AdTech · Traffic arbitrage · US · 2016 to 2026
Ten years building the ad ops platform behind a traffic arbitrage business
Campaign launch time
Launching a campaign took 30 to 45 minutes by hand and takes under 5 minutes on the platform.
- Before
- 30 to 45 min
- After
- <5 min
Challenge
Media buyers built every campaign by hand inside each ad network and tracked spend in Excel. A launch took 30 to 45 minutes and losing campaigns were spotted hours late.
Outcome
A platform with a campaign wizard, ad network integrations, predictions and an autopilot. Launches take under 5 minutes, one buyer runs ten times more campaigns, and margins rose 10% right after launch.
Key results
- 10x
campaigns per media buyer
- <5 min
to launch a campaign, down from 30 to 45
- 20x
faster reaction to performance changes
- +10%
margin right after launch
- Client
- Spinemedia, US
- Engagement
- 2016 to 2026
- Team
- Scaled with the business from 3 developers to 30 people at peak, including data analysts, data engineers and developers
- Status
- Completed
The situation
In arbitrage the margin is the business, and it is thin. Spinemedia's media buyers worked directly inside each ad network, built every campaign by hand and tracked spend and performance in Excel. Launching one campaign took 30 to 45 minutes. A campaign that had started losing money got noticed when someone spotted it in a spreadsheet, usually hours later.
What we built
-
A campaign wizard for budget, pacing, targeting presets and creative assignment, with checks before anything goes live.
-
Two-way integrations with Facebook Ads, Taboola and Outbrain that keep campaigns, creatives, budgets and performance data in sync.
-
A prediction engine for campaign performance, built years before LLMs were usable, and an autopilot that applies optimisation changes with a visible diff, an audit trail and one-click revert.
-
Near real-time reporting on spend, impressions, CTR, CPM and CPC by campaign, creative and traffic source.
-
Prebid.js set up and customised on the client's websites, plus ongoing support for the sites themselves.
-
LLM content generation inside the campaign wizard, added as soon as the models were good enough.
Results
Launch time dropped from 30 to 45 minutes to under 5. With the autopilot and predictions handling routine adjustments, one buyer could manage ten times as many campaigns, and the team reacted to performance changes about 20 times faster. Margins rose 10% (relative) right after launch and kept growing as we added automation. The team ran over $10M a month in ad spend through the platform.
How the engagement evolved
The first version shipped in under three months in 2016 with three developers. The team grew with the business, up to 30 people at peak, including data analysts, data engineers and developers. When growth levelled off and feature work slowed, we scaled down to 12 people, including an engineering manager, a data analyst and a data engineer.
For ten years we also absorbed every API change the ad networks shipped, so buyers never had to stop work because a provider changed something. The engagement ended when Spinemedia wound down its arbitrage operations.
They were a reliable and proactive partner. We implemented a handful of features that were researched, planned and delivered by Pixelteh. We were happy with them as a dev partner, and with the tech partner they evolved into over 8 years. They are capable and reliable, and that's more than you can ask for, based on my experience.
Stack
Related technologies
Industries