The Hidden Dangers Of Supply-Path Optimization

Data-Driven Thinking” is written by members of the media community and contains fresh ideas on the digital revolution in media.

Today’s column is written by Ari Paparo, CEO at Beeswax.

The latest ad tech buzzword is supply-path optimization, and conveniently it comes with a tidy acronym – SPO – to save us all typing time.

For those not familiar, the basic concept is that, due to the proliferation of duplicate auctions inherit in header bidding, it is incumbent on demand-side platforms (DSPs) to optimize the “paths” to buy supply.

While there’s been no shortage of words written about the benefits of SPO and the imperative to implement it, not as much time has been spent delving into the challenges it presents media buyers looking for the cleanest, most transparent route to their audiences.

Why SPO? Duplication.

Duplication is the obvious reason why DSPs are anxious to shed supply. If the same impression is available from two or more sources, why should anyone waste resources evaluating and bidding multiple times?

While this generally makes sense, the observation that duplication exists does not immediately suggest the right mechanism for eliminating this duplication. And when you get into the details of how duplication can be handled, you end up asking some pretty hard questions about the impact various techniques may have on advertiser results.

Suppose I told you that Exchange A and Exchange B were perfectly duplicated, with the same auctions available on both in real time. The naive approach would be to choose one exchange or the other and, voilà, the problem is solved.

But from the buyer’s perspective, the two exchanges are not identical unless the following criteria are met:

  • Neither exchange has variable or buy-side fees
  • Both exchanges are 100% transparent, with no hidden costs
  • Publishers have the same take rates with both
  • They both use exclusively first-price auctions
  • They both have the same cookie-match tables with the buyer

Obviously not all the criteria are possible. Even with perfect duplication of inventory, the well-intended naive SPO strategy will produce suboptimal results by missing some valuable auctions that would otherwise have produced incrementally better performance, and by overpaying for other auctions that could have cleared with an alternative supply path.

The next question: What if the deduplication is not naive, but using artificial intelligence (AI), machine learning (ML), blockchain or some other buzzword?

First, let’s make it clear that it is not technically feasible to deduplicate in real time due to the speed and scale of RTB auctions. You cannot receive an RTB auction request, wait around for duplicates to appear from other exchanges, compare them and throw out the ones you don’t want – latency kills, kids.

Inevitably SPO systems use some kind of static, per-placement, per-region mapping of duplicate paths to supply and then make some kind of intelligent (or not) decision about which to bid on. The problem with this approach is the same as the “naive” example above: Unless the exchanges are 100% congruent, the value of each impression will not be identical, and any algorithm will lose some results.

To use a trivial example, assume our two exchanges are identical, so the algo round-robins on bids between the two. What happens when an impression from Exchange A is supposed to be ignored, but it has a valuable matched cookie, while the whitelisted auction from Exchange B has no user identifier?

Changing Dynamics

To make the challenge of optimizing SPO even more daunting, you have to consider that the exchange landscape is not static and is subject to feedback loops.

Consider the effect of a single large DSP switching off Exchange A entirely in favor of Exchange B. If we’re in an environment with only first-price auctions and no yield management, the first-price results should be the unchanged.

But in the real world, where most exchanges are conducting a mixture of auctions and transparency is not assured, the clearing prices on Exchange A will decline while those on Exchange B will rise. The DSP that is “optimizing” its supply path will put all of its buying into the more expensive channel.

When exchanges are actively yield managing through dynamic floor prices and the like, SPO can cause faster and more aggressive price adjustment as the “winning” exchange can quickly adjust prices (higher) to reflect the buyer’s more active bidding.

Funny Business

The imperative of SPO for DSPs is all about costs. Listening and responding to millions of queries per second is really expensive. And when exchange No. 47 comes knocking on your DSP’s door with “premium supply from top publishers,” the buyer might be a bit skeptical about whether that supply is really worth the bother.

This situation causes an economic misalignment where exchanges are anxious … cough … desperate … cough … to get the DSP’s bids, lest they lose their yield edge with publishers, and, God forbid, get kicked out of the header.

So, we have a situation where desperate exchanges are selling into margin-starved DSPs – what’s the worst that could happen?

This issue was recently detailed by AdExchanger’s Sarah Sluis.

DSPs are allegedly grabbing margin at the expense of second-tier exchanges, and not to the benefit of the agencies and marketers that trade through those platforms. If DSPs “optimize” their supply path for their own benefit, they will be working against the customer’s interest most of the time.

A less insidious scenario may be playing out among exchanges that also own DSPs. Not sure I really need to spell this out for you

What To Do?

First, the good news. As more exchanges move to strictly first-price auctions with reduced fees, the buy side has far less to worry about when choosing between supply sources. Many of the risk factors I’ve described become much smaller relative to the bids. For example, the effect of dynamic floor prices is minimized in a first-price environment. The risk of “second pricing yourself” is also reduced significantly.

The less good news is that buyers need to take hard looks at the optimizations of their DSP partners to understand how exchanges are being throttled, eliminated or deprioritized on their behalf. Many buyers, reasonably, look at sites and domains as their optimization lever, but with SPO, the exchange or inventory source becomes a critical way to cut the data and evaluate partner access.

Finally, buyers should build relationships with exchanges, even if they end up transacting through an intermediary. Buyers and their agencies can learn an enormous amount about the changing landscape through supply meetings (hint: They’ll buy lunch), and if nothing else, these conversations will help drive productive and deeper conversations with DSP partners.

Follow Ari Paparo (@aripap), Beeswax (@BeeswaxIO) and AdExchanger (@adexchanger) on Twitter.

Enjoying this content?

Sign up to be an AdExchanger Member today and get unlimited access to articles like this, plus proprietary data and research, conference discounts, on-demand access to event content, and more!

Join Today!


  1. I don’t know the feasibility, but it sounds like we need an impression ID that would match across all exchanges.

    • Matt Clark

      One challenge with an impression ID is that many of the sell-side participants (including publishers) would be incentivized to fake it. There are material advantages to publishers in inflating their impression counts, and to SSPs in having their supply appear unique.

  2. Thank you Ari for writing on this important topic. I don’t think anyone can disagree that the pioneers of supply path optimization made a lot of assumptions, it wasn’t about the single impression being duplicated. At a macro level, the rewards far outweighed/outweigh the risks. We mitigated spoofing. We reduced the chance of buying Banner masquerading as Video. It also helped birth tools such as Ads.TXT. The biggest hidden danger is having a stagnant method for SPO vs a process that is continually evolving (and evaluating).