How do you work with someone whose working style clashes with yours?

The architect who owned our warehouse models works almost entirely in writing. I default to walking over and asking, and for my first few weeks I read his short replies and long delays as him not wanting to be bothered. What it actually cost was time: a question about the grain of a customer dimension took me nine days to get resolved across a scatter of half-answered messages, and I had built against a guess in the meantime. So I tried it his way. Instead of a one-line question I wrote the query I was running, the row counts I expected, the counts I was getting, and the two interpretations I could see, and asked him to pick one. He answered inside a morning, with the reasoning, and pointed out that my dimension join would have double-counted returns — roughly 58,400 extra rows a night once it hit the nightly load. The one thing I asked for back was that when he rejected an approach he add a sentence on why, because a bare no left me guessing. He does. Since then my questions to him get answered the same day, and I write the expected numbers down before I ask anyone anything, which turns out to be useful with everyone.

I ran the ingestion side of a warehouse migration with a counterpart who would not accept a decision that had not been through a written proposal with a comment period. My team lived in tickets and short calls, and during the scope reduction that mismatch started costing us: the previous round of trimming had taken six weeks, mostly waiting for comment threads to settle, and my engineers were rebuilding against models that were still under debate. I stopped fighting the format and adopted a smaller version of it. I wrote a one-page template — the decision, the options, the reconciliation impact, a named decider, a date — and committed to circulating it two working days before any cut was made. In exchange I asked for one thing: that we close each one on a twenty-minute call rather than in comments, with whatever was unresolved going to the named decider on the spot. He agreed to try it for the next round. That round closed in eleven days rather than six weeks, we retired twelve of the thirty-one candidate models with both teams signed off, and drift on the shared marts stayed under 0.3 percent through the migration. We have kept the template, and I have stopped assuming that a request for something in writing is a delaying tactic.