< Back

Building Upwork's First Marketing Mix Model Across an $88M Media Portfolio

Marketing Science • MMM • Attribution • Executive Analytics

Key Results

$88M

ANNUAL MEDIA PORTFOLIO

25%

MEDIA BUDGET REDUCTION AS SALES GREW

1st

MARKETING MIX MODEL IN COMPANY HISTORY

At a glance. Client: Upwork. Role: Principal Brand Analyst. Scope: an $88M annual media portfolio across digital, TV, radio, and out-of-home. Methods: marketing mix modeling, multi-touch attribution, segmentation, experimentation, and executive dashboards. Result: a 25% media budget reduction while sales continued to grow.

Executive Summary

Upwork was investing $88 million a year in media across digital, TV, radio, and out-of-home, and leadership needed to know what that spending actually produced. As Principal Brand Analyst, I built the company's first marketing mix model (MMM), along with the attribution, segmentation, and reporting infrastructure that turned it into a working decision tool for senior leadership.

The Challenge

At that scale, every channel has its own reporting, and each platform makes a confident case for its own contribution. The walled gardens claim credit generously, last-click attribution favors whatever happens closest to the sale, and brand media is almost invisible in a dashboard. Leadership needed one measurement view that could compare local and national media, digital and broadcast, and brand and performance spending on the same terms, then show where the next dollar would do the most work.

Research and Diagnosis

I built the brand awareness analytics infrastructure from scratch and owned measurement across paid digital, out-of-home, TV, and radio. Marketing mix modeling estimated each channel's incremental contribution to business outcomes, and multi-touch attribution filled in the digital detail the model can't see. When COVID and the shift to remote work changed demand for freelance talent almost overnight, I built impact models so the budget could respond to what customers were actually doing.

What We Did

I modeled financial returns at the segment level to sharpen targeting and acquisition efficiency. I designed and ran A/B experiments on creative, messaging, audiences, and channel mix. I built Looker executive dashboards that the C-suite reviewed weekly, which set the rhythm for cross-functional performance reviews. The work ran on SQL, Looker, Tableau, Power BI, Kantar, and NetBase.

Results

The model gave leadership the evidence to reduce media spending by 25% while sales continued to grow, by moving budget out of channels whose reported performance overstated their real contribution.

What It Means

A marketing mix model earns its keep when it changes how a leadership team allocates money. At $88 million a year, even a few points of misallocation is real money, and the companies best placed to recover it are the ones willing to measure every channel against the same standard. The same discipline applies at any budget size, which is why a marketing audit often starts with the question this model answered.