With continued economic uncertainty and ever-tightening ad budgets, the gap between supply and demand continues to grow. This has put added pressure on the programmatic ecosystem, leaving the sell-side with a surge in available inventory that needs to be accounted for and further squeezing the buy-side to make the most out of their budgets.
As an industry operating in real-time, at billions of transactions every second, efficiency and performance are key. We only have to look at The Trade Desk’s call for bid deduplication—and the reasons behind it—to see that ad operations plays an important role in improving both the workings of programmatic advertising and as a result, the buyer/seller relationship.
The questions become: How can we improve ad operations for a better ecosystem? And what tools can we introduce to support infrastructure in a way that builds a bridge between supply and demand, making sure that both sides are happy with the outcome?
That’s where traffic shaping and machine learning come in.
Using machine learning for better traffic shaping can mean huge strides forward for healthier ad operations, the future of the programmatic ecosystem and two-sided bid optimization.
But, traffic shaping itself needs to be well understood and approached in a meaningful way to achieve its full potential.
We sat down with Benjamin Hansz, Vice President Strategy for Simplaex’s Rivr, to discuss traffic shaping.
Gavin Dunaway: What is traffic shaping?
Benjamin Hansz: In the world of programmatic advertising, traffic shaping refers to the practice of selecting a subset of requests to process based on various signals either within the auction data itself or by enriching it with data from outside the auction process.
When traffic shaping is applied to help an SSP determine which DSPs should receive a specific bid request, we call that Demand Selection. This helps deal with a problem that plagues all sides of the programmatic ecosystem: noise. Referring to the requests sent but not answered, or wasted, and contributing to bottlenecks and compromised margins for the supply partners sending the bid requests out and the demand partners receiving them.
When executed correctly, bid requests are sent only to the relevant demand partners i.e. those who are most likely to participate in the auction in a meaningful fashion, as opposed to sending all requests to all partners. This has become increasingly important with the unpredictability of ad spend and the growth of header bidding that causes an overload of requests—fewer of which actually lead to impressions and revenue, and many of which are duplicate.
GD: How does it work?
BH: There are different ways to execute traffic shaping depending on the set up of the supply side party executing the auctions. Commonly, you’ll see the supply side either let QPS limitations determine which ad-requests and which bid-requests are processed, or select which demand partners to cut in bulk based on rules. Some might develop their own logic based on a couple of dimensions, such as country and device, to guide filtering (better, but time and effort consuming), or even better, implement tools, like Rivr, that plug into the bid stream to analyze the completed auctions and make recommendations based on specified features.
For example, based only on the information from the bid stream (i.e. completed auction events), Rivr is able to establish which DSPs are most likely to bid and which can safely be removed without compromising revenue. Rivr does this all using audience level analysis, which acts as a translator, empowering both sides to understand the inventory and the audience that they’re trying to match.
This is done by using higher value metrics, i.e. instead of looking at the ad format, device type and some contextual clues, audience analysis delivers each demand campaign based on similar audience types displayed by the desired behavior to previous ads (i.e. clicks, views, etc). This allows for media-agnostic optimization and a better understanding of the traffic in question.
The resulting reduction in bid requests sent—an industry range of 30%-60%—has a direct impact on infrastructure costs, performance and efficiency, all while retaining practically all revenue.
GD: Why use a traffic shaping tool?
BH: “Waste” or unanswered bid requests currently account for an obscene portion of most bid streams. Over 90%. This “waste” places strain on SSP performance and has monetary implications—including server resource costs—that hurt a company’s bottom line. Traffic shaping tools cut this waste drastically and have a positive impact on the entire equation.
Traffic shaping allows the supply side to expedite auctions, and simply put, make more from less. A few of the high-level benefits include:
- Infrastructure cost savings as a result of fewer requests to process.
- Higher quality traffic: improved relationship with DSPs due to sending only relevant traffic to each relevant partner.
- Increase in revenue per 1M bid requests which increases efficiency for both sides. We’ve been seeing 24% uplift.
- Improved DSP-SSP relationship: the demand partners gets more relevant traffic as well as cost savings for a healthier relationship.
- Internal efficiency: savings in time & energy for the SSP employees for them to focus on USPs. The benefit is moving it off a roadmap and into reality.
GD: What does traffic shaping have to do with Supply Path Optimization (SPO)?
BH: Supply Path Optimization, put in very simple terms, is a way for the demand side to take back some control and streamline operations. For example, many demand partners have made moves to decrease costs by cutting out underperforming traffic sources, reducing the number of intermediaries in an auction, making calls for deduplication and other improvements for efficiency.
Traffic shaping helps to create a cleaner path to placement and improve the outcome for both auctions by doing the work of cutting out the ‘dead weight’ in the programmatic supply chain.
GD: What is the impact of traffic shaping on the supply-demand relationship?
BH: When done right, the impact of traffic shaping ultimately strengthens the supply-demand relationship by making it more efficient. One way to interpret the end goal of traffic shaping is that it optimizes the deal flow between the buy-side and sell-side. When all the noise is stripped away, the supply side is able to present the demand side with the opportunities it’s genuinely interested in.
Now, whether a specific demand partner actually bids, and is successful is another matter. Put another way, by stripping away all the dross, traffic shaping allows supply and demand to get straight to business.
GD: Is there an impact on revenue with traffic shaping?
BH: Depends how you look at it, and how you play it. It’s possible to do traffic shaping (i.e. cutting out waste) without sacrificing any revenue. However, in practice, it’s not unreasonable that there may be a small revenue sacrificed.
Also, there does come a point when it may be desirable to sacrifice some revenue to significantly increase the reduction in data and server costs, e.g. for sources of high volume, low-value traffic. It all comes back to how the trade-off between lost revenue and cost savings looks in each unique setup.
This ‘sweet spot’ where the savings and efficiencies far outweigh the revenue impacted, will be different for everyone and is determined with our partners on a case by case basis.
GD: What has been the pushback from other industry partners about SSPs employing traffic shaping?
BH: It’s a matter of trust. Some DSPs may prefer to see “all the data” for fear of being left out of pockets of traffic they think are worth it. Or of the greater fear that the DSP’s own algorithms/selection criteria may be gamed by the SSPs. Some publishers may also want to get their ad requests in front of every possible demand partner, and hearing that their SSP is employing traffic shaping may lead them to think they’re missing out.
At the end of the day, ultimate KPI for both scenarios is the result—i.e. when the DSP gets better quality traffic or the publisher gets better revenue.
GD: What can an SSP expect from an out-of-the-box traffic shaping tool?
BH: You should expect three things:
Time to market: an average SSP could easily spend a year or more developing the most basic traffic shaping tool. By contrast, an off-the-shelf solution should be up and running in a month or two.
Adaptability: as each SSP has its own idiosyncratic infrastructure, by necessity, a solution provided by a third party would be extremely flexible in how it works within the existing process and auction logic of that SSP. Thus it basically becomes native to the auction set up as the models learn
Best in class solution: a third-party vendor would have the advantage of sharing best practices across different industry partners. While most companies hire talent from industry peers, the insights and knowledge those hires bring from their peers begins to age from their first day.
A third-party that is actively engaged with a number of industry peers, not only is able to share the best learnings across a wider range, but also execute practices that are the most up to date.