Most recently, we’ve participated in a project for the US market, where we were implementing the Ad Exchange module, which automatically resells traffic between SSP and DSP, which do not have direct integration. We’ve created an exchange system that was able to process more than 30,000 QPS (Queries Per Second), filtered more than 70% of bot traffic and had an income of more than 300,000$ per year. To gain this, we dived deeply into the RTB(Real-Time Bidding) and formulated a few key points on how to launch a project to extract revenue from the traffic resale scheme.
So how can you earn hundreds of thousands of US dollars?
Number of Martech solutions
1. You must understand the auction business model
You have to answer these questions:
- What exactly will be the criterion of success?
- Will you be using filtering and targeting?
- Will there be the closed auctions for the “chosen ones”?
- Will you deliver traffic to which this DSP does not have access (due to the lack of support for the OpenRTB protocol by DSP)?
In our project, it turned out that the analysis of historical data of transactions gives us enough for profit. Thanks to the AI (artificial intelligence) algorithm, the system does not send meaningless requests to those DSPs that did not respond to similar requests earlier or gave too low rates. Besides, we provided filtering of bots (about 70% of traffic). This functionality appeared after checking the initial hypothesis that it can increase margin by 50%. The parameters of all filters are available to the client from the control panel, which allows adapting the application to the market.
There is also another way to enrich traffic. You can implement targeting using various parameters from the DMP (data management platform). By the way, independent DMP platforms are actively evolving now. They aggregate more and more data about the audience, which allows you to build more and more complex targeting, depending on the needs of the advertiser. We just completed such a DMP project for the American customer (before that, data segment management was on the side of AppNexus).
However, you have to use DMP carefully. DMP analytics requires a certain amount of time and resources. Since the automated procurement proceeds are very sensitive to response delays, data analysis duration can’t be very long.
2. SSP and DSP must be synchronized with businesses
In our opinion, this is one of the main elements that determine the success of a project. Different systems offer and buy heterogeneous traffic, so you need to choose a combination of SSP and DSP, where the interests of the seller and the buyer overlap, but they are not directly integrated. The final income will depend on how much this niche is demanded.
A smart approach is to integrate with multiple SSPs and DSPs to increase impressions. For example, at the start, our client integrated two DSPs and four SSPs, and later the number of partners was increased to 20. We’ve implemented mechanisms to balance the number of requests to partners per unit time to avoid overloading.
3. Technical skills of the team are crucial
Integrations are not only about the initial arrangement with the partner. You have to avoid technical problems or you can lose your integration with a partner. Most of the systems are sensitive to response time. Moreover, the income directly depends on the flow of incoming requests.
For example, in the early stages of our client’s project, the exchange received about 245 million requests per day (about 2,800 operations per second(QPS)). Successful was about 0.3% of requests, and revenue from 1,000 impressions is about 10 – 50 cents. With such a volume of requests, the revenue of the service at the MVP stage was about 50 to 300 US dollars per day.
After integrating with new partners, the number of operations per second increased to 30,000(estimated peak level 100,000 op/sec). With this frequency, the customer reached very comfortable profit levels.
More requests – more money, but these requests must not only be received but also processed. If the implemented service delays the process (respond too slowly or do not respond at all due to overload), the partners will simply refuse to integrate with it and choose a competitor. Therefore, from a technical point of view, the exchange should be optimized as much as possible. This requirement imposes stringent conditions on the technical skills of the team.
Here you can find more detail on the technical side of our solution. We’ve wrote an article about it.
4. Optimize your distribution of effort
It makes no sense to implement the entire complex traffic enrichment system at the start. The better way is to select some basic elements from the existing auction model and incorporate them into a working prototype to test the initial hypothesis.
Unfortunately, not always RTB expert has enough technical expertise to make such selection on his own. Thus, the quality of the technical team plays a decisive role.
In our last project, we presented a working prototype in three months, and MVP in six months. At the same time, the system supported only the basic functionality. It had to answer the question of whether it is possible to earn money by optimizing requests to the DSP through analysis of previous transactions.
After receiving confirmation that the auction model works, we’ve implemented modules that provided increased profits – bot traffic filter, daily reporting for SSP and DSP (number of requests, impressions, cost), metrics about the service’s operation(server load, delayed responses from and to partner systems, etc).
5. Think about the future
The market is changing, new players are constantly appearing, more sophisticated bots, etc. If you miss the situation and, for example, allow a larger percentage of bot traffic, which will be filtered out already at the partner level, you will have a reduction in revenue, and perhaps a break in partnerships. Therefore, if you plan your solution to exist for a long time(and this is necessary for payback), you must think about the future.
Each request is the overhead of the Ad Exchange, and the only way to get a substantial profit for a given income from one transaction is to increase the number of transactions per conventional thousand requests. How exactly to do this depends on the competence of all involved parties. In our project, while we were finalizing filters for traffic and a reporting system, our customer was simultaneously looking for new integrations and negotiated. In the end, our customer received an income of about 500 thousand US dollars during the first year.
Conclusion
We believe that there is room for a wide variety of ideas in this market, and we can offer our technical expertise. We are ready to consider different interaction models, for example, revenue sharing, where the client becomes a partner and an agreement is reached on obligations and % on revenue sharing. In this case, we could launch a technological solution in 3–6 weeks, depending on the number and complexity of integrations at the start. Thus, in collaboration with us, you can quickly and relatively cheaply check out the new Ad Exchange model for specific SSPs and DSPs before even launching a full-fledged project.