Roman Garbar
Guest post by Galia Lahav, Digital Marketing Manager @Persona.ly
User acquisition is one of the most important pillars of running a successful mobile app. User acquisition strategies stretch from organic (ASO) and owned media, to paid media (social platforms, networks, and programmatic).
As a mobile marketer, you most certainly tried various marketing channels and platforms. While paid social seems to be the easiest nut to crack, it’s never a good idea to put all the eggs in one basket as it causes dependency on one traffic source. While there is nothing wrong with working with networks (the good ones, of course), there are more efficient ways to purchase traffic. Programmatic channels allow marketers to diversify sources by buying traffic from various ad exchanges in a smart and transparent manner.
What is programmatic?
Programmatic in adtech refers to the use of technology to automate the buying and selling of advertising. This includes using a programmatic platform—Demand Side Platform (DSP)—to purchase ad space in real-time, targeting specific audience segments, and analyzing data to optimize ad campaigns.
Programmatic advertising allows for more efficient and effective advertising, as well as greater targeting capabilities for advertisers. Humans simply can’t match this level of efficacy due to the complexity of multiple factors’ combinations and the speed with which changes can occur.
RTB Ecosystem and Process
Advertisers can purchase programmatic traffic through real-time bidding (RTB), an open auction-based marketplace using a demand-side platform (DSP).
In the programmatic advertising ecosystem, there are several key players: publishers, advertisers, SSPs, and DSPs.
- Publishers are the owners of digital properties such as websites or apps who offer ad space to monetize their product.
- SSPs (Supply-side platforms) act as a bridge between publishers and buyers, providing a platform for publishers to sell their ad inventory.
- DSPs (Demand-side platforms) are platforms used by advertisers to purchase ad inventory in real-time by connecting with multiple SSPs and other ad exchanges.
- Advertisers are the ones looking for available ad space to serve their campaign and reach their target audience.
What makes programmatic stand out?
SSPs originated from ad networks—companies connecting multiple publishers (websites and apps) to access their advertising inventory and monetize by displaying ads to their audience.
Ad networks provide access to a large inventory; however, they require a lot of manual work and, in some cases, lack transparency. Ad networks are mostly based on performance-oriented payment such as CPI (cost per install) or CPA (cost per action). (We will cover what’s wrong with this in the next paragraph). SSPs provide programmatic (read “automated”) inventory and work exclusively with the CPM (cost per mille, or cost per 1000 impressions) model and allow greater versatility in targeting practices.
While the CPA-based user acquisition seems like a great deal (advertisers pay per install/target event), there are some cases where the results are too good to be true. The lack of transparency leaves room for manipulation, as ad networks don’t disclose the traffic sources to avoid direct competition with a publisher. As a result, advertisers might end up paying for fraudulent installs or events. There are multiple anti-fraud solutions providing a safety net for advertisers, but the damage done by fraud is beyond just monetary. Potential damages include harming ASO and cannibalizing organic traffic.
With programmatic buying, advertisers purchase the impressions from publishers through trusted SSPs. Further, the DSPs ensure the right targeting based on the event probability predictions developed by machine learning algorithms. While the KPIs are set in a performance-based manner (and might involve eCPI, eCPA, or ROAS), the advertisers pay for the number of times the ads were displayed to users.
The “secret sauce” of programmatic buying is to find the perfect combination of audience characteristics and to sift through irrelevant audience segments. This approach ensures that ads are displayed at the right time to the audience that would genuinely be interested in the product and be more likely to perform the target actions. Therefore, there is no reason for programmatic DSPs to hide their traffic sources and publishers.
On the one hand, this approach allows advertisers to lower costs. On the other hand, it eliminates the risk of any fraudulent activity, as multiple anti-fraudulent methods are involved in both SSPs and DSPs.
Types of DSPs
While programmatic is just a smarter way to purchase traffic, DSPs enable advertisers to set up certain KPIs to achieve success. The different DSP types include rule-based DSPs and ML-based DSPs.
Rule-based DSPs
For rule-based DSPs, the bidding is done based on simple rules (mostly set up by an expert) that the platform is following. While this is a viable way to purchase traffic, it’s lacking efficiency and requires a lot of manual adjustment.
Most self-serve DSPs are rule-based, as managing an ML-based algorithm requires a vast knowledge of the platform and algorithm (and a lot of data science).
Here is an example of a setup for a rule-based DSP:
| Store Category | Placement | Device Type | Bid |
| RPG | Banner | Tablet | $0.24 |
| Casual | Video | Smartphone | $3.50 |
| Puzzle | Native | Smartphone | $1.10 |
Machine learning-based DSPs
For machine learning-based DSPs, the targeting is based on the conclusions that the algorithm picked up from studying big-data. The platform considers multiple factors, processing millions of combinations to predict the probability of a certain event by a certain audience segment.
Once an SSP signals that there is a new auction available, a DSP only has around 200 ms to process whether to bid and how much to bid based on the expected outcome.
The click, install, and event predictions are based on previous audience behavior and happen almost simultaneously, straight after the SSP sends data on the available inventory. Then the algorithm decides whether to bid—and if yes, at what price.
Bid shading is another distinctive characteristic of ML-based DSPs. Based on the predicted user value and past auction data (think millions of ad auctions per second), the algorithm defines the bid price and places the bid. This allows the advertiser a more balanced bid price and ensures that the advertiser is not overpaying in the first-price auctions realm.
Here is a very simplified structure of the bidding process for ML-based DSPs:
The key differences between the rule-based and ML-based DSPs are listed below:
| Rule-based | ML-based | |
| Real-time bidding | √ | √ |
| High-quality inventory | √ | √ |
| Scalability | √ | |
| Basic targeting definition | √ | √(in the early stage of learning) |
| Automatic targeting adjustment based on performance predictions | √ | |
| Self-service | √ | |
| Managed service | √ | √ |
Data sharing in programmatic – the cold start problem
Data sharing in mobile user acquisition involves providing a UA partner with access to an audience list.
An ML-based DSP, having access to a large dataset, can then generate lookalike audience segments to solve the cold start problem (and significantly reduce the exploration cost). This approach allows the advertiser to ensure that they’re targeting the right audience from the start. Simultaneously, the audience list serves as a suppression list to prevent targeting any existing users.
Another feature of ML-based DSPs is that they are able to recognize the behavioral patterns of different audiences, classify the segments, and even serve different creatives based on the segmentation. This way, campaigns focus on acquiring new and relevant users, rather than wasting advertiser’s budget on targeting existing users or those who will not convert.
Getting started with the programmatic channel
1) Do your research.
Remember that programmatic is about technology and quality inventory. Some players might position themselves as a DSP while being an advertising network. In order to avoid confusion, make sure to check the case studies, research the company, and ask the right questions about the traffic sources and methodologies.
2) Share your goals and KPIs.
For a machine learning algorithm to perform proper targeting, it requires a very specific campaign setup where the goals are clearly defined.
3) Share data.
As mentioned before, data sharing is a critical step in mobile advertising. There are multiple safe and data privacy-compliant ways to share the data (and it’s not a .csv file).
4) Be patient but careful.
Depending on the app vertical, it might take a different time and budget for ML to establish a robust install-deep funnel event compression. Make sure to be in touch with the platform’s customer success manager for a progress evaluation.
About the Guest Author
Galia Lahav, a Digital Marketing Manager at Persona.ly, a mobile-first DSP operating worldwide. With access to over 2.5 million ad auctions per second, proprietary bidder and machine-learning algorithms, Persona.ly offers transparent, performance-driven, highly targeted UA and retargeting solutions at scale. The company is trusted by Rapido, Games24x7, Papaya Gaming, Ubisoft, Tilting Point, and many others.
