. Outcome labels sitting behind a database grant the simulation role does not hold. Runs seeded once and replayed with zero network calls, so a result can be reproduced rather than remembered.

I think of it as one continuous job rather than a career change. Analytics taught me what a defensible answer looks like; AI engineering is how I reach one faster and at a scale I could not manage by hand. Someone should be able to look at what I built, understand why it says what it says, and bet something real on it.