F.A.Q

Modified on Thu, 8 Jun, 2023 at 9:43 AM

How is Bidease preparing for Google's potential elimination of the MAID on Android devices?

Bidease is using its own resources to tag and label data accordingly while building its own datasets to ensure the company is not dependent on 3rd party providers to access the inventory in order to correctly identify Device IDs.

How does the sell side determine the winning bid? Is it the same across all exchanges?

Yes, Bidease is using the auction of first price where the highest bid is the winning one to determine the winning bid across all Ad Exchanges.

If fraud occurs, who is responsible for making the appropriate adjustments to ensure it doesn’t happen again?

Bidease is responsible for all potential fraudulent activity - the system will adjust its algorithms immediately in case of potential fraud detection to ensure our data is not corrupted.

Does Bidease deal with ad verification vendors, ie. DoubleVerify, Integral Ad Science, or Zefr?

We do not. We have explored them in the past and may do so again in the future.

How are we better than RevX or MAAS? What about Smadex or Adelphic?

Each company uses its own proprietary algorithms and models - we believe our models perform at least on par with the leading global DSPs and are able to demonstrate comparable performance.

What type of optimization does your platform have: CPI, CPA, ROAS? What type of events can you optimize towards?

We have all types of optimization using CPI, CPA, and ROAS (IAA/IAP), such as, but not limited to - Subscriptions, First Purchase, Retention, Install.

We can optimize all types of events that we are receiving via MMP/S2S.

What does the learning period of the algorithm look like? How much time does it take to finish initial learning? What amount of events does the algorithm need to finish initial learnings?

At Bidease, we believe the progression contains the following stages: learn, explore, then optimize.

In the Learn Phase, we don’t spend a single dollar. We build models that are trained off of unattributed postbacks to put the best foot forward out-the-gate
In the Explore phase, we launch a fleet of models that spend and collect conversions
In the Optimize phase, we identify the conditions with the best performance and invest more spend into those pockets

Total process may take 1 to 2 months, depending on the first-party data amount. The learning and optimisation curve starts from the very first impression, action and/or event, while additional key events equal better and more precise models of target users.

How does the algorithm work with the budget? Does it always spend all the budget or it corrects itself based on the performance? If you don’t spend all the budget all the time, who interferes: AdOps team or the algorithm itself? Can you work with an open budget and achieve target ROAS or target cost per conversion?

During the learning phase, we are eager to collect data aggressively based on recommended and defined budget caps if not agreed differently, which can be a no-cap performance-driven approach based on set KPIs (CPA/ROAS) to gather more data to improve our machine learning model and build improvements, but not lower than the minimum recommended amounts. Additionally, during the learning phase, we use the budget allocation between the most efficient campaigns and a dynamic optimisation of creatives.

After the learning phase, we can work strictly based on KPIs (CPA/ROAS) with a scalable approach without budget caps (open budget).

Additionally, our teams can also manually limit/raise budgets based on defined criteria, eg agreed limits or performance.

Which geos have the most potential on our platform? Can you give us a top-10 based on potential?

We do have global coverage with more than 7.2+ billion unique monthly user profiles at our fingertips with constant MoM growth. Where Americas 275M+, Europe 155M+, Asia 545M+, Africa 70M+, Oceania 3M+ Daily User profiles reachable.

TOP10 based on our internal expertise are:

What exchanges are you integrated with? Are there unique exchanges? What is the potential of these unique exchanges? In which geos do they have traffic?

We are integrated with all main Ad Exchanges with the best quality traffic:

DoubleClick Ad Exchange by Google
Facebook Audience Network
AppLovin
Unity
Fyber
Chartboost
IronSource
Vungle
Bigo
Anzu
Yandex
Mail
And others

Additionally, we do constantly expand and add Direct Deals and PMP Deals with various App Developers and Publishers, Networks and Data Providers which results in cheaper and qualified traffic with 1st party data available for targeting and optimisation purposes. All our traffic sources are always fully transparent for our customers.

What creative formats work on your platform? What amount of creatives per campaign can we use? How does the algorithm make decisions on the creative level?

Video 1080x1982 / 1920x1080 (up to 10 Mb, 15 - 30 sec) with end cards

Native 1200x627 jpeg/gif (up to 500 kb)

Static interstitial 1080x1982 / 1920x1080

Native banners 320x50 / 300x250 (jpeg/gif)

HTML5, Playable, Rewarded

The minimum requirement is 2 per campaign.

We are using dynamic creative optimisation and budget allocation based on performance from quantitative and qualitative data collected. Diverse formats and styles of creatives will lead to a better understanding of performing placements and designs for our algorithms to use as main ad sets and traffic sources having rest in the least priority.

We also have our in-house production team to support you with creative ad sets.

What kind of information/data from our side can help you during the initial learning period: LAL lists, unattributed purchase+revenue postbacks, lists of top publishers in top geos from other DSPs?

All of the above: LAL lists, unattributed purchase+revenue postbacks, and lists of top publishers in top GEOs from other DSPs if that improves efficiency of our forecasting models

Campaigns will run on LAL models built on non-attributed data during the learning phase, and algorithms will be trained also based on all newly collected data within our platform. During that time predictive models are created based on accumulated and qualified data.

Additionally, we can optimize by top-down approach from the conversion funnel if it is significant and if there is a correlation of the events in the funnel which lead to primary KPI (CPA/ROAS).

However, there are two very important data sets we need to collect beforehand:

Suppression Lists to prevent us from wasting spend on users who have already been previously acquired
Unattributed Postbacks to seed our training data and get better performance out-the-gate

Which postbacks do you need for your algorithm to work properly? How do you use them?

More data and postbacks we can receive from you will result in better and faster model building to foster performance. Mandatory Postbacks are used to get information about conversions and/or KPI events for calculation and optimisation purposes.

Here you can find documentation on Tracker postback links and additional resources.

Which scaling and optimization principles work best on your platform?

Probability theory and machine learning (Wiki) in connection with Multi-Layered Bid optimization algorithms which work on layers like Placement, Campaign setup, Ad Set, Ad Type and other data.

What is your approach to iOS? Do you use postbacks from MMPs or do you work with SKAN models? If you use SKAN, please describe how your iOS algorithm works.

We are using postbacks from MMPs. We are not using SKAN as of now. Our tight integration with MMPs does work on the probabilistic model. If it is possible to have iOS with probabilistic attribution enabled and advanced privacy turned off, our algorithms can build better models for NONIFA traffic. Otherwise, the models can be built for IFA traffic only

How do you prevent fraud?

We have our proprietary algorithms that spot bot/unusual traffic & are eliminated at the source (pre-bid), which the client is not charged for;
We partner with Tier1 Ad Exchanges/SSPs which have built-in anti-fraud systems in place;
We are integrated with 3rd Party Anti-Fraud tools such as MOAT, and Double Verify;
We are integrated with top MMPs which have their fraud solution which we use for data analysis and insights;
We have feedback loops with SSPs to analyze anomalies and flag potential issues for all ecosystems;
And lastly, we are a fully transparent mobile DSP platform providing full access to our platform for customers where all traffic sources are visible with data and with no User IDs being hashed for your further usage.

What makes you different from other top DSPs on the market?

Customizability, transparency and efficiency.

While each DSP has its algorithms and models where the main difference may lay in the accessible reach, the number of traffic sources and partners, their quality, 1st party data availability, how data is collected etc. we believe that our proprietary prediction and machine learning algorithms provide the most advanced audience, gender, and age predictions with the ability to target by interest and behavior.

We have invested and continue to for the last 6 years with proven success among the diverse types of customers across the globe supported by our employees from New York, Seoul, Dubai, Beijing and European offices.

We also provide full information about traffic sources and data points we collected so our Clients are fully aware of quality of our buying team’s work, inventories we touch and how the money is spent.

How can you send us impressions, clicks and spend data: API, S3?

API would be the most suitable solution. Overall, our IT dept. and engineers are flexible and able to provide customized and tailored solutions for your needs.

What are the targeting features you use?

Full list of campaign level settings and targeting options with full description can be found here.

How do you pull Ad Revenue data? Can you optimize using it? Do you have a direct deal with IronSource?

We are able to pull Ad Revenue data via MMPs/S2S - but MMPs are more preferable than the others. Also, we have a direct deal with IronSource.