About
I started in digital marketing in 2014. Google Analytics felt like a superpower: finally, you could measure things. The more I looked at the data, the more I saw that the tools were counting, not answering. Which channel drove that pipeline? Did that campaign create demand, or just claim credit for it? Would those customers have converted anyway?
So I went deeper. I did not study statistics; I needed it, so I learned it. Same with Python, SQL, BigQuery, machine learning and causal inference. Every technical skill I have came from hitting a real problem I could not solve with what I knew.
I have consulted for companies in the USA, Canada, Mexico and Finland. Today I work in marketing analytics for B2B software, build TrueLift to hold my own forecasts accountable, and take on a small number of engagements where measurement actually matters. I am based in Kitchener-Waterloo, Ontario, and work with teams across the Toronto corridor and remotely.
Three things set the work apart. I do both the data foundations and the modeling on top of them. I work on B2B data: pipeline, CRM, long sales cycles, not consumer checkout. And I hold my own work accountable: forecasts are graded against what happened and decisions are recorded.
Selected work
Four engagements where measurement actually mattered. Teams are unnamed; the methods and the outcomes are as they happened.
A B2B tech marketing team needed to know which of their fifteen channels was driving pipeline, not just touching it. I built a multi-touch attribution analysis comparing four model families and surfaced a counterintuitive finding: one channel behaved like a step function. Below a touch threshold it contributed almost nothing; above it, conversion rates jumped.
The finding reshaped budget allocation.
Python · BigQuery · Shapley values · Bayesian modeling
A marketing team ran billboard campaigns across three Canadian cities and needed to know if they worked. I built counterfactual cities from a donor pool to isolate the causal lift in branded web traffic, with directional confidence intervals and a framework for designing measurement into future buys.
Genuine lift separated from background noise.
Synthetic control · causal inference · Python
A marketing ops team had email engagement in Marketo and web behaviour in GA4 and could not connect the two at the person level. I designed a Marketo lead ID to GA4 user ID mapping through UTM tokens and GTM, feeding one BigQuery layer that made post-email behaviour analysis possible for the first time.
No more reconciling spreadsheets. One queryable data layer.
GA4 · BigQuery · GTM · Marketo · LookML
A B2B SaaS team ran a 20%-off promotion across Q4 and saw a 34% lift in conversions. Leadership wanted to double down. The same month also brought budget season, a product launch and more ad spend. A synthetic control and difference-in-differences design built a counterfactual from comparable untreated segments and measured the bias directly.
True incremental lift was 12%, not 34%. The team avoided over-investing in discounts the next quarter.
Causal inference · synthetic control · diff-in-diff
Writing and research
Long-form pieces on causal inference, attribution and forecasting, and interactive studies you can run yourself, live at martinguzman.ca.
Start with the interactive proof of the Frisch-Waugh-Lovell theorem: adjust confounding strength and watch naive regression diverge from the true causal effect.