9. Who are our users?
We are a two sided marketplace
One side is influencers like
• Celebrities
• People active in their community
The other side is their buyer audiences
22. It's about the
absolute volume
you can capture
Bid
Volume
Sometimes good enough
A budget of the big company
23. 2. We didn't have the right data
We wanted to estimate for each audience
• The optimal long-term bidding strategy
• Its true addressable size, and overlaps
• Product affinities
1.
2.
24. 3. We were just slow
Many products/audiences › many campaigns › a lot of
mistakes
Optimizing existing campaigns
Exploring new opportunities
Traditional rules not sufficient
Traditional templates not sufficient
1.
2.
3.
25. Sample setup
Creatives
• set for newsfeed
• set for sidebar
Campaign setup (all dark
posting)
• Conversions for 300k+
audience
• Clicks + CPC
• Engagement – with direct CTA
• Retargeting + CPM
1.
2.
3.
26. Sample setup #2
Audience tests & research
1. Owned pages data
2. Interest targeting
3. Past website data
4. Custom audiences
5. Lookalikes
6. Behavior on site
Setup pixels
Setup tracking urls
Budget / Optimization
• Initial $10 per ad set (rigorously
tested)
• Quick scaling up
• Frequency cut-off
• End of campaign/relaunch
1.
2.
3.
39. TL; DR
We did custom automation because it was
• difficult to get the ground truth to existing tools
• difficult to optimize bidding and spend with existing tools
• still time consuming even with some automation
We built our own stack on top of Facebook API to automate
parts we didn't find elsewhere – it was tough but it worked