Reducing Cost Per Engagements/Views in Facebook Advertising

OK, here's something that I have tested in the past: Choosing to promote Facebook ad sets for a bit longer (with the same lifetime budget) to see if it helps reduce the cost per engagement and /or view. 

Here's the theory: when you tell Facebook (or any ad platform for that matter) that here's $X and here are my limited dates, I need to achieve my ad objectives. GO! Facebook understands it as a matter of urgency...think about it as something that needs to communicated urgently to the target audience before it's too late...instead, what if you added a few extra days (where permitted) and then checked if the CPV and CPE did in fact decrease...wouldn't you be slightly happier?!

Of course, correlation doesn't necessarily imply causation. But if you've been running FB ads for your brand, you would have a good idea about which ad sets to exclude (ad sets where targeting is unusual, contests, buying models are different). Once you've removed the data outliers, customize the columns in Facebook ad manager to include 'Start' and 'Ends' dates from the Settings tab. 


Do a simple excel formula to calculate the number of days the ad set was promoted for (End - beginning date + 1) and you now have number of promotion days on the X axis and cost per view/engagement on the Y axis. Right click anywhere on the data and Add Trendine along with the linear equation and  you now know if there's a negative relationship between days promoted and cost per engagement /view - which is something great to know.

This approach might probably work best with always-on content where you can know for sure that the content has remained more or less the same (to compare).

What do you think of this approach? Something that you'd like to test?

Use of VPN in Middle East and Country Level Filters In Google Analytics

If you've looked at any Middle East based site report in your preferred analytics tool, you're quite likely to see traffic from US/UK. While not everyone uses it but several reports suggest 20-30% of online users in the Middle East are in fact, using VPN for various reasons (anonymity, access to content).

VPN Traffic presents a small challenge in Analytics - how do you know the actual country where the traffic came from...? You can't. Now add a layer of complexity where you need to setup country level views in Google Analytics. If you were to choose a custom filter where the (RegEx) filter pattern matches say, United Arab Emirates, you'd miss out on the VPN traffic from users accessing the site.

Working around the country level filter works best if you have a multi-country website, have a generic TLD with sub-folders for each country, say for UAE and for Saudi Arabia.

The RegEx pattern here is the /countrycode that is being picked up as the Request URI in your browser (Request URI would be the relative URL of a web page. So would become /sa/products.

The major advantage with using the Request URI approach is that even if a VPN user country shows as US, s/he would still want to access the local country page, say /AE. So, regardless of your country, you're looking at the pages browsed. The drawback is that this will also include traffic from other countries. e.g. If users in Oman do not have their own sub-folder, they would either be redirected to another store based on IP or would choose a specific country website to access.

To complete the Request URI approach, the steps required are:

  1. Create a new filter
  2. Choose Custom
  3. Include
  4. Request URI
  5. /countrycode

and voila! you're done. 

Keeping up with digital - My Feedly reading list for analytics and digital

Just thought of sharing my Feedly list here to let everyone know how I try to keep up with all the articles that come out. Here goes:

Web Analytics:



Does Social Media Drive Conversions?

A little while ago, an acquaintance expressed concern that social did not really work for the brand in terms of conversions. At first sight, it might appear to be the case. However, what’s required is a deeper look to understand how channels work together to deliver assists or drive conversions.

How Is Conversion Reported and Why Is It Important To Understand Attribution?
The standard Conversions report uses the last-click model to account for conversions. Making marketing mix decisions based on last click models could become problematic if the budgets are skewed towards particular channels.
I couldn’t find a simpler and better analogy than this lightning of a goal scored by a certain Real Madrid player in the Copa Del Rey 2014 final against Barcelona.
Watch this clip for the first 30 sec (or more…) and answer the question:
Who scored the winning goal for Real Madrid in this match?

Easy one. Gareth Bale!
Now what if the question was tweaked to:
Which Real Madrid players helped score the winning goal in the dying minutes of the final?
Yep, you’ll get a very different answer….

Messi’s cross gets intercepted by Carvajal. He heads it off to Isco, who does a short pass to Coentrao. Coentrao sees Bale and send off a through ball. Bale gets the ball and makes the run of his life to score and probably justify the heavy investment in him by Real Madrid.
Think about your marketing efforts for a bit. You could almost replace the player names with channel names (and investment with your marketing budget) and have a similar scenario where a standard last-click conversion report always paints channel X as the hero. The truth is that convincing people takes time (and sessions). This is the same reason why return visitors almost always have a much higher conversion rate. When you look at the complete effort required in making that conversion happen, you would agree that the assisting channels are not too shabby after all (in the case of soccer, making a defender/midfielder’s job not so thankless…).

Understanding channel contribution is all the more important for “soft” channels such as social due to the possible disconnect in timing between the brand’s content and the customer’s needs. A first time visitor to the site who lands via social media might not be willing to make a purchase immediately, but could agree to sign up for a newsletter and probably convert at a later stage in the future.

Exploring Channel Funneling
A good starting point in understanding social (or any other channel) attribution is the Assisted Conversions Report in Google Analytics under Conversions > Multi-Channel Funnels. For each of the channels listed, the key columns to look at are Assisted Conversions (all assists except last click), Last Click Conversions and the ratio of these two metrics.
Assisted / Last Interaction Conversions Ratio helps explain the importance of the social channel.
·         A ratio of higher than 1 implies that social as a channel helps more with assists than actual conversions based on last click.
·         A ratio of 1 means that social equally serves in assists and last click conversion.
·         A ratio of below 1 is when social directly drives your sales than assists.
Knowing this piece of information is perhaps the starting point of deciding how much credit to give to your channels. Attribution modeling is a topic on its own and has been beautifully explained by Avinash Kaushik in this post. The key is in deciding the level of importance to give to the interactions that take place via different channels on the path to conversion.

If you’re not working in analytics, walk up to your friendly analyst and start a conversation around this topic. Hopefully, you will find some nuggets of gold related to social channel that will help in appreciating the true value of social at your brand. The truth is that social can serve differently for brands but the only way is to start digging out such data to help make more informed decisions. 
What are your thoughts on this topic? Does social media truly work for your brand?

Two Improvements Analysts Would Love To See In Facebook Advertising

Even though I love Facebook Ad Reporting tool for exporting just about any campaign related data, there’s scope for improvement. Any brand using Facebook Advertising understands the importance of segmentation – not all Facebook users are equal and not even all Facebook fans equal either. The key here is segmentation – the more granular you can get, the more differences you’ll see in behavior for key metrics such as Cost Per Reach, Cost Per Click, Cost Per Engagement etc.

Keeping this in mind, here are the two improvements that analysts could benefit from:

#1 – Ability to segment Connections Targeting under Data Breakdown
Connections Targeting is crucial as it helps determine whom to target (or not) i.e. fans, friends of fans, users not connected to the page. Each of these segments exhibit a very different pattern in terms of engagement rate, the type of engagement and the cost per reach/engagement. In my experience, it’s almost an inverse relationship between the cost per reach and engagement for these three groups. Reaching fans is the costliest (as it’s very targeted) but it also produces the highest engagement rate out of the above three groups. Moreover, users not connected to your page are more likely to like a post while fans and friends of fans are more likely to comment or share on content.

If you were planning post promotion for your next post, you would ideally want to know the pattern and how to allocate the total budget between the various connection targeting options possible. However, this option can only be utilized if you have already created separate ad sets for each of such connection targeting options. By doing so, you can rank the cost per reach, engagement, website click, video play etc. So, if you ran a post promotion for 25-44 and chose ‘All’ under Connection Targeting, you cannot retroactively find out how different groups behaved with your content.

#2 – Ability to know Negative Feedback for Advertised Content
If you’re onboard with point #1, then you’d also want to know the actions taken by users after seeing your ad (any type of ad content). Within the existing reporting tool, there are plenty of metrics that you can download depending on the type of your content. It could range from page likes to post engagement to conversions via Facebook. It can then be broken down by the existing options in point #1 to find out the best performing segments.

What’s missing right now is the Negative Feedback (hide post, page unlike) that you can get for the entire post from Facebook Insights. For any paid content I run on Facebook, I’d surely want to know the kind of paid content that triggers a negative reaction from the user. Example, if friends of fans feel overwhelmed from promoted posts in their newsfeed and end up hiding the post, I would want to keep that in mind for the upcoming post promotions. There are way too many possibilities to list here but the idea is clear. Avoid the low performing ones (or negative performing) and go on with the positive performing ones.

What do you think can be improved within the existing reporting tool to help analysts make better decisions?

3 AdWords Video Optimization Tips To Improve View Rates and CPV

When it comes to any channel, segmentation is the key to understanding the differences in performance with regard to the peaks and valleys in trends.

As you'd expect from AdWords for Video, there are several ways in which data can be segmented to bring out the best in your campaign. If you don't have any data available from past campaigns, it'd be best to gather statistically significant data and then turn on the optimizations.

1. Segment Performance By Hour Of Day and/or Day Of Week

While it's currently not possible to create a pivot out of both these variables, the next best option is use them independently to find out patterns in viewership. When are users more likely to watch your content )(especially when it comes to pre-rolls via In-Stream Ads?

In my experience, both of these especially 'Hour Of Day' for shorter duration campaigns can help understand how your ad fits in between the busy lives of your audience. Is your ad viewed more during late evening / early morning hours (or not) when your audience is unwinding?

View Rates can be laid out by the hour (0 for midnight, 1 for 1:00 am...till 23, 11:00 pm - midnight) and be judged against the average view rates for the campaign. This would result in any of the three possible scenarios:
  1. Hours during which View Rate = Average View Rate
  2. Hours during which View Rate > Average View Rate
  3. Hours during which View Rate < Average View Rate 
As you can imagine, the third possible scenario will likely have higher CPVs and therein lies the opportunity to pause campaigns during low performing hours. This helps you improve your Avg. View Rate along with decreasing Avg. CPV (win-win!)

Note: In trying to show ads only during certain hours, care needs to be taken that the audience is not over-exposed to impressions. While this might not be an issue when targeting a wider demographic, capping impressions (on a per day/month basis) could helpful.

2. Making The Best Of Video Remarketing

While this is definitely a topic on its own, Remarketing in AdWords For Video is very interesting to me based on the options offered. 

The first thing to do is to come up with a list. This list can bucket users who have either visited your website (placing remarketing tags on your site), or visited your channel, seen an ad/vdieo, engaged with your channel etc. The full screenshot is below. The key idea is that remarketing cookies need to be placed on user devices to be tagged as individuals who have shown interest in your brand via various means. 

Once you know whom you want to remarket to, the list needs to be populated with users (minimum 100) for it to be used as a target. 

Think about the original ad for Mio Squirt ("Eye of the squirter") and how annotations were used to create back stories via separate videos. 

Remarketing the backstory to audiences who have viewed the main ad would most definitely result in our win-win (Higher View Rate / Lower CPV)!

Another example is how Crazy Egg realizes that I have visited their website in the past and shown some interest in tool. As a result, I have been targeted on YouTube (on a few occasions) to watch their pre-roll ad containing an introductory video to Crazy Egg and the product benefits explained. Suddenly, the content becomes much more relevant and makes me likely to pay attention to the video content. 

3. Segment Performance By Format and Network

Starting out your campaign, you might not have sufficient data on it but once it starts rolling in, it's a good time to checkup on Ad Formats + Network to see the difference.

Ad Formats can be In-Stream / In Display while Networks can include YouTube Videos, Search and GDN.

Using the screenshot option to breakdown performance metrics by these factors helps us understand our win-win targets. Is the View Rate much higher for In-Stream vs In Display (Note: In-Display counts a view after it has been clicked on while In-Stream counts it after 30 secs) or is there any significant difference in the CPVs?

How this helps is because Max CPV bids can be customized (increased / decreased) for the two formats based on information that we have now analyzed. Again, the two key variables that we are looking to improve are the View Rates and CPVs.

What are your ideas for optimizing AdWords For Video? Do let the readers know via comments.

Difference Between Boost Post and Promoted Posts In Facebook

Since both options are available in Facebook, thought I'd do a quick post on them and what are the pros/cons of each. 

Option 1: Boost Post
Boost post allows admins to quickly promote a post from the post itself or from the Insights > Posts tab.

Option 2: Promoted Post
Requires ad managers to manually enter demographic targeting for posts. This obviously takes longer in setting up.

Here's a quick summary of what are the main differences and why Promoted Posts option is much better than Boost Posts.

Criterion Boost Post Promoted Post
How to find it? Insights > Post tab Ad Manager / Power Editor
Duration Maximum 7 days None
Ad Placement Control  None Desktop Newsfeed, Mobile Newsfeed and Right Column
Interest Targeting None Keywords can be chosen
Behavior Targeting  None Digital Activities, Mobile users and Travel
Connection Targeting Fans and friends or demographic targeting used in posts All users, People connected to page (or not), Friends of fans
Bidding Options None For engagement, Clicks or Impressions
Ad duplication for multiple posts None Ad details can be copied and used for multiple posts

Here's a longer version of the main stuff from above:

Ad placement control: Having tested Right Column ads in Facebook, I think that paying for these would not be wise. It's almost as if users develop blindness to right column ads (on Facebook or other sites as well). If in doubt, you can always sort your post promotion data by placement to see check how your past promoted posts have performed (CPM, CPE etc).

By choosing Boost Post option, you lose control over where your post gets promoted. Under Promoted Post option, you can always decide if you want your promoted posts to appear on (desktop or mobile) Newsfeed and / or Right columns.

Interest / Behavior Targeting: Interest Targeting allows ad managers to choose keywords that may be directly related to their brand. Behavior Targeting helps choose particular mobile OS, device manufacturers among other options (think promoting a post to competitors).

Connection Targeting: This option lets Promoted Posts have a big upper hand on Boost Posts. The former allows users to create separate Ad Sets for friends, friends of fans, People not connected to the page, All and then monitor performance of each Ad Set (Also, because Ad Manager cannot separate ad data to show performance levels for such connections...Creating different Ad Sets is the only way out!). 

Boost Post has only 2 options: Promoting to all friends and fans OR using demographic targeting (applied in a post) to show ads to that audience.

Hope this post was helpful in understanding the difference between Boost Post and Promotion. Looking at the two options, Post Promotion is definitely the way to go for Ad Managers and allows for narrower targeting and detailed performance review.

Meetup #2 for Digital Analytics Dubai - A/B Testing For Meeting Business Goals and Attribution Modeling

Organizing Meetup#2 for the group - Digital Analytics Dubai. Will be happening on 10th May / 7 pm / Pascal Tepper.

This time, hope to exchange ideas around two topics:

  1. A/B testing for meeting business goals: In a very time pressed environment (as we all are), we need to prioritize tasks (and tests) that we can realistically conduct and implement to help meet our business goals. Hope to exchange cool ideas around best practices, tools and how to map the process of testing and learning.
  2. Attribution modeling: This one is a very interesting topic on its own. The feature is already available in most WA tools (In GA: Conversions > Attribution > Model Comparison ) to check out the different models. Each one has its pros/cons and hope to discuss these models.
Full details of the event are here:

[RECAP] Digital Analytics Dubai: Meetup # 1 - 27 May '14

Thanks to everyone who turned up at the first meetup event for Digital Analytics Dubai. It's always good to put a face to the id. Since this was the first meetup, a lot of interesting topics were discussed, so here's a rather detailed recap for those who could not attend:
  • Attribution for success of offline marketing: While attribution modeling within digital itself is a topic on its own, the fact remains that offline marketing still accounts for a major chunk (Perhaps 95%? in MENA). How can a site know that a offline channel really worked?

    • A common solution could be coming up with unique voucher codes for each channel (online or offline). What if there's no voucher code involved?

    • Another way around it can be the use of 'Vanity destinations' that can then be redirected to the original website to understand which channel worked. e.g. Website runs a billboard along the highway and wants to know how many people saw the billboard and came to the website. What if the billboard read as A person landing on the latter page gets redirected to and hopefully, converts on the site. Now, we have some attribution going on for the offline channel. Avinash Kaushik has discussed this in great length on his blog (a very old blog post but still relevant):

    • A third way would be the use of Custom User ID in Universal Analytics (by Google Analytics). In simple terms...To identify a user, GA places a cookie that has a unique identifier for that device (same user will have different unique cookies for his/her laptop/tablet/smartphone/desktop...thereby, being counted as 4 users)...What Universal Analytics does (now out of Beta) is allow sites to have a Custom unique identifier (Custom User ID dimension) that truly tracks a user across channels and devices. It has to be non-PII (personally identifiable information)...Think of it as using the primary key in a table to identify the row in table which has all the possible data...when it comes to sites, that would be Email addresses used in log-ins...for a telecom provider, it would be the phone number of the user. Even though the use of this feature requires some dev work, it is live and very much available
  • YouTube Analaytics: We also had a small discussion around how the 'Audience Retention' tab in YouTube Analytics can help channel owners find out the drop off rate for viewers as the length of content increases. What the audience retention rate tells a channel owner/advertiser is the average percentage of a video that was viewed and where did the drop off happen. For advertisers, this can be further broken down by Organic and Paid Views to see the difference in audience retention rates - Something which I think is a good metric to add to Views (BFF to it...)
  • Frequency of meetup events for Digital Analytics Dubai...We aim to meet once a month (probably 2nd Tuesday of every month) and starting with the next event, we hope to tackle one topic at a time. 
Stay tuned for information on the next event. To end the blog post, here's my favourite quote from Avinash Kaushik "All data in aggregate is crap!". Enjoy.

First Event for Digital Analytics Dubai Meetup Group

Excited to get the first event moving for Digital Analytics Dubai group off the ground and will be having the first Meetup event on Tue, 27th May at Pascal Tepper, Dubai Media City.

Hope to get to meet other community members face to face, have an open forum on questions and exchange information on topics and probably even set the frequency of events for the group.