Running 25 SSPs because your ad ops team added them one by one over five years is not a header bidding strategy. It is latency debt disguised as demand. We ran 90 days of controlled A/B tests across 50 publishers to identify which demand partners actually move eCPM, which ones are free-riding on your inventory, and what the right stack actually looks like in 2026.
The header bidding premise is sound: more demand sources competing simultaneously produces higher clearing prices. In practice, many publishers have implemented this principle without discipline, adding SSPs until their page load times are unacceptable and their auction logic is a black box.
The result is a stack where 8 to 10 SSPs generate 85 to 90 percent of actual revenue, and the remaining 15 to 20 partners add timeout risk, latency, and complexity without meaningful incremental yield. The problem is identifying which partners are in which category, because every SSP will show you reporting that makes them look essential.
Every SSP has numbers that make them look good in their own reporting. The only number that matters is incremental eCPM contribution when you run them against a clean control group. That is what almost no one measures.
Over 90 days, we ran controlled experiments across 50 publishers with varying traffic profiles (content category, geography, device mix, and audience scale). For each publisher, we created three inventory pools: the control (existing full stack), a reduced stack with the bottom 10 partners removed, and a reduced stack with the bottom 20 partners removed.
We measured eCPM, fill rate, page latency impact, and timeout rate for each configuration. We then calculated the incremental eCPM contribution of each SSP partner by measuring the delta between "stack including this partner" and "stack excluding this partner," controlling for traffic composition and seasonality.
The results were consistent enough across publishers to draw generalised conclusions, with some variation by traffic type noted below.
Across all 50 publishers, removing the bottom 20 SSP partners (by incremental eCPM contribution) produced an average eCPM increase of 18 percent. This is not because those partners were competing for the same impressions, but because reducing the number of active bidders per auction reduced latency by an average of 340 milliseconds, which meaningfully improved the quality of auctions being held.
Faster auctions clear at higher prices because buyers can allocate budget with better accuracy when the supply signal arrives reliably within their bid evaluation windows. Slow, unpredictable auction timing degrades buyer confidence and suppresses bids below true willingness-to-pay.
The "high-value" SSP partners in our analysis shared consistent characteristics: low bid timeout rates (under 4 percent), stable latency (p99 under 180ms), unique demand sources not duplicated elsewhere in the stack, and transparent auction mechanics. Partners that failed on any two of these dimensions consistently showed low or negative incremental eCPM contribution.
We cannot publish a specific ranked list of SSPs because performance varies significantly by traffic type. What we can share is the framework for evaluating which partners deserve a place in your stack:
Tier 1 partners provide consistently high bid density, unique demand sources, and low latency overhead. For most publishers, this is 5 to 8 partners. These should be prioritised in your auction logic and given the widest access to inventory. Removing any of them should produce a measurable negative eCPM impact in controlled testing.
Some SSPs have strong, unique demand in specific content categories or geographies. A gaming-specialist SSP may be Tier 1 for your gaming content and irrelevant for your news content. Run Tier 2 partners selectively on the inventory segments where they demonstrate genuine incremental value.
Any SSP that fails to demonstrate positive incremental eCPM contribution in a 30-day controlled test should be moved to quarterly evaluation and ultimately removed if the pattern persists. The burden of proof should be on the partner to demonstrate value, not on your team to prove their absence makes things worse.
Partner evaluation is most effectively run as a quarterly process rather than a one-time cleanup. Markets change, DSP budgets shift, and SSP performance evolves. The publishers who maintain the best-performing stacks treat it as ongoing maintenance rather than a project.
The practical process: every quarter, run a 30-day holdout experiment on your lowest-performing 20 percent of partners. If they do not demonstrate positive incremental eCPM in that window, serve them notice. This keeps your stack lean, your auctions fast, and your relationships with core partners healthy.
The discipline required to remove a partner is harder than it sounds. Sales teams, historical relationships, and "just in case" reasoning all push toward accumulation. The data should be the deciding vote, not relationship management instincts.
Monthly analysis on programmatic trends, publisher strategies, and advertiser playbooks. No fluff. Just data and insight from the team building at the frontier of adtech.
No spam. Unsubscribe anytime. Monthly cadence.