fbpx

Category: Advertising / Marketing

How Facebook Machine Learning Prioritizes Users In Audiences

Facebook’s machine learning is meant to do 1 job. The job is serve your ads in the cheapest price possible.

What this means is that if you run an engagement ad and select a worldwide audience for it, Facebook will get you the engagement from the cheapest geo-location which could often be Philippines or India or another country like that.

Similarly, if you run a conversion ad and put in a large amount of countries in your audience section, Facebook will start prioritizing your traffic towards countries that are likely to convert but also where cheaper impressions can be served. So you’d see a large ad spend being done in Brazil or Mexico or Kuwait etc.

Many times, marketers want to sell their products or services worldwide but also have a preferred or prime geo-location which they want to serve first or primarily or in majority.

If you want to sell your goods or services in US and your goal is to have 50% buyers from US but also have some buyers across the world, you should ideally segment your ad-sets. What this means is you should create a separate ad-set for US and another one for rest of the world.

As a media buyer, it is no longer your job to find buyers. That is something Facebook does for you. Your job is to understand how Facebook’s machine learning works, which does it’s job well, but is far from perfect. As a media buyer you’re supposed to identify its shortcomings, and create your ads in a manner to get the best our of their machine learning.

How Facebook Campaign Objectives Work – A Case Study

It is a matter of common sense for years that Facebook tags groups of people from within an audience for various objectives. What this means is if you run an ad for “cooking” interest that may have 100 million people in the audience, and your campaign objective is engagement, Facebook will only show your ads to users tagged as engagers from within the 100 million audience size.

Similarly, if you run a conversion ad on the same interest of “cooking”, Facebook will show your ad to “purchasers” from within the 100 million audience.

But there’s more to the story.

A friend of mine asked me whether he should select campaign objectives that are cheaper in nature for really warm audience. For example, he wondered if “reach” or “engagement” ads should be run for custom audiences of people who have added the products to cart or initiated check out etc.

I told him that this probably won’t yield better results than the conversion objective since Facebook doesn’t just work by tagging users but takes many other data points in consideration. So I ran a test. And here are the results

Click on the image to see full version
  1. Both the audiences are 100% identical. RT Reach has campaign objective of Reach. RT has campaign objective of Conversion.
  2. The CPM of reach objective is 1/10th. It was 10 times cheaper to run the reach ads on my warm audience.
  3. Despite the warm audience, CTR was 1/3rd for reach ads.
  4. Despite lower CTR, more clicks were generated for reach ads. (10 times more impressions and 1/3rd CTR = 3 times more clicks)
  5. Despite more clicks and cheaper CPC (0.11 for reach & 0.28 for conversion), cost per ATC, cost per IC, cost per Purchase was much higher for reach ads.
  6. Cost per purchase for reach ads was nearly 2.5x of cost per purchase of conversion ads.
  7. Reach ads saw ROAS of 0.76 while conversion ads saw ROAS of 2.22

Conclusion

While Facebook could be taking thousands of data points in consideration, I want to highlight the basis of why the reach ads underperformed.

Campaign objective doesn’t just tag users, but also decides whether your ads will appear early in the newsfeed or down below capturing higher attention vs lower attention of the same warm users.

It may also take into account the time of the day for when each individual purchaser makes a purchase. By running reach as campaign objective you’re reaching your “purchasers” at a time when they won’t initiate a purchase.

The conversion objective could take individual placements and platforms in account for each individual user etc.

In summary, the campaign objectives do not just work by “tagging” various groups within the audience but also take into consideration many other data points.

Gaming The System & Little Tricks

I finished yesterday’s blog by asking if gaming the system is a good idea to make money or not. I have gamed the system all my life. When I ventured into internet marketing, the first platform I drove traffic from was Digg.com

It was a social news site kind of like reddit that doesn’t exist in its original form anymore. It was an insane source of traffic and Zeeshan mentored me well on how to really make the best out of it. We could get most stories to the front-page on a daily basis. We drove 10s of thousands of unique visits. Not just that, our content reached the eyes of editors of the largest publications in the world which would often result in backlinks and further traffic as well as SEO juice from them. I built my blog SmashingLists almost entirely out of Digg and sold it for a pretty hefty amount back then.

I loved it. It wasn’t just traffic, it was really high quality traffic. After the demise of digg and trying a few other things like StumbleUpon etc, I ventured into viral Facebook marketing. I did that briefly for about 3 months. It certainly was the darkest shade of grey. I don’t encourage anyone to choose this shade in their lives. It’s not worth it.

After that with my co-founders Saad and Zeeshan, we leveraged the organic traffic from Facebook by building and acquiring Facebook pages. Very white hat and we did it for the longest time. We built many websites and made a ton of money.

In summary, gamification of the system has been the heart of our internet marketing journey. I’ll go on to the point to state that if big tech companies claim that they haven’t done it to “hack growth” they lie. Wasn’t Facebook built by scrapping off the student list of all Harvard students? Aren’t AliExpress affiliate ads served on Torrent website popups? I have seen all these mainstream apps like ride-hailing, food-delivery, pretty much everything, capitalizing the grey areas.

Growth hackers study the systems, the AI, find the shortcomings, and capitalize on them. That’s what they are designed to do.

But some people suggest gamification is a small guy game. A few days ago, PG published this tweet

I agree with him.

I have seen or known 100s of people who have made millions and tens of millions all by capitalizing the “little tricks”. It’s totally possible. It works. There are probably a million case studies of millionaires who made it through beating the system.

Although, really big money, the unicorn status, the billions, are not made with little tricks and gamification. They are made by solving a problem so big that it helps millions and tens of millions people use the service or the product. I’m still willing to bet though, that the growth of these companies are still carried out using the “little tricks”.

Since people from emerging and under-developed world are often not so well off, to them $100 seems like a big deal and they would happily settle for little tricks and gamification as long as it provides them the opportunity to make that $100 and a road that would eventually lead them to become somewhat wealthy.

To finish this off, if you game the system, you’ll make it. If you build a product or service that helps millions of people, you’ll make that every hour what you’ll make with gaming the system in your lifetime. But even while you build a product or service that helps the people, don’t forget to game the system along the way.

Building Lookalike Stacks, And Why They Work

I have been following some fellow marketers run this strategy for a while and I had tested it out myself too a few weeks ago. I’ve seen decent results with it. Unfortunately, I didn’t have a lot of data to work with so I couldn’t test it at full scale but I believe in the concept for sure.

The strategy is called lookalike stacks and it’s dead simple to implement. You create 1% lookalikes of various options like 1% of view content, add to cart, initiate checkout and purchase and stack them together in 1 ad-set. You can do the same for 2%, 3% and so on.

At nearly $100 spent, purchase conversion value is $255 yielding ROAS of over 2.5

There are two reasons why I think this strategy works better than most other lookalike configurations.

Firstly, Facebook prefers broader audiences over narrow audience. When you let Facebook work with broader audiences, it has more room to play with, find buyers, and generate cheaper sales in turn. When you stack 1% of everything together, the audience size is much larger than 1% of individual lookalikes. It is often 2-3 times larger. Sure you could individually use 3% of purchase or 3 % of view content to achieve same audience size. However 3% is never going to be as good as 1%.

Secondly, best lookalikes are created when you not only use more qualified events but also have a large amount of data in them. For example, if you have 1000 ATCs and 100 purchases, your ATC lookalike is likely to be better than your purchase lookalike because Facebook had 1000 people to create lookalike from. Although purchase is a more qualified event than ATC, it will create a more qualified lookalike once there are large number of purchases in the data-set. When you create a stack, you’re able to leverage from the best of VC, ATC, IC, Purchase together in 1 ad-set. This kind of ad-set is the best of both worlds as it has both: lookalikes of most qualified events (IC, PUR etc), and lookalikes of events with most data in it (VC, ATC etc).

I hope I was able to explain myself just fine. If I didn’t, please feel free to ask me questions.

Scaling In Non Mainstream Geos

The past few days I have been working on a highly competitive product that I know hundreds of sellers are working on. There’s demand for it, but the product has lower sale price of $20 and is competitive to sell with paid media.

After analyzing competition and discussing strategy with Saad, I decided to sell the product in non-mainstream geolocations primarily targeting middle-east. This allowed us to generate sales at cheaper CPA than US/UK/CA/AU/NZ/EU and we were able to scale freely in untapped markets.

The CPA for “other geolocations” can be seen $2 lower than US and $6 lower than UK/CA/AU/NZ /EU

Sales from Middle-East, Asia, Africa & South America costed us 25% less and also accounted for 48% of total sale volume with US only accounting 17%.

Our average order value was 40% higher in Qatar and UAE in comparison with average of all other geolocations. 8% of our sales came from Mexico. We also received substantial volume from Chile, Hong Kong, Taiwan, Brazil, Oman, Bahrain and Caribbean states.

Our seed data was built using Singapore and Hong Kong which I think have high conversion rates and purchasing power. The seed data was then used to create custom audiences and lookalikes to scale in all other locations.

Over-all I’m really excited by the new avenues this unlocks because a few years ago, I couldn’t have imagined this. Selling in these geo-locations is one thing, but scaling there is a different ball game.

Diagnosing Your Sales Funnel

I received an email from a reader who needed some advice regarding the diagnosis of sales funnel. I’m going to keep this short as I’m busy with launching some more campaigns right now, but I hope that I leave some value here.

There’s no rocket science here. The first thing for you to do is to be aware of all the steps that your potential customer is going to take in order to purchase something. Starting off from seeing your ad on Facebook or other platforms, watching/engaging with it, clicking on it to reach your landing page, reading the product description, adding the product to cart, initiating checkout, adding shipping/payment info, and eventually committing a purchase. These are the steps that the customer goes through in most of my advertising. It can be different for everyone.

The second thing for you to do is to find where the breakage is. If you’re failing to see results, you need to identify the point where something is going wrong. If you’re doing video ads and have good watch time, your creative and your targeting should be okay. So you’ve diagnosed this step of the funnel and should move forward. If you have a good CTR (1%+ for Facebook), it means your ad copy was convincing and the customer is interested in knowing more about your product.

If your bounce rate is low and your time on site high, it means your landing page was engaging and informative. If over 10% of the people on your LP add the product to their cart, it means your ATC button placement, color etc is good. If over half of those who added the product to cart then initiate check-out, it means your cart page is not broken and created as it should. If over half of those who initiated check-out, purchase the product, congratulations you’ve made it.

Key metrics that I really like to focus on: average video views in seconds: 10 seconds or more. Average video views in percentage 25% or more. CTR minimum 1%, ideally 2%+. ATC rate, I like it over 10%. Initiate check out rate, I prefer having over 5% and conversion rate should ideally be 2.5% or more. Below are the today’s stats for one of my stores

I hope you find this useful.

This Seems Relevant Today

This could have been me had I stopped yesterday which by the way I wanted to.

This is me instead because I hung around longer.

I spent the past couple of days trying to optimize a new product launch. All metrics looked great. Every step of the funnel just as I wanted. I had low CPM, high CTR, low CPC, low CPATC, low CPIC, but.. also low conversion rate. For those who don’t know what am I talking about, I had low cost for everything, but the number of users purchasing were also low which was something I really didn’t expect to happen.

Due to this my cost per acquisition was higher than where I wanted it to be. Instead of making money, I was losing money until I launched the retargeting campaign.

For those who don’t know, retargeting is reaching warm audience or potential customers again. People who showed purchase intent but didn’t purchase. My retargeting campaign brought me really cheap sales. So cheap that it offset all the loss that other campaigns caused. Not just that, it turned the overall campaign around and made the product launch profitable.

This showcases two things. 1) Retargeting is really really powerful. 2) When you’re thinking of giving up, hang around just a little bit longer.

Using Case Studies For Marketing

One of the things that people tell me is that when they run ads they get a lot of irrelevant traffic or leads although they are confident that their targeting is accurate. When you’re selecting a large interest with an audience size in millions, you’re obviously going to reach many irrelevant people just because of the sheer size of the audience.

One of the things that we do to improve this traffic quality as well as our conversion is to do case studies on the pain points and their solutions. For example, if you’re trying to sell a SaaS subscription, instead of trying to reach your potential customers directly with the ad of your product, you should do an ad of the case study.

If your product is an e-commerce product discovery tool, you should do a case study about “how a store owner made $37,000 with this product discovery strategy”. Once you run an ad for this case study, you’ll be able to collect very relevant clicks. You can then retarget this traffic with your product ad. You could also create a lookalike of this case study audience, and then run your product ad for them.

The more expensive your product is, the more number of case studies I recommend you to do.

Why Selling Digital Products Could Be A Good Idea

While e-commerce is a great business and my focal point of attention these days, I’ve also spent a ton of time selling digital products as an affiliate which were mostly ebooks, newsletters, courses and forum memberships.

I have written much about the upsides of e-commerce, but probably not so much about the downsides. The first major downside is that tangible products have repeat costs. In order to fulfil each purchase, you have to source the product as well as pay for the shipping. This can obviously be avoided in a digital product where you spend a one time cost in manufacturing your product and can sell it again and again.

The second major downside is the liability. I’ve never been bothered by the risks of selling advertising on websites and also not too much bothered by the risk of selling digital products. Physical products, however, can go wrong. They can malfunction, cause damage to the consumer and this is a risk worth considering.

As already mentioned, digital products come with higher margins and are more profitable. This is also why you see many marketers resort to selling courses in the end because it is a higher margin business. You create the course, may be also incur a cost in doing so, but on an on-going basis your only cost is marketing. This leaves a much higher budget for you to make a profit.

I do think that most course sellers do it out of desperation and many are not even fully equipped with the knowledge that they try to sell. However, selling courses itself is not necessarily a bad thing as many of them come with a lot of value. The fact, although, remains that you can make more money by telling people how you make money and less by actually trying to do it.

Why Should You Always Duplicate Your Ads

If you’re familiar with Facebook advertising, you may have seen that some people always run multiple copies of the same ads in an ad-set. Those unfamiliar with this strategy always wonder, why would someone create 2 identical copies of the same ad and place them in an ad-set. Here’s the reason why.

When you target a large audience (for example 1 million to 100 million) which Facebook also encourages you to do so, not every person in your audience (interest/behavior) is going to be identical.

When you place two identical ads in an ad-set you’re hoping that your first copy will be seen by a small pocket of your large audience, and your second copy will be seen by a different small pocket. Based on the performance of the audience in those pockets, Facebook will continue to find similar audience using it’s machine learning capabilities.

It is obvious that one of the pockets of the audience would be superior to the other one and by having multiple copies you’re giving their machine learning a better chance of spending budget in your interest in a more optimal manner.

I found this difficult to convey over the text, but I hope that I’m able to do so. If you have any questions, please feel free to ask in comments.