Adgebra’s Audience Segmentation Engine

audience segmentation
Tweet about this on TwitterShare on LinkedInShare on Google+Share on FacebookEmail this to someone

In this blog, we will try to answer few questions related to how adgebra’s audience segmentation engine works.

  1. How does ‘adgebra’ audience segmentation work?  
  • Adgebra code placed on various partner (publisher) sites identify each user uniquely with identifier dropped in user’s machine. Based on users ‘content consumption pattern’ on partner sites, he/she is categorized in numerous segments. There are 20 higher level segments (Travel, Entertainment, finance etc.) and 120 niche level segments (Eg: – under finance, there are niche segments like loan, insurance, etc.). In simple words, segmentation is based on user behavioural pattern.

2. Does ‘adgebra’ offer demographic segments?

  • As of now, adgebra does not offer any demographic segments like age, gender, income group, etc. Adgebra believes in sharing what is near real rather that depending on probability algorithms. Adgebra offers behavioural audience segments. There are few instances where publishers do record demographic data of a user which can be passed to adgebra. Though, such data comes at additional cost and handled on a case to case basis.

3. Does it mean that if a user reads only 1 article about football will make him football fan?

  • No, this is where ‘adgebra – audience segments’ are different than any other DMPs (data management platform). Adgebra has very a unique way of assigning ‘confidence score’ to each user segment. Confidence score varies from 1 to 5 star based on how often user consumes the content is one particular segment. For example, if ‘person A’ is reading 3 finance related articles in a day versus ‘person B’ reading 3 finance related articles  in a week will share different confidence scores. There are many other parameters that affect confidence score.

4. Now, over the period of time, each user will become 5-star rating?

  • The confidence score can go down to zero if the user does not consume content of that particular segment for a certain period of time. For example, if ‘person C’ reads a lot about football during world cup might get 5 rating, but immediately after world cup he stops reading about football, his confidence score will start coming down and eventually become zero after certain days of no activity. This feature also ensures that accidental and seasonal readers are flushed out of the segment regularly.

5. Will advertiser get access to Confidence score feature while targeting?

  • Yes, confidence score feature can be used to target niche and relevant audience while creating a campaign in adgebra. Though, one should remember that higher the confidence score lower are the volumes (reach).

6. Does adgebra provide custom audience?

  • Yes, adgebra has provision to bucket user in custom segments. This is possible only when partner publisher agrees to pass additional information about that user. Eg: Age group, Salary bracket etc. OR there is a need to add new segment very specific to advertiser requirement. This definitely comes at a cost and takes more time to scale.

7. Can I use, custom audience feature to run remarketing campaigns?

  • Yes, this is in-fact one of the best feature of adgebra. Same confidence score logic can be used to identify users’ level of interest in advertiser based on his/her content consumption on advertiser site. And remarketing campaigns can be run using this newly created custom segment.

8. Is there a provision to expose Adgebra’s audience in 3rd party DSPs like DBM / Appnexus?

  • Technically Yes, though we prefer not to expose our audience segments to outside world. This can be taken care from case-to-case basis.

9. Can single user fall into multiple segments, if Yes, how?

  • Yes, same user might have different area of interests and can fall into multiple categories based on what he/she is reading on a regular basis. For Eg: ‘person X’ might be reading a lot about lifestyle as well as Bollywood gossip. So he/she might fall under both ‘Lifestyle’ and ‘Entertainment’ segments. Remember that, ‘confidence score’ might vary for each segment based on recency and repetitiveness of user reading content under respective segment.

10. What is the validity of the segment?

  • As mentioned earlier, ‘confidence score’ of any segment going down to zero indicates that the user does not belong to that segment anymore. This is nothing but expiry of each segment. Each segment has different expiry period based on the level of importance.

11. Where is all the user segmentation information stored, if cookies, then what happens if a user deletes the cookie?

  • Yes, you are correct, we do store segment and confidence score in the browser cookie. If a user opts to delete the cookie, our system will have to restart the segmentation from scratch. We are working on a new technology called ‘fingerprinting’ to identify user machine again and reload the segment information from the stage where it got deleted.

12. If advertiser decides to target Segment A with confidence score 3, what does it mean?

  • If users of Segment A with confidence score 3 are targeted, your ad will be served to all the users under Segment A with confidence score 3 and above (4 and 5). Ads will not get served to users falling under Segment A with confidence score 1 & 2.

13. What are stats of adgebra audience segments?

  • As of now, adgebra has segmented 150mn unique machines across the globe. There is a churn of 10%-15% of the audience, segments do expire and new get added. Our monthly reach is around 50mn unique visitors.
849 Total Views 1 Views Today
Tweet about this on TwitterShare on LinkedInShare on Google+Share on FacebookEmail this to someone
Rohit Bagad

Rohit Bagad

Rohit is the Founder & CEO of Inuxu. He strongly believes that the optimal use of data, be it audience behavioral data, quantitative data or offline data, can open infinite opportunities for every entity in the digital media ecosystem.

All stories by: Rohit Bagad

Leave a Reply

Your email address will not be published.