Vizury

Blog

Effective Data Management for Advertisers

Effective Data Management for Advertisers

In the first blog of this series, we recognized the motivation for sharing data effectively and also introduced a framework that Advertisers could apply to manage the process. In a nutshell, the framework requires advertisers to understand their data, the various sources and the impact of each data set or field from a marketing/advertising perspective. Let’s explore the data sources/ingestion aspect of the framework in more detail in this post.

Data sources

For obvious reasons, most digital marketing channels are largely interested in two blocks of first party data: User interaction data (what products/categories are of interest to a user?) and the product catalog (what is the advertiser selling?). Both blocks can be used in isolation but are best used in conjunction to enable meaningful one-to-one conversations (customized messaging) with end users.

1. User interaction data: Sourced almost exclusively from the website, this block of data is extremely crucial for targeted marketing channels like retargeting. Data is ingested from the website using some form of affiliate tags (parameterized, DOM scraping, hybrid etc.). These tags are also used for user tracking (cookies). Off late, this block of data is also expected to include CRM data for users who are registered on the website (anonymized/aggregated data) to build richer user profiles that can be used for more effective marketing. This means that existing linkages between back-end systems like CRM and CMS need to be more fluid or there should be support for off-line exchange of data normalized on User IDs (cookie, CRM ID).

While cleaner ingestion methods are not expected any time soon for this data block, Advertisers should address the following concerns:

  • Ensure data sufficiency and accuracy by cleanly bridging back-end systems and providing a flexible interface for sharing data through tags on the website
  • Force affiliates to engineer their tags for minimal impact on user experience (advisable to occasionally monitor latency & design of these tags)

2. Product Catalog data: The scope of catalog data is also being constantly redefined. Advertisers have every incentive to include information on special offers, channel specific images etc. along with the regular payload (list of products). Sharing this block of data is a more straightforward instance of systems-integration (between Inventory Mgmt. Systems & affiliates’ data storage systems). Rather, it would have been so if there were homogeneity in the technology/product stack used by different advertisers. To mitigate long, drawn-out manual server-to-server integrations, historically, most ad-tech companies chose to scale faster by using a relatively ‘sub-optimal’ integration method – ingesting XML/text files at some given frequency. This was great in the short run, but issues like discrepancy between website information and marketing messages (difference in prices, products being out of stock etc.) are being noticed in such instances. These kind of issues are best avoided by Advertisers interested in safe-guarding their brands and the experience they offer their customers (and not just on the website!). The problems are compounded for marketplace websites where different vendors could be following different update cycles for their inventory. So, if you are an advertiser you should:

  • Invest in technology to disseminate catalog data in almost real-time: the good news is that most open-source/commercial components support programmatic or rule based export & publishing of data. Utilizing these features is the next best thing to having a clean web-service (API – interface) based exchange of data.
  • Build latency expectations for ingestion with affiliates and have clear SLAs defined for freshness of data.

The issues or recommendations highlighted in this post are not meant to be exhaustive and would have been part of the natural outcome of a data dictionary building exercise.

Clearly, both affiliates and advertisers have to work their way towards exchanging data more effectively to get the best results for everyone. While getting close to the ideal scenario may be difficult without standards in place, there are 3rd party solution providers who are trying to bridge this gap. In the next post, we will take a look at some of these solutions and how they stack up in different scenarios. As always, we would love to hear your take on this. So, please drop a comment or write to us at marketing@vizury.com.

How to Get The Best Baner Ads on Your Website – I

How to Get The Best Baner Ads on Your Website – I

vizury interactive banners

 

Advertisement of a product has stepped beyond presenting product information to becoming an engine for user persuasion. Ads have been used for years together to craft the need for the product, often positioning the product in ways beyond the core functionality to connect with the user at an emotional level. While broadcasting ads and print ads have primarily been focused at crafting product perception web banners have essentially been a follow up to these ads, very often carrying specific product related information to the user who already has a fair degree of brand recall.

With the advent of e commerce, web banners were not only a marketing artifact but also lead generators for a transaction to happen on the site. Banners now had to work at multiple fronts. They had to persuade a user who may or may not be interested in the product, has developed a behavioral blindness owing to the tendency to segregate important from the unimportant, and may not be in a position to consume a lot of information.

This behavioral blindness only grew more acute as end users started consuming more content. Their attention spans reduced at exponential rates and browsing between diverse types of content became more organic way of browsing the web.

No one ever opens a web page looking for an ad banner. A Banner now has to do a lot more to gain that small quantum of attention that the user has which is a tall order. Crisp and limited communication, persuasion through imagery, colors and typography and creation of a sense of urgency in the user became critical for a banner to perform, simultaneously also competing with other banners who were doing the same set of things. With the degree of ambiguity built into this process multiple attempts were being made to arrive at a stable, scalable methodology to arrive at high performance banners which will consistently deliver high ROI, something that generic banners were hardly delivering.

vizury banners

 

The concept of retargeted ad banners proposed to plug the problem at precise points of origin. They were in a sense not unsolicited. Any user had to do a certain kind of self-initiated behavior on the advertiser’s site to see these banners. Since they show products / services that users have shown certain interest in, the likelihood of user ignoring these was extremely small. Advertisers now had the knowledge and precision to extend values and benefits to only those who would avail of them, in the sense not wasting money on a non-buyer. So as a concept, retargeted banners would definitely add the value in the space left by generic web banners.

Vizury offers all the advantages of a retargeted banner and tops it up with high degree customization. We constantly experiment and optimize elements like banner fill, animation states and transitions, call to action copy and placement, number of products displayed etc., to be able to churn out high performing banners and replace them with other similar banners before any banner fatigue sets in.

In the posts that follow we’ll discuss the gaps existing in the banner presentation and how we go about identifying and bridging them.

Keep watching this space for more banner facts.

 

OLAP infrastructure at Vizury

Ad-Tech companies like Vizury are constantly querying multi-dimensional data based on client’s request or internal reporting purposes. For example, one needs a system which is able to answer queries such as:

  • What was Vizury’s revenue for the month of April 2014?
  • What was the breakdown of impressions served by banner sizes during the first week of April 2014?
  • What were top 10 campaigns by clicks during the period of April 7 to April 10?

The system which enables queries like these is called OLAP: Online Analytical Processing. OLAP stores the most granular data in a database and sums up the rows as needed. This leads to reading large number of rows for queries.

Olap at Vizury:

OLAP at Vizury

    1. The underlying database: We use Vertica, which is a columnar database useful for persistent storage of the most granular data.

 

    1. Middle layer: We use Mondrian server to store multi-dimensional data structure. This middle layer serves as a “smart cache”. For example, if the first query looks at data for April 1st, then the Mondrian layer caches it. For a subsequent query for dates April 1 and 2, the Mondrian layer retrieves data for April 1st from its cache and queries the database for the data for April 2nd.

 

  1. Front end: We use Saiku as the front-end server. It provides a rich user interface to perform multidimensional queries.

At Vizury, we use open source components wherever possible. Both Mondrian and Saiku are open source and the advantage with open source solutions is that you can modify the source code of various components and customize them to suit your requirements.

Tips for implementing OLAP infrastructure

Here are some key lessons learnt while deploying and tuning our OLAP infrastructure.

    1. Use of columnar database is a must: Initially, we started out by using MySQL, but were soon faced with scaling issues. So we switched to Vertica and our scaling woes were gone. We realized that OLAP workloads work best with columnar databases. This is because of two reasons:(a)In row oriented databases like MySQL, entire row needs to be read even though you may need only a few columns. On the other hand, in columnar databases, storage is column by
      column, and hence only required columns need to be read. Now, in a typical OLAP query a few out of many columns need to be queried and so when using columnar databases, I/O is reduced.(b) Since the storage is columnar, each column can be efficiently encoded and compressed, further decreasing I/O.

 

    1. Columnar databases are CPU bound: Having used MySQL earlier, we were accustomed to the idea that “database workloads are disk bound”. However, in scaling up Vertica, we realised that it is bound by the CPU capacity. For optimum Vertica performance, the CPU utilization should be mostly below 70%. If it is not so, shifting to higher number of cores or more machines helps boost Vertica performance. Once the CPU is utilization is below 70%, we were able to derive small amount of performance boost by shifting to a disk with higher IOPS.

 

    1. Some columns are more compressible than others: If there are fewer values over a large number of rows, then the compression can be very efficient. For instance, Vizury has over 500 campaigns (each campaign is a string), while the number of rows are more than a billion. The campaign column can be compressed to a very small size by using a dictionary to map campaigns to integers.

 

    1. Aggregated tables help: We figured that while we store the most granular data in the database, most (>50%) queries never access some of the dimensions. Hence, we aggregated the tables across those dimensions to come up with aggregated tables. We saw up to 30% increase in average query times by using aggregated tables.

 

    1. Vertica design tool is not of much help: We tried changing projections as per Vertica design tool, but were not able to derive noticeable performance gain due to this.

 

  1. Mondrian needs to be customized: We were able to make several changes to the Mondrian server to maximize its performance. In particular, we found that Mondrian makes needless “read tuple” queries apart from the required “segment load” queries. By avoiding a large fraction of these “read tuple” queries we observed a huge increase in the performance.

We were able to bring down the average response times of OLAP infrastructure continuously, while the number of rows in the database continued to increase, as depicted by the following graph:

2

 

A big shoutout to Arvind Jaganathan who worked on all the above the improvements, and brought them from theory to practice! We continue to work on several improvements to further bring down the response time.

Vizury Central: Your gateway to in-depth Campaign Data

Vizury Central: Your gateway to in-depth Campaign Data

On the eve of our fifth anniversary, we released Vizury Central – A unified interface for our clients and media partners to manage their campaigns and relationship with Vizury.

We have made considerable progress in adding new features on the lines of superior analytics, real time campaign management and personalization.

Superior Analytics:

Our drill down dashboard now provides single-click filtering and drilldown into multiple dimensions of data.

vizury central_superior analytics

Questions like:

  • What was the Ad spent on Facebook Exchange for the month of April?
  • What is the trend of CTR for non-Facebook inventory for our apparel products?
  • Which days in the week are more profitable than the others? Etc can be easily understood through our drilldown dashboard. A key aspect of this screen is that, there is no “Submit” button anywhere!

Central also provides an insightful view into a client’s website conversion as well. Our Conversions Report provides Website Conversions, View Through conversions and Vizury attributed conversions all in the same graph. Data export on all order ids is just a single click away!

vizury central_analyze conversions

Real Time Campaign Management:

Clients can now manage their campaigns in real time by setting category wise CPC prices. This feature allows the user to set differential pricing for clicks across 20 categories. This will be a powerful tool to our clients to manage spends across different categories.

vizury central_category based pricing

Language Localization:

Keeping in view the geographical spread of our clients and media partners, Vizury central is now available in 6 languages: English, Chinese (Traditional), Chinese (Simplified), Japanese, Spanish and Portuguese.

We plan to further enhance the user experience of Vizury Central by introducing more indepth analytics around our client’s data and campaign management options. Stay tuned!

Please visit https://central.vizury.com to login to Central.

Cross Channel Retargeting: Mobile web to mobile app

You must have read our Article on In-App Retargeting that describes how Retargeting through your mobile apps can help you drive sales and boost your ROI. Vizury has always focused on driving more clicks and sales volumes for our clients and hence we realized that Mobile Web Inventory Retargeting alone could not complete your mobile marketing strategy. Most mobile users switch between mobile web and apps in the blink of an eye and it is essential to reach out to users as they swapped between the two channels on their mobile device.

cross channel mobile retargeting

The mobile marketing eco-system is fragmented between web and app at a single user level and hence the need to reach out to users through both the channels is essential. After the first release of Vizury MobiConvert (which covered only Mobile Retargeting), we realized that Cross-Channel Retargeting is the next big step that can drive significantly higher volumes for our clients through their mobile assets. But the major issue for Web to App Retargeting is that In-App Advertisement ecosystem operates on device ID recognition mechanisms (Android Advertising Id for Android and IDFA for iOS) while Web Retargeting works with third party cookies. Our challenge was to build a user identification mechanism which is not dependent on cookies and by which we are able to identify a user across web and apps – all this while still complying with user privacy norms.

Safeguarding user privacy has always been one of our priorities during this product development. Vizury MobiConvert can effectively identify users across mobile web and apps without utilizing any user data that is PII. Vizury has integrations built to all major inventory partners in mobile web and mobile app and hence we are able to mine this user behavior data across mobile web and mobile app to identify a unique user across the two channels.

Currently Mobile Web to Mobile App Cross-Channel Retargeting with MobiConvert is in its beta phase and is able to identify users across mobile web and apps with an accuracy of more than 80%. MobiConvert’s Cross-Channel Retargeting is available in Japan and India and our clients have recorded up to 20% boost in their mobile web clicks/sales volumes. We will soon launch Mobile Web to Mobile App Cross-Channel Retargeting in all other markets as well.

Keep reading this space for more updates on Vizury MobiConvert.

Contributors: Shiju Mathew, Rahul Aggarwal, Akshatha Kamath, Harsh Vardhan

Technology behind Vizury Central (Part – 1)

This is the first of 3 blog posts that talks about how we built a scalable, feature-rich, front-end app at Vizury with an emphasis on site speed & performance, we call it Vizury Central.

Over the years we have accumulated terabytes of data through various collaborations in the advertising industry. One of the larger goals for us is to give meaningful insights to our clients that empower them to make critical business decisions.

We effectively wanted to build a platform, where both our advertisers & publishers can interact more intuitively with their data by slicing & dicing various aspects. We started working towards that goal a few months back.

Last December, on our 5th anniversary, we released the first version of Vizury Central, with a Dashboard for our customers. The Dashboard provided campaign overview in one glance. Over the last few months, we’ve been hard at work in incrementally adding features around our prime offerings Vizury WebConvert & Vizury MobiConvert.

Where we started

Our very first version of Vizury Central was built around the Spring Framework on the back end, the server end rendered templates & JQuery with Google Charts on the front end for presentations & visualization. While it solved the problem, the overall development cycle was getting really long with the additional overhead of interacting with an external agent for UI design. Also,    the visualization generated using Google Charts were more of static content where data needs to be re-fetched whenever the date-range changes.

Enter all-JS stack

We had literally hit a roadblock. That’s when we decided to rewrite the complete stack from ground-up, from the server to client, all in JavaScript. Yes, you read that correct. Our current stack that serves Vizury Central is written completely in JavaScript with node.js on the server side & angular+d3 on the client’s end.

Architecture of Vizury Central
ui architecture_vizury central

With browsers becoming faster by the day, we decided to move the entire rendering on the clients’ side. Our servers remain responsible for:

  1. Exposing easy to use REST HTTP API
  2. Enforcing access controls based on session
  3. Querying & transforming data from various Databases

Once a client visits Vizury Central login for the first time, the server sends the static HTML page with compressed JS & CSS bundles with far Expires Headers. All the view definitions for the application are bundled into a single compressed JS file.

Further communication uses the concept of ‘data-on-the-wire’, where only the JSON data is exchanged between the client & server enabling the transition between the different states of the application. I leave you with the awesome Vizury Central Dashboard.

Best user interface_vizury central

In the next part, we will talk about: how angular JS integrates with D3 & Cross Filter to provide an amazing, interactive drill down to slice & dice over 70,000 rows of raw data on the client’s side! Phew!

 

 

Retargeting in the Automotive market: Tata Motors Case Study

Automakers were among the first to recognize the value of an online presence. The trend picked up fast and the leading automotive manufacturers lost no time in taking online engagement in their stride. An online marketing strategy has become a must-have for every automotive manufacturer to stay on top of the game. Retargeting has become an essential part of every online marketing strategy in the present day and the auto industry is no different.

Tata Motors is India’s leading auto manufacturer. It owns some of the leading brands in the passenger car segment. Vizury has been a proud partner in their online marketing efforts for passenger cars. It helped Tata Motors attain a CTR that is three times the world average for display ads. The infographic depicts the Tata Motors and Vizury association which has generated great results.

IND-Tata Motors

If you want to re-use this infographic, please write to marketing@vizury.com for terms of use.

 

Learnings that helped us grow and sustain in China

China is one of the fastest growing ecommerce markets today and thus an irresistible prospect for companies on the lookout for new markets.

See also: China Is Dominating Global Ecommerce Sales, Study Says, Ecommerce in China – Online consumer on the Rise

However, things were not the same four years ago when we first stepped into China with our desktop retargeting offerings. Ecommerce was still booming and Vizury was the first company offering retargeting in China, some of our clients even called it “magic”. We had an adventurous ride thus far and have grown from a two member team operating out of coffee shops to twenty five Vizurians working out of three offices across China. Here are some important learnings that we picked up along the way:

China is a very traditional country like India and both the countries have been close since the days of early civilizations. However, China is very different from India in terms of the technology, internet growth, legal requirements and culture. Our first learning – standard advertising rules do not apply in China. For example, while one user group is price conscious and displaying the discount percentage on the ad helps in engaging these users. However, a vast group of users prefer to be associated with the brand especially for luxury goods. For such users, the brand logo on the ad helps grasp user attention more than the discount percentage!

Our second learning – China has swift market changes, so gear up and adapt to these changes. For example in 2011 DSP was very nascent but booming by 2012. Similarly, ad-exchange was a small phenomenon in 2012, but 2013 saw every ad network or DSP setting up an ad-exchange. You must be equipped to evolve your business strategy and products to around these changes to stay ahead in the Chinese market.

550+million internet users show such a diverse behavior that you cannot treat China as a single market. Users from cities like Beijing, Shanghai are well versed with ecommerce whereas users outside these cities are not. It is more like a group of diverse and complex markets and therefore essential to tweak your solutions based on these diversities and so our third learning – localization is the key to win here.

Mobile domination is a phenomenon worldwide and so in China. However, the mobile eco-system in China is very unique in terms of the mobile devices used, mobile OS, App stores. So our fourth learningA mobile solution that is tailored for China is a “must have” in your offerings.

Effective Data management for Affiliate marketing

This multi-part blog post recommends a framework for Advertisers to effectively manage data sharing with their partners and affiliates and boost their returns from various channels. Let’s begin by taking a look at some of the complexities involved with data sharing and a brief introduction to the Vizury framework. In subsequent posts, we will dive deeper into the framework with the help of specific examples and understand how some 3rd party players are advancing the evolution of this aspect of the ad-tech ecosystem.

Partners, affiliates and data

Channel partners or affiliates need data to engage users and lead them to transact on an advertiser’s website. A price comparison site like Junglee, Coneco or eTao might need product (catalog) data only. Sophisticated channels like retargeting need both user-intent and product data. As Advertisers begin to work with a motley group of affiliates, managing their individual data needs often starts turning nightmarish. It’s just data, why do we need a framework/process to manage it, you say? Well, here is what could go wrong:

  • Data security: Safe-guarding personally identifiable information (PII) or other sensitive data needs to be a part of the organization’s DNA. Like disaster recovery solutions, everyone knows it’s important but rarely is anything done till it’s too late.
  • User experience: If not done right, how you distribute data can impact other systems which may hamper users’ experience with your brand/website. For e.g. badly engineered affiliate tags could bring website page loads to a crawl.
  • Consistent and correct messaging: Imagine the after-taste when a user sees and clicks on an ad for a discount offer that actually expired days ago! Not great, right? Messaging is important and advertisers should moderate what affiliates say to potential customers like orchestra conductors.
  • Sub-optimal ROI: If affiliates aren’t enabled with the right data, advertisers are actually getting in the way of their delivering the best results they are capable of.

As an organization that works with hundreds of clients, we are no strangers to the aforementioned challenges. Catering to different verticals and advertisers requires us to deal with disparate data sets, each with a mind of its own! To make our lives easier, we have come up with what we believe is a handy framework to effectively visualize, monitor and enhance our data input setup. And, we would like to share this with everybody.

The Vizury Data Management framework

The following are the fundamental tenets of the framework:

  • Segregate data sources: Identify the different sources of sharable data across systems and segregate them based on the source type. E.g. Catalog data may be extracted from inventory management systems; User interaction data needs to be sourced from content management systems/website and so on.
  • Build a data dictionary: Enumerate the different data fields and define each field in detail. Definition could include attributes like purpose, format, nature of aggregation, temporal sensitivity etc.

Data Dictionary
data dictionary

 

  • Map data sources/fields to affiliates & share effectively: Understand who needs what data and whether the best practices for ingesting a particular data source are being adhered to or not.

Segregation and Mapping
Data Segregation and Mapping

In the next blogs in this series, we will dig deeper into each stage of the framework and go through detailed examples of sharing catalog and user intent data. I hope you will find this method useful. Keep watching this space for more!

What Makes Retargeting Successful?

An introduction to how the concept is different from regular display campaigns.

Retargeting as a concept and a tool has been around for some time now. However, its understanding and usage in Asian businesses is still not as mature as in the developed markets of U.S. and Europe. So what exactly is retargeting, how is it different from our regular display campaigns, and how can it fit within my digital marketing mix?

If the above questions interest you, we’d suggest you read on.

what is retargeting

How Does It Work?

Marketers use retargeting platforms to target these users with personalized banners when the latter browse other websites. Users find the banners useful to their current need and display a higher propensity to click it and revisit the brand website, this time to complete a purchase or transaction.

retargeting banner example

For example, let’s suppose user A visited the Agoda China website looking for a hotel room in XYZ checking in on July 10, and left without making a booking. Agoda will retarget user A when she visits a different website (e.g., news website) with a personalized banner. The banner may have several dynamic features like search, scroll, mouseover, change dates, etc. These make the banner content more contextual and useful for the user and as a result they deliver high CTRs (click-through rates).

What Makes Retargeting Successful?

The success of a retargeting campaign is dependent on getting a combination of three things right:

1. Targeting the right audience: The ability to identify the interest level of a customer, the revenue and margin potential, and urgency help increase the ROI from serving ads to each user.

2. Display the right content on the banner: The more relevant the content, and sometimes the higher the number of options displayed, the better the CTR. The way to get more clicks out of users is best based on their preference and your understanding of what they can be up- or cross-sold to.

3. The right kind of reach: With users browsing the Internet for a million different purposes all through the day, using a plethora of devices, finding the right time and place is critical to driving purchase intent.

What are your views on retargeting? As a marketer, do you think it’s something you need, would love to experiment with, or has it provided you the right kind of returns already? Tell us in the comments section!

Original article had appeared on Clickz.