Why I’m starting this blog

meta
Notes on what I’ll write here, what I won’t, and why I think a writing habit matters more for a senior practitioner than it does for a junior one.
Author

Umut Altun

Published

February 7, 2023

I’ve been a senior data scientist for a few years now and I’ve avoided writing about the work publicly the entire time. That ends here. The reasons I avoided it were the usual ones — too busy, too proprietary, not enough new to say — and they were all weak. So this is the first post, and a brief note on what’s coming.

What I’ll write about

The work I do clusters around three areas: LTV and subscription prediction, marketing analytics (MMM, bid optimization, attribution), and agentic AI / LLM systems for data work. Posts will mostly be about one of those, occasionally about the cross-section.

Some things I’d actually like to write:

  • How a cohort-based LTV system handles the gap between “we have 30 days of data” and “we need a prediction for month 12.”
  • Why power-law fits keep beating neural nets on retention curves, and where they finally break down.
  • The honest version of when Marketing Mix Modeling is useful and when it’s hand-waving.
  • Building a Text2SQL agent that’s actually used in production — what works, what’s still broken.
  • The places where agentic AI is genuinely changing the shape of data work, and the places where it’s still cosplay.

What I won’t write

  • Generic “here’s how to do A/B testing” posts — there are already a thousand of those, written by people who explain it better than I will.
  • Anything that would expose proprietary data, architectures, or numbers from work I’ve done for employers or consulting clients. The systems I write about are real; the specifics will be abstracted enough to be safe.
  • Listicles, trend-chasing, AI hype. The half-life on that is about three weeks and I’d rather write things that hold up.

Why a writing habit matters more later in your career

This is the part I think about most. When I was junior, the case for blogging was straightforward — it builds your reputation, it’s a forcing function for learning, it gets you noticed. Fine.

But for a senior practitioner the calculus actually skews more in favor, not less:

  1. The thinking is the bottleneck, not the doing. When you stop being graded on whether your code runs and start being graded on whether your choices were the right ones, writing them down is the only way to check yourself.
  2. You accumulate strong opinions and never sanity-check them. Twenty colleagues nodding along in a meeting is not the same as a stranger telling you you’re wrong.
  3. Your defaults become invisible to you. When you’ve made the same kind of decision a hundred times, you stop noticing you’re making it. Writing forces you to surface the defaults so you can check whether they still apply.

That’s the bet. We’ll see if I cash it in.

If you have thoughts on any of this — or if there’s something specific from the three areas above you’d want me to write about first — email me.