Havas, an agency holding company heavily invested in the programmatic ad space, last week launched what it is calling a “meta DSP,” or demand-side platform. Real-Time Daily wanted to learn more about the “meta DSP” and spoke with Lawrence Taylor, global head of Affiperf, the new programmatic division within Havas in which the “meta DSP” is housed.
Real-Time Daily: How does the "Meta DSP" work?
Lawrence Taylor: The rationale behind the Meta-DSP is really simple: no single DSP uniformly dominates the competition. We schematically describe a DSP as a combination of 4 macro components: inventory, data, features, algorithms.
The role of the meta-DSP is to propose, for each campaign, the best combination of DSPs. To achieve this, we need several things.
First and foremost, we need to deeply understand and know DSPs, from the tiniest feature they've just added to the macro architecture of their APIs. A theoretical knowledge is clearly not sufficient, we need first-hand knowledge. [We have] more than 30,000 campaigns spread across several DSPs -- [that is our] first source [to] understand [each] DSP’s respective strengths and weaknesses. To finish, we've ran sessions of experimental head-to-head campaigns in order to measure discrepancies on all important domains: bidding algorithms, inventories access, median CPM for the same inventory, etc.
Second, we need a way to streamline the process of campaigns creation and monitoring. A meta-DSP campaign is a computer program that executes on DSPs. That means we make heavy use of DSP’s APIs.
RTD: Does the advertiser use Affiperf to run a campaign across multiple DSPs at the same time, and the advertiser just simply doesn't
care which DSP is ultimately being used?
Taylor: Generally a campaign will be composed of sub-campaigns spread across several DSPs. But this is not [always] the case -- there are cases where the use of several DSPs might not be justified. This is the case where the friction costs associated to a multi-DSP campaign is greater than the advantages that could be gained from it.
RTD: How do you prevent the DSPs from competing against one another to drive price up if they are ultimately bidding for the same advertiser?
Taylor: This is a point we take very seriously. First, this problem might not be as big as it seems. We ran trials where we did not find any evidence of self-competition. DSP strategies are diverse, and collisions seems to be a rare event.
That said, we cannot afford taking such risk. So, in a nutshell, we take the following measures in multiple DSPs campaigns:
First, we make sure that, at campaign level, for any given ad exchange or SSP, a single DSP is allowed to bid. This is a decision taken at campaign setup. In order to still maximize the benefits of multiple DSPs campaigns, we allow for a fine grain allocation decision, e.g. allowing a certain couple (creative size, ad exchange) for DSP 1, and another one for DSP 2.
Second, for strategies with really high bids (e.g. retargeting), where the risk of self-competition is really high, we take an even more conservative stance, by allowing those strategies to run on a single DSP. This is a decision we might relax as the time being, on a ad exchange/SSP by ad exchange/SSP basis.
Third, we run rules that seeks for self-competition execution prices patterns. Those reports, ran at high time frequency, could act as an emergency stop-loss. Hopefully we haven’t encountered such alerts so far.
RTD: What's in it for Havas/Affiperf? Where does your cut come from? How do you remain neutral?
Taylor: Affiperf's aim is to create the greater possible value for each of our advertisers. Having one open infrastructure means we are more independent from technology changes, we can adapt faster to new solutions.
RTD: What DSP partners are currently signed up?
Taylor: Currently MediaMath [and] AppNexus. And new partners are added all the time, all we require is an open API.