About

I’m Matthew Reda, a Bayesian measurement scientist. I sit with marketing and finance leaders, figure out what’s actually being decided, and build the smallest Bayesian system that can answer it honestly. Sometimes that’s a sixty-line model. Sometimes it’s a year-long platform. Either way, the goal is the same: tell people what the data supports, and — just as importantly — what it doesn’t.

This blog is where I think out loud about that work: marketing mix modeling, measurement, regression done carefully, and the occasional result that surprises everyone, me included.

How I got here

I studied physics at MIT, where I did undergraduate research on single-molecule microscopy — STORM imaging, point-spread-function fitting, Bayesian deconvolution. When every pixel starts as noise until you treat it as a measurement problem, you develop a particular instinct: write the model down first, decide what would falsify it, then look at the data. Same instinct now, smaller microscopes and bigger budgets.

Since then I’ve spent my career in measurement science:

• Senior Measurement Scientist at Choreograph (now WPP Media), 2024–present — building MMM and incrementality systems for portfolios in the high-eight-figure media-spend range, owning the methodology and the fitting infrastructure.
• Measurement & Modeling at EssenceMediacom, 2022–2024 — hierarchical regression for brand health, and the start of a migration from frequentist regressions to fully Bayesian posteriors.

What I believe about measurement

A few things I keep coming back to, and write about here:

• Start with the decision, not the data. A model exists to support a decision. If there’s no decision on the table, there’s no model — only homework.
• Pre-specify, then commit. Pick a likelihood, priors, and identification strategy before you see the fit. Specification shopping is the fastest way to get an answer that flatters the brief and survives nothing.
• It’s almost always the data. Most “modeling” problems are upstream data problems. Validate schema, coverage, and collinearity first.
• Diagnose before reporting. If $\hat{R}$, ESS, or divergences fail, the summary statistic is fiction — and a fiction shipped to a CFO is worse than no answer.
• Report what the data supports. HDIs, not point estimates. “We don’t know yet” is a complete sentence.

What I build

Most of my work is open source. A few of the projects I write about:

• mmm-framework — a pre-specified Bayesian marketing mix modeling toolkit on PyMC. See Building a Pre-Specified Bayesian MMM.
• atlas — budget optimization over any predictive model. See Atlas: Budget Optimization Over Any Model.
• BayesInsight — config-driven Bayesian modeling. See Bayesian Models as Configuration.

There’s a fuller tour of the work — methodology, projects, and how to engage — on my portfolio.

Elsewhere

• Portfolio: Bayesian measurement systems
• GitHub: @redam94
• Email: m.reda94@gmail.com

If you find a friend who lights up at the words “posterior predictive check,” hire them. In the meantime, thanks for reading.