Questions you Should be Asking Your DSP
summary
Machine learning DSPs treat ad budget as an investment, not spend, optimizing for ROAS and cash‑flow/payback rather than publisher lists; they claim access to every smartphone user via exchanges but recommend letting models find users instead of hard allow/deny lists. Key operational checks: what inputs (MMP events, SKAdNetwork data, incrementality tests) they need, how many bids per second they can process, and whether cohorts show ROAS over day 30 cohorts so you measure long‑term value and run A/B tests to verify.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:08Rethinking ML-driven DSPs
- 4:27Questions not to ask
- 10:25DSP inputs and outputs
- 13:15Complex measurement ecosystems
- 14:43Long-term value of DSP users
- 15:25Relevant DSP benchmarks
- 16:32Testing DSP advice
- 17:37Q&A
related talks
19mScale for Success: A Developer's Guide on UA Strategy & DSP Partnerships
29mAI and ML for Computational Advertising
23mThe unreasonable effectiveness of data for better monetisation
21mDemystifying Machine Learning for Programmatic UA & Retargeting
46mFrom correlation to causation: measuring ad effectiveness at scale
19mSession: Re-Engagement In 2026: How To Design & Scale Winning Retargeting Campaigns
44mOptimizing Marketing Spend with Auto-Bidding Algorithms
44mOffer Optimization, a Dual Approach