Blogs

Gaurav Jain's picture
July 23, 2014
Gaurav Jain

A digital identity is data that uniquely describes a person. This unique identity may not necessarily be traceable to an actual person, yet it helps in providing a very rich experience. Publishers can use this information to show personalized content on their websites and DSPs may use this to show highly personalized creative.

A digital identity is the lifeblood of the advertising ecosystem and it is this identity that helps differentiate an online advertising medium from its mass channel counterparts. Identity is used in all stages of advertising – namely collection, identification, personalization and attribution.

For years, cookies were the main mechanism of user identification on the digital medium. Though it was not the most efficient mechanism, it was highly functional. Cookies are unique to a domain and are also unique to the browser being used. The Advertising ecosystem built its own layers on top of cookies so as to leverage the data in a much more efficient way.

When Apple launched its app store, advertisers started using UDID to track installs. UDID had a problem - It is a unique identifier based on hardware and had no user opt out. In 2011, with the launch of iOS 5, access to UDID was denied by Apple to reduce the privacy concerns. In absence of any identifier, mobile technology companies started building their own identifiers such as OPENUDID and ODIN. In 2012, Apple launched a resettable ID named IDFA which allowed advertisers to use a privacy compliant ID.

Similarly, advertisers have been relying on Android-ID for tracking devices in the Android ecosystem since Android Froyo release. This is a non-resettable ID, which is specific to hardware. Just recently, Google too came up with a new Advertising ID which is resettable and privacy compliant.

Android ID and IDFA, at times, are passed raw and other times hashed by using different algorithms such as md5 and sha1.  

identity_blog_01 

Device ID mapping is required not only for targeting and showing relevant content but also to maintain the frequency capping. Since correct attribution is the backbone of digital advertising, fragmentation of device IDs is severely hampering the ability of attribution. Without a reliable identifier, advertising platforms have no way to understand the user’s interaction with the ads.

There are three places where DSPs interact with the user – Data collection, Targeting and Attribution. For a non-retargeting platform, user identity mapping is not important in the first two stages.

Data collection: Interaction happens between advertiser and DSP

Targeting: Interaction happens between SSP and DSP

Attribution: Interaction happens between advertiser and DSP  

The advertising industry is moving towards converging of passing raw device IDs in IDFA format for iOS and Google Advertising format for Android. In the meantime, advertisers may choose to use a single identity which binds these multiple identities. There are broadly two ways – probabilistic and deterministic.

Some advertisers have chosen to use probabilistic approach, commonly known as fingerprinting, in which on the basis of different properties such as browser version, device type, country, time zone, language settings, user agent, browser resolution, browser add-ons etc. This might help reach huge scale but it may not be very accurate. Different players claim to achieve accuracy levels from 60-80%.

Other advertisers are using a deterministic approach in which a common identity such as an email ID, phone number or social network IDs are used. This can be used to tie different identities to a single user.

The next level of user identification lies in binding these multiple identities across devices to a single common identity which can then be leveraged for cross device marketing strategies. Given the nascent stage of the industry and the high fragmentation, this is a very challenging problem to solve. At Komli, we are excited about these challenges and the opportunity it provides for marketers.

Sanjiv Jha's picture
June 19, 2014
Sanjiv Jha

Using a combination of predictive and causal modeling techniques to deliver higher ROI

First, let’s get on the same page for the definition of predictive modeling and causal modeling. Causal modeling is used to understand what events or actions influence others. It is an estimation approach based on the assumption that the future value of a variable is a mathematical function of the values of other variables.  On the other hand, predictive modeling is the process by which a model that best predicts the probability of an outcome is created. However, when it comes to predicting human behavior such as clicks and conversions, predictive model has its limitations. At Komli Media, we use the combination of both causal and predictive modeling practices to overcome the limitations and best optimize a campaign. We call it “Predictive Causal Modeling”.

We use predictive modeling to target advertisements to consumers and we observe that targeted consumers purchase at a higher rate subsequent to having been targeted. However, we cannot say with certainty whether the advertisements influenced the consumers to purchase or predictive models did a good job at identifying those consumers who would have purchased anyway. So it’s reasonable to assume that both are true and accordingly optimize campaigns.  We use site map analysis of advertiser's web sites to identify audience segments. We then determine their purchase intent intensity using various user signals, through randomized controlled experiments. Based on these observational data we typically create two kinds of campaign strategies. One strategy that targets high intent segments which largely consists of consumers who would have purchased anyway. While the second strategy targets relatively low intent consumers to whom we then show advertisements multiple times, converting them to high intent consumers over time.

new_image_blog

When it comes to user behavior, historical data can be unreliable to predict the future. Any model based on historical data implicitly assumes that there are certain steady state conditions or constants in the system. This cannot always be true when involving people. For example, purchase intents change drastically when a flash sale is announced. What makes predicting accuracy difficult is that with more data there is more probability of noise. To overcome this problem, we use an in-house developed casual modeling technique of min/max bid price analysis. This technique removes tipping points from the predicted value of conversion that pushed the bid price over the edge leading to wastage of impressions. After filtering outliers or tipping points we significantly reduce the wastage of impression. This helps us improve margins across campaigns and deliver significant value to advertisers as we can now target a wider audience for the same budgets.

By combining the best practices of predictive modeling driven by prediction of outcome and casual modeling driven by observational data, we have significantly improved our capability to accurately predict the purchase intent of a consumer. It has helped us improve our margin and also reach out to a wider consumer set for the same budget. It has also helped to target relatively lower intent consumer and convert them to purchasers. After applying the casual modeling technique to predictive modeling we have observed that the number of new consumer’s purchase has doubled, providing evidence of the success of Predictive Causal Modeling techniques. 

Atul Shrivastava's picture
June 03, 2014
Atul Shrivastava

Leveraging data to drive higher ROI

Website browsing data helps you understand what your user was looking for along with the strength of his intent. In the previous post of this series (Data is the difference!), we saw different user data signals that can be extracted from your website or app. This post will focus on using those signals for crafting a personalized message and segmenting valuable customers for ROI optimization.

venn-diagram2

Communication strategy using Contextual signal

The contextual browsing data signals like category, product visited, etc. can be used to devise a relevant communication message for the user. A customized and relevant message goes a long way in driving a positive response towards the ad. Few examples are:

  • Somebody who's searching for mandarin collar shirts or was found browsing multiple shirts of that type can be shown an ad showcasing other mandarin collar shirts in your collection matching the price range or color that the user was looking for.
  • Somebody who put a certain pair of leather shoes in the shopping cart (but abandoned the transaction) can be shown an ad with picture of those exact shoes saying "Your favorite shoes are waiting for you! Use this Coupon code XYZ123 to avail a special 10% discount and make them yours!"

There are two ways you can achieve this:

  • Manual method: In your remarketing platform, you can create segments for each category and target these segments to show the top 5-6 static creatives in their respective categories. Needless to say, this method is not scalable if the number of categories or subcategories in your product catalog is huge.
  • Automated method: This method is far more effective in delivering a personalized communication and is scalable to the longest of product catalogs. If your remarketing platform supports Dynamic Creatives, the task of Ad generation is taken over by its recommendation engine and is executed for each user in the runtime. Products similar or equal to the ones browsed or cart-ed by the user are filtered out from the product catalog (remember the inferred signals!) and are used to generate an ad specifically for that user.

ROI optimization using Intent signals

The intent signals help you tell an accidental visitor or casual onlooker from somebody who's seriously considering a purchase. These signals can be leveraged to spend selectively on the users (in terms of advertising cost or individual discount offers) and hence boost your ROI. Since most of the remarketing platforms use RTB to buy inventory from Ad exchanges, this translates into placing high bids on a strong intent user to win every opportunity of serving an impression to him and not bidding or bidding conservatively on someone who is not likely to convert. Few examples of selective bidding are:

  • Somebody who was seen on the website regularly (3-4 sessions) for the last 1 week, checking out multiple Items in the Bed category with significant time spent and page views per session can be considered to be deep in the research phase with high category coherence, contemplating a purchase soon and hence, should be bid at aggressively.
  • Somebody who put a certain product in shopping cart (but abandoned the purchase), or compared it with other items in the same category or added it to his wishlist is considered to be deep in the purchase funnel and should be bid at aggressively or even offered a personalized discount to give the final push.

Just like in case of contextual signals, there are two ways to achieve this:

  • Manual method: In your remarketing platform, you can create segments based on recent views, frequency, funnel stage etc. and set a different bidding strategy for each of them.
  • Automated method: If your remarketing platform supports Rule-based bidding, you can assign different bid-boost values to each intent signal like cart, recent views, category coherence etc, so that a user scoring high on multiple intent signals is computed a high bid value as well. Some sophisticated platforms leverage machine learning to compute dynamic bids for each user based on the intent signals. These predictive models decide the weight for each intent signal (in computing the bid value) by looking at the conversions and intent signals registered then.

Limitations of Browsing data based retargeting

This post cannot be complete without mentioning the caveats in browser data based retargeting. Since the entire user browsing pattern is tracked against a cookie here, this method suffers from limitations around cookie persistence and matching.

  • Few browsers like Safari don't allow third party cookie persistence, while some users surf in the incognito mode where cookies can't be set. Systems with high safety settings flush the cookies after every session, making the retargeting effective only within the session.
  • Cookies work in the silos of a browser i.e if a customer uses multiple browsers or devices to access your property, he will be treated as multiple customers and the unified view of his browsing pattern will be lost. This also means a cookie can be retargeted only on the same browser/device. However, the industry is trying to overcome this limitation with cross-device user identification which is essentially tying up multiple cookies/devices using keys like email id.
  • If multiple users are sharing the same machine, their browsing pattern gets tracked against the single cookie. The remarketing platform then thinks of them as a single user which compromises the efficacy of remarketing.

Conclusion

Contextual data signals can be leveraged to deliver a personalized and relevant communication for the user based on his past browsing behaviour. While intent signals can be used to optimize the ROI by spending selectively on your retargeted base. Creating segments is the easiest way to use these signals, but if your website has a huge product/category catalog or you want to use multiple intent efficiently without exploding the segment count, you should go for Remarketing platforms with dynamic creatives support and automated bidding controls. Cookie-based tracking used to capture these signals has some limitations which the industry is trying to overcome by using fingerprinting techniques or tying cookies with actual user account.

Amar Agrawal's picture
May 21, 2014
Amar Agrawal

Adaptive Ads

A banner ad typically is a rectangular advertisement placed on websites either on top, bottom or sides of the website’s regular content. Here are the 3 most important parts of a banner ad-

1. Call to action

2. Color, images and background

3. Value proposition

Online advertisers use A/B testing to test out how different variations of each of these can better the performance of the ad. The traditional way to do this would be to upload all possible ad variations into an ad-serving platform and let these run randomly on test group of users. The advertiser then pulls out a report by ad-variation to compare the relative performance of each and consequently continues to promote the best performing ads to run across all users.

Though A/B testing is the only true way to objectively test the optimizations that can be made to ad creatives, this is obviously tedious and time consuming. This is where ad technology comes to the rescue!

Algorithm powered A/B testing

Ad-serving platforms today allow advertisers to auto-generate many ad variations in bulk and upload them to the ad-serving system at the click of a few buttons. The system also optimizes the ad serving to the best performing variations automatically so that the advertiser need not do manual performance monitoring and optimizations. 

Adaptive Ads

Komli’s proprietary re-marketing platform takes this one step further, by allowing advertisers to create truly personalized ad copies for each individual user a.k.a Adaptive Ads!

Adaptive ads helps the advertiser:

1. Create more engaging ads that help continue the conversation with the user based on past interactions on the advertiser’s website

2. Keep creatives fresh by showcasing the latest catalogue of products

3. Achieve operational efficiency by uploading a single ad template that is customized to each user with the right products and messaging

DCOGraphic

Performance Insights

Adaptive ads leads to better user engagement and higher purchase propensity. A leading retail advertiser in India used Komli’s Adaptive Ads to re-market users who dropped out from the advertiser’s website. The advertiser saw a 4x increase in CTR and 2x higher conversation ratio as compared to standard banners.

The Komli Advantage

Adaptive Ads is seamlessly integrated into Komli’s remarketing platform so that advertisers have complete transparency and control on how these ads run. This includes product feed setup, user data collection & audience creation, upload of dynamic templates – all integrated into a simple-to-use efficient self-serve interface.

Product_Screens_New

And that’s not all! Adaptive ads is powered by a state-of-the-art predictive modeling technology that can optimize bids based on user intent signals like purchase funnel, time of last visit, time spent on the website, etc. to enable a truly turn-key approach to remarketing.

Learn how Adaptive Ads is enabling advertisers to deliver personalized promotional offers to users to incentivize them further to complete abandoned purchases in our next blog post.

administrator's picture
April 28, 2014
Atul Shrivastava

The importance of various data signals for Remarketing

User Remarketing is one of the effective techniques used by marketers today to get their site-visitors to purchase on their online store. While SEM, social and branding campaigns drive new traffic to your website, remarketing encourages the site visitor to go further down the purchase funnel by showing a highly relevant message, while being selective in going after the users so as to optimize your ROI. Remarketing depends on user data (website, CRM, social, emails etc) to be able to do so and the success of any remarketing campaign largely depends on how good the data collection is. The browsing pattern of the user on your property (website or app) can be a great asset to you when it comes to remarketing. The footprints users leave behind on your property can help you understand what he was looking for and how strong was his intent. For an existing customer the browsing data can be combined with purchase history or profile data to deliver even more effective messaging.This blog touches upon the key data signals marketers should focus on extracting and the mechanics of extraction.

Types of User data signals

The schematic representation below represents different signals you can capture under browsing data on an ecommerce site.


 Atul_Blog_Venn_Diagram

While this list covers most of the important data signals one can use, it by no means represents an exhaustive set. Also, other verticals like travel sites, finance, classifieds etc, may differ in terms of contextual signals but the signals depicting the intent strength remain pretty much the same across verticals.

Mechanics of Signal extraction

Since most of the online businesses (like ecommerce, travel, classifieds etc) allow user to access the website/app content without logging in, majority of the browsing data is tracked against the cookie. Remarketing platforms use pixels or tags to track the browsing pattern just like any web analytics solution does. The remarketing pixel should be placed on every page of the website, either directly or through a tag manager solution like Google Tag Manager.

Extracting Direct Signals

With every page load, the pixel fires, drops a cookie  (if the user doesn't have one) and tracks details like timestamp, page url, referrer url (source of traffic), device type etc against that cookie on the server-side. Other signals like category, product, and funnel can be extracted using one of the following ways:

·       Parsing the URL, if it is structured.

·       Pixel crawling through the page and reading the values from certain DOM elements.

·       Website owner passing the values explicitly to the pixel as parameters (params).

However, not all the signals can be captured at the time of pageload like  user going through review section, photos within the product page or clicking a button (like add to wishlist) which doesn’t reload the page. In such cases, pixel is configured to listen to such user interaction signals and trigger a server call when it happens. Alternatively the website owner can fire the pixel explicitly with appropriate parameters when such an event happens.

Extracting Inferred Signals

Signals like recency, frequency, time spent on page, category coherence are computed on the server-side from individual events recorded by pixel. Product attributes like price, brand, offer etc can be inferred by looking up product_id (captured as a direct signal) in the catalog (usually provided programmatically as an xml feed). In absence of an xml feed, these signals can be captured as direct signals from the page itself by passing key values as params in the tag.

The above mechanics are applicable when the pixel is placed on your desktop or mobile-optimized site. In case of a mobile app where cookies don't exist, the user is identified by device-id which needs to be passed to the pixel. The pixel is deployed using a tag manager SDK already integrated with your app.

Marketers who have user history or profile data can also bring these high value signals for leveraging in their remarketing campaigns. Such information can be passed via params and subsequent audience buckets can be created for remarketing. For eg: a fashion ecommerce site can pass whether the user is a male or female based on the profile data of its logged in user. The richer the data extraction about the user, the better will be the audience segmentation which will enable a high ROI for your campaigns.

Conclusion

Website browsing data is instrumental for an effective remarketing strategy. Different signals (like product, funnel, frequency, recency etc.) can be extracted by placing a remarketing pixel on your website and are used for crafting a personalized message and segmenting valuable customers for ROI optimization.

Komli Remarketing DSP platform provides the best-in-class data collection and segmentation functionality in the industry today. We have built advanced remarketing predictive model that leverage direct and inferred signals to predict the bid price and customize the marketing message. If you would like to know more, please visit our website http://www.komli.com/in/content/komli-remarketing or contact our sales team at remarketing@komli.com for a live demo.