Shuai Yuan

VP, Data Intelligence at Verve · London

Shuai YuanPh.D., Computer Science, UCL

I build data and machine learning organisations, and the real-time pricing systems that ad exchanges run on.

Real-time inference workflow. Online: a bid request passes through feature lookup and a model to a price, at over 4 million requests per second in under 20 milliseconds. Offline: auction outcomes flow as events into the lakehouse, feed training, and retrained models are deployed back to the online path. 4M+ requests per second under 20 ms Bid request Features ML models Price auction outcomes Events Lakehouse Training retrained models
The loop my team runs. Models price every bid request inside the auction. Outcomes stream back through Kafka and Flink into the lakehouse within 15 minutes, and retrained models return to serving.
14+ years
in computational advertising, from research to running the org
4M+ QPS
real-time inference served inside live auctions
30+ people
data scientists and ML engineers led across four countries
12+ papers
peer reviewed, including KDD full papers and a best paper award

About

I lead data engineering, machine learning engineering and data science as one team, because pricing models are only as good as the pipelines feeding them.

I started in research. My Ph.D. at UCL was on real-time bidding: how exchanges should set reserve prices and how buyers should bid against them. Since then I have built teams from zero at two startups, both acquired, and run a global organisation of more than 30 at FreeWheel, Comcast's video ad business.

Today I am VP, Data Intelligence at Verve, where my team owns the pricing algorithms, the real-time inference stack and the streaming data platform behind its supply-side exchanges.

  • LeadershipManager of managers. Teams across the US, UK, India and China. Hiring, career ladders, cloud budgets.
  • Pricing and auctionsDynamic margin and floor optimisation, bid shading, forecasting, reinforcement learning, demand shaping.
  • Data platformsFlink and Kafka streaming, BigQuery and Iceberg lakehouse, dbt, Airflow, low-latency model serving on GCP and AWS.

WorkLondon, 2010 to now

Verve

VP, Data Intelligence

Jun 2025 to now

  • Lead an organisation of 18 across data engineering, ML engineering and data science, with three engineering managers.
  • Shipped dynamic margin and floor price optimisation in production, lifting both revenue and net margin against control.
  • Built real-time inference that serves 4M+ QPS inside the auction at 99.99% reliability.
  • Cut network egress cost by 97.6% with demand shaping models, with no loss of fill rate or publisher revenue.
  • Rebuilt the streaming lakehouse (Flink, Kafka, BigQuery, Iceberg) to process 40 TB a day across 8 regions. Data freshness went from over 5 hours to under 15 minutes.

FreeWheel a Comcast company

Director of Data Science

Jan 2021 to Jun 2025

  • Ran a global organisation of 30+ data scientists and ML engineers through four managers, and doubled it in two years.
  • Owned the ML roadmap for the premium video ad marketplace: inventory forecasting, dynamic bidding, traffic shaping and incrementality measurement.
  • Governed a multi-million-dollar AWS and vendor budget.
  • Founded the GenAI initiative for 800+ engineers: LLM developer tooling, code assistants and evaluation guardrails.

Beeswax acquired by FreeWheel

Director of Data Science

Apr 2020 to Jan 2021

  • Built the ML engineering team from scratch at the first Bidder-as-a-Service platform.
  • Launched the Bid Shading product in four months. It reached $1M+ ARR.

MediaGamma acquired by Beeswax

VP of Data Science, earlier Lead Data Scientist

Sep 2014 to Apr 2020

  • Built the data science and engineering team from zero and led it through the acquisition.
  • Developed bidding models with classification and reinforcement learning that returned 30 to 50% better ROI than manual baselines.

Advisory

Data and ML advisor

2010 to 2020

  • Machine learning and auction work for AppNexus (now Xandr), Bright (acquired by LinkedIn), Illuma and Volo.

ResearchReal-time bidding, auctions, multi-agent systems

The full list of 12+ papers is on Google Scholar.

Education

Speaking

  • Sep 2024 MLOps in Real-Time BiddingODSC Europe
  • Oct 2018 Open ETL for Real-Time Decision MakingSpark+AI Summit Europe · Slides
  • Mar 2018 Enabling Deep Learning in Real-Time BiddingDeep Learning in Retail & Advertising Summit
  • 2015, 2016 Real-Time Bidding: Mechanisms & AlgorithmsTutorials at WSDM 2015 (slides) and ECIR 2016 (slides)

Contact

[email protected]

I am based in London. Write to me about pricing and auction problems, data platform work, speaking, or senior roles in AI and data.