Hello, I am Demet Dagdelen.
I am an applied scientist at Automattic, where I build and lead the ML, experimentation, and agentic systems that move the numbers behind WordPress.com, Woo, Jetpack, Tumblr, and others. I’ve been there for 10+ years, first as a data scientist, now leading the teams that build the opinionated systems that make WordPress.com grow, grounded in statistical rigour and focused on true business impact.
Measurement & growth
Proved $35M was non-incremental.
I built GeoX, our geo-experimentation platform, after finding that standard attribution methods couldn’t tell us what our ad spend actually caused. It showed $35M/year was non-incremental; we then stopped all advertising for two years.
Rebuilt growth from first principles.
I then got to sit on the other side of my own result, running WordPress.com growth for a year. I led a 35+ person cross-functional team across engineering, design, data, and marketing, rebuilding the growth funnel from first principles: acquisition, activation, conversion, and retention (14% incremental lift; published industry results sit around 5-6%).
Systems that run themselves
Causal ML + lifecycle, automated.
I rebuilt WordPress.com’s largest campaigns as ML Sales: uplift models that email only the users an email can actually persuade, with an always-on holdout on every release. It nearly doubled incremental cash while emailing 70% fewer people, has run autonomously since 2023, and drives 25% of post-signup subscriptions.
Foundational data platforms.
Underneath all of this sit pipe, the self-serve ML platform I kicked off in 2017, now scoring 50M users a week; ExPlat, our Bayesian A/B testing platform, which I helped build and then spent years growing into a company-wide practice: experiment reviews, embedded data scientists, and results always reported with their uncertainty attached; and A8Cmail, our in-house email platform, designed to handle 500M+ sends a month with holdouts on by default and ML-based segments available to anyone.
Current focus
Agentic growth.
My current focus: agents that propose, run, and evaluate their own experiments across landing pages, ads, and lifecycle email. Most of my work there goes into evals, instrumentation, and tying what the agents produce to actual revenue. Underneath that sits the data work: making a decade of user, experiment, and revenue data self-serve, for people first and now for agents.
This blog is where I think out loud about all of it.
causal inference · incrementality · ML systems · safe, reliable, and useful agents