How to Evaluate a Prebid Server Provider: The Questions Most Publishers Don't Ask

You've already made the hard decision. Server-side header bidding is the direction, and now you're evaluating providers. The problem is that most vendor comparison frameworks were built around the wrong questions.

Visit any "Prebid Server provider comparison" roundup and you'll find the same criteria: supported SSPs, dashboard design, contract length, maybe something about managed setup. Those are table-stakes questions. They don't get at the structural issues that actually determine whether a Prebid Server deployment performs well two years from now — or becomes a slow-motion problem.

This post gives engineering and ad ops leaders a sharper framework: the questions that reveal infrastructure quality, incentive alignment, real latency commitments, and what "managed" actually means in practice.

Why Most Prebid Server Evaluations Miss What Matters

The typical enterprise software evaluation is feature-driven: list requirements, map vendor capabilities, score the matrix. That approach works reasonably well for CRMs and project management tools. It fails for Prebid Server, because the value difference between providers isn't in features — it's in architecture decisions, business model, and operational depth. None of those show up in a feature grid.

Two vendors can both offer "multi-region infrastructure," "managed configuration," and "100+ bid adapters." One runs those regions on shared cloud tenancy and hasn't updated adapter configurations since Q3 last year. The other runs isolated, single-tenant infrastructure with dedicated engineering monitoring adapter health weekly. The feature list looks identical. The performance gap is not.

The evaluation questions below surface these distinctions. Work through them with any provider before shortlisting.

Infrastructure: Who Owns It, Where It Runs, What Happens When It Fails

Infrastructure is where the most important — and least-discussed — differences between Prebid Server providers live.

Ask: Is infrastructure shared or dedicated?

Shared-tenant infrastructure means your auction traffic runs on the same servers as dozens of other publishers. During traffic spikes — the kind that come with breaking news, sports events, or product launches — you're competing with those publishers for the same compute resources. Latency spikes at exactly the moments when impression value is highest.

Dedicated or isolated infrastructure costs more to operate, but it means your performance is determined by your traffic patterns, not someone else's. Aditude runs every client on isolated AWS infrastructure, with full configuration isolation and no shared-tenant risk.

Ask: Where are the server regions, and how is geographic routing determined?

The speed advantage of server-side header bidding disappears if the Prebid Server endpoint is 200ms away from your users. Any serious provider should have multi-region deployments with intelligent routing. Ask specifically: which regions, how is routing determined, and what's the measured latency from your primary audience geographies — not from the provider's home region.

Ask: What does failover look like, and how long did your last outage last?

Every infrastructure setup has a failure mode. The question isn't whether outages happen — it's how quickly the system recovers and whether publishers are protected during that window. Ask for a recent post-mortem. If they don't have one, that's either because they've never written them up (a process problem) or because they're not willing to share (a transparency problem). Either is worth noting.

Ask: What does the uptime SLA say, and does it include penalties?

SLA language varies significantly. "99.9% uptime" sounds strong until you notice the measurement window is monthly, excludes "scheduled maintenance," and carries no financial consequence for breach. Push for the specific terms: what's the measurement window, what's excluded, and what remedies exist if the SLA is missed. Aditude's Prebid Server carries a 99.99% uptime SLA — but more importantly, ask any vendor what happens contractually when they miss theirs.

Conflict of Interest: How Does This Vendor Make Money?

This is the question that makes vendor representatives uncomfortable, which is precisely why you should ask it first.

Prebid Server is publisher-side infrastructure. Its entire purpose is to run a fair, competitive auction that maximizes publisher yield by forcing demand partners to compete on price. The moment the entity operating your Prebid Server also has a financial relationship with one or more of the SSPs bidding through it, you have a structural conflict. That vendor has two masters — and in ad tech, the SSP relationship is often the larger revenue line.

Research into publisher ad tech stacks has documented cases where SSPs overcounted reported revenue by material margins — in one documented instance, an SSP was reporting wins it wasn't paying out on. Publishers running those stacks only discovered the problem through independent impression-level counting. If your Prebid Server provider has commercial ties to the SSPs bidding through it, the chance of that oversight surfacing is much lower.

Ask: Does this vendor operate an SSP, ad exchange, or demand-side business?

If yes, that's not automatically disqualifying — but it requires you to understand which demand partners are in the stack, whether any receive preferential treatment (earlier timeout windows, flooring exemptions, tiebreaker logic), and what disclosure you receive. Aditude's Prebid Server is built exclusively for publishers — not retrofitted from SSP infrastructure — and every design decision is oriented toward publisher yield rather than demand-side revenue.

Ask: What is the pricing model, and what does the vendor optimize when you're not watching?

Revenue-share sounds aligned, but the incentive structure matters. A vendor taking a cut of your revenue should theoretically want to maximize it — the practical question is aligned with what? Maximizing gross CPM this month, or sustainable yield over time that requires hard decisions like culling low-quality demand? Understand the incentive structure before you sign.

Ask: Can you see all bid data, and does the vendor take any margin on demand?

Some managed Prebid Server providers take a hidden spread on impressions flowing through the stack. This is legal but often undisclosed. Ask explicitly: does any revenue from my inventory flow to this vendor beyond the agreed fee or rev-share? Request contractual language on this point, not just a verbal assurance.

Latency: What Are the Actual Commitments and How Are They Measured?

Latency is the metric every Prebid Server vendor will claim to win. It's also the metric most frequently stated in ways that obscure meaningful comparison.

The right question isn't "what's your average latency?" — it's "what's your p95 latency, measured at the client request level, in my primary audience region?"

Average latency can look excellent while p95 — the 95th percentile, what 1 in 20 auctions actually experiences — is catastrophic. For publishers running 100M+ impressions per month, p95 is the number that determines your tail-end user experience and the bid adapter participation rate that comes with it.

Ask vendors to provide:

• P50, P95, and P99 latency figures (not just average)

• The measurement methodology (server-side measurement is not the same as end-to-end client measurement)

• Region-specific figures for your primary geographies

• How latency is affected during traffic spikes

A properly configured Prebid Server can deliver up to 40% latency reduction relative to a pure client-side setup — but that figure assumes well-optimized infrastructure. Poorly maintained or underpowered server-side deployments can erase that advantage and introduce new latency that didn't exist with Prebid.js. Aditude's Prebid Server delivers auction results in under 100ms globally, with multi-region infrastructure to support that commitment.

Also ask about timeout configuration. Timeouts set too conservatively cut demand. Timeouts set too aggressively let low-quality demand drag latency. A competent provider manages these dynamically — not with a single static value applied to all partners. For a detailed breakdown of how timeout calibration affects fill rate, see Prebid Server Architecture for Enterprise Publishers.

Configurability: What Can You Control, and What's a Black Box?

"Managed" is a word that can mean many things — ranging from a genuinely full-service operation to a vendor that controls everything and gives you a dashboard you can look at but not touch.

Ask: What configuration decisions require going through the vendor, and what can your team change directly?

If timeout values, floor logic, SSP priority, and identity configurations all require a ticket and a two-week turnaround, that's not a managed service — that's a dependency. The difference matters most when you need to respond to a revenue problem: a publisher team that can isolate an underperforming adapter and adjust its configuration today has a structural advantage over one waiting on a vendor support ticket.

Identity and first-party data configuration is one of the areas where generic Prebid Server deployments consistently fall short. Cookie sync rates drop when requests originate from remote servers rather than the browser, and correcting for this requires active management, not a one-time setup.

Ask: How are adapter updates and maintenance handled?

In 2024 alone, Prebid Server processed more than 960 pull requests to the open-source codebase, averaging 18 per week. Not all adapters are maintained at the same quality level, and misbehaving adapters — ones that take the full timeout and return no bid — drag overall auction performance. Ask specifically: how often are adapter configurations reviewed, how are degraded adapters identified, and what's the escalation process for flagging one to your team?

Ask: What does reporting cover, and is there impression-level granularity?

Summary dashboards showing daily revenue by SSP are table stakes. The useful questions are: can you see bid-level data? Can you reconcile your impression count against each SSP's reported count? Can you isolate performance by device type, geography, and ad unit without a custom data pull? The answers determine whether you're managing your inventory or being managed by it.

Migration: What Does Onboarding Actually Look Like?

Migration risk is consistently underweighted in Prebid Server evaluations. Publishers focus on steady-state performance and underestimate the complexity — and potential revenue impact — of the transition period.

Ask: What is the typical onboarding timeline, and what are the revenue impact scenarios during transition?

A provider who says "two weeks, no problem" without asking about your current stack complexity either has a very standardized onboarding process that may not fit your setup, or is underselling the risk. A credible answer includes a discovery phase to understand your current configuration, a parallel-run period where both old and new setups operate simultaneously, and a cutover strategy that minimizes the gap between baseline and new performance.

Ask: Who owns the work during onboarding — your engineering team or theirs?

Some managed Prebid Server providers hand you documentation and leave the integration work to your team. Others take ownership of the configuration, line item setup, and SSP mapping. The answer matters for resourcing and timeline. Aditude's managed approach means hosting, demand configuration, and ongoing maintenance are handled by Aditude's team — reducing engineering overhead on the publisher side.

Ask: What happens to your existing Prebid.js setup?

Server-side doesn't mean replacing client-side. Most sophisticated publishers run a hybrid architecture — some demand partners client-side for match rate reasons, others server-side for latency. A provider pushing full server-side migration without discussing hybrid architecture either hasn't thought through the identity implications or is optimizing for operational simplicity rather than your yield. Aditude's Cloud Wrapper and Prebid Server are designed to operate together in a hybrid configuration, matched to each publisher's inventory mix and yield priorities.

Ask: Is there a data migration path for your historical reporting?

Switching providers creates a reporting discontinuity. Some vendors help bridge this; others don't. If your ad ops team relies on historical comparisons for optimization decisions, understand what the gap looks like before you commit.

The Prebid Server Provider Evaluation Scorecard

Use this scorecard when running a formal evaluation across multiple providers. Score each criterion on a 1–3 scale: 1 = unclear or unsatisfactory answer; 2 = verbal commitment; 3 = specific, contractual, verifiable commitment.

A provider who scores 3s across infrastructure, conflict of interest, and latency — and gives clear, contractually grounded answers — is worth serious consideration. A provider with strong feature coverage but weak answers in those three areas is a risk you're accepting knowingly.

Bringing Aditude Into Your Evaluation

Aditude's Prebid Server is purpose-built for publishers: isolated AWS infrastructure, 99.99% uptime SLA, sub-100ms multi-region latency, and a business model with no SSP-side conflicts. If you're running a formal evaluation, bring us in — we're comfortable being compared on the questions that actually matter.

Related reading:

• Prebid Server Architecture for Enterprise Publishers: What Works at Scale

• What Is Publisher Infrastructure? A Definition for Ad Tech Teams

• The 5 Questions Every Publisher Should Ask Their Monetization Partner

• Prebid 2024 Year in Review — Prebid.org