Senior Applied Scientist-Measurement

Applied scientists at TTD work closely with engineering throughout the lifecycle of the product,
from ideation to production and monitoring. Our applied scientists are end-to-end owners. You
will participate actively in all aspects of designing, researching, building, and delivering data-
focused products for our clients and traders.
Bringing objective, transparent measurement to the open internet is core to TTD’s strategy, and
retail conversion data is an essential part of that. This role owns that data. This particular role is
responsible for the research and application of state-of-the-art statistical and modeling
techniques to solve measurement problems centered around retail conversion data. This role
will be the team’s expert on retail data — understanding the ins and outs of every source we
use, how each is collected and where it can mislead, and owning the statistical integrity of these
datasets end to end. Day to day, this role will explore new data sources, build new-buyer and
conversion reports, project and scale attributed numbers up and down defensibly, impute
missing data, monitor data quality, and reason carefully about how retail data behaves inside
various measurement models. The work of this role ensures our retail measurement solutions
are statistically sound and robust, and it helps steer product decisions toward methods that hold
up.

 

The main job directions include:
- Own our retail conversion datasets end to end — be the team’s expert on these sources, how
they are collected, and where they can mislead.
- Explore and process data from a variety of retail sources, building the datasets and pipelines
that downstream measurement relies on.
- Apply rigorous statistics to scale and project our measurement numbers, impute missing data,
and produce new reports — always with careful attention to bias and uncertainty.
- Monitor data quality and the stability of our metrics, distinguishing real shifts from noise.
- Reason about how retail data feeds conversion lift, geo lift, and attribution, and validate that
our methods are robust on real data.
- Partner with cross-functional stakeholders and communicate learnings in compelling ways that
steer product toward statistically sound, robust solutions.
WHO WE ARE LOOKING FOR
• Proficient in Python, SQL, and PySpark, with a strong passion for enhancing and
expanding your technical skills. You are strong at data processing — exploring, cleaning,
and transforming large, messy datasets — and have a deep understanding of the
foundations of statistics, including estimation, sampling, and measurement error.
• Hands-on experience building statistical solutions and data pipelines at scale, with a
track record of owning a project end-to-end (from research to production) and
partnering with
a cross-functional team of scientists, engineers, and product managers. You are
comfortable becoming the go-to expert on a complex, messy dataset.
• A keen sense of data intuition and statistical rigor: you can reason about how a data
source biases a result, defend a number that cannot be directly measured with an
honest error bound, and tell a solution that is statistically sound and robust from one
that only looks good in the short term. Achievements like first-author publications or
clear project successes are a plus.

 

WHAT YOU BRING TO THE TABLE
We do not expect you to know every technology we use when you start at TTD. What we care
most about is that you can learn quickly and solve complex problems using the best tools for the
job. However, we find that the most successful candidates typically come in with something like
the following experience:
• BS/MS with 4+ years or a PhD with 2+ years of experience working in a DS or ML role
that involves bringing products from ideation to production.
• Experience working with retail, panel, survey, or conversion data, and reasoning about
how a data source biases a downstream estimate (match-rate composition, coverage and
panel skew, deduplication).
• Experience estimating population quantities from partial or biased samples: projecting
an observed or matched subset up to a full population via sample weighting or
calibration, and attaching honest error bounds (e.g. via resampling).
• Rigorous missing-data imputation practice that carries the added uncertainty through to
the final estimate, with the judgment to recognize when data is missing in a way no
imputation can fix.
• Proficient in Python and SQL.
• Experience building monitoring, anomaly detection, and data-quality checks on
production metrics and data feeds is a plus.
• Experience in causal inference and lift measurement is a plus.
• Experience in programmatic advertising is a plus.
• Experience running heavy workloads on a distributed computing cluster (especially EMR
or Databricks), leveraging technologies like Spark to work with large datasets preferred.
• The ability to communicate with diverse stakeholders, making architecture
recommendations, ensuring effective execution, and measuring quality of outcomes