Reebok: American-inspired, global brand that creates sports and lifestyle products.

In early 2019, we came across an opportunity to consult the media buying division of one of the biggest brands, Reebok, and help implement campaigns that meet and exceed their business KPIs and improve the overall performance of their campaigns.

With the introduction of new ad types, advances in custom audience targeting, and ever-evolving rules and regulations, staying on top of Facebook ad options is tough and you need a mastery of the space to remain competitive. By working together with Reebok KR, we focused on three core objectives to help them structure for scale:

Objectives:

1. Getting through initial Learning and Test phases faster with tips on how to give the Facebook auction maximum signals
2. Leveraging Facebook’s machine learning to drive more efficient CPA’s and reduce inefficient ad spend
3. Running effective brand awareness campaigns with the use of Reach & Frequency.

Ad budget

$300,000 USD per month for Facebook and Instagram.

When it comes to scaling up Facebook ad campaigns to anywhere from $1,000 to $10,000 in ad spend profitably, having the right systems and framework in place is key to ensure maximum results generated from every dollar that goes into it. This ad system will allow you to a) finesse all touchpoints and every single separate interaction that the product has with your prospects, and b) automate the entire process so it becomes hands-off once you dialed it in with the right rules and sequences.

From a high-level strategic standpoint, The 10X Ecommerce Ad System™ is a proven concept that systemically nurtures new prospects and efficiently communicates to bring about consistent top-line results to the business, without cannibalizing user traffic.

First and foremost, account simplification is something that Facebook recently came out to emphasize when it comes to delivery best practices. Account simplification is less of a hard strategy, but more of a mindset and an approach to how you buy on Facebook, where at its core the primary aim is to remove all constraints from the system and let Facebook’s decision-making and delivery optimization pull more conversions together and make better choices.

‍

‍Objective 1: 👉 Simplify Goals & Account Structure

We understood that Reebok had different pools of goals within their sub-departments (i.e. soccer, basketball, and running). Going into the test, as soon as we took over the account, we noticed that there was a significant amount of audience overlap – they had about 175 campaigns and we were worried that it was hurting their ad efficiency.

So we reviewed them and came up with a test plan where we essentially tested their existing efforts – BAU (business as usual) – vs. a more consolidated, simplified approach with just 15 campaigns. The main hypothesis was that we would be able to increase our scale at a higher efficiency since we were having a much stronger conversion signal.

In order to do this, we launched with 1 broad with no targeting parameters to start and had let the system find what their core customers look like and ran 1-10% segmented lookalike audiences off their core customers, with some manual bids against their historical CPA levels.

They immediately saw a 3x boost in their conversions from a 7-day test runway.

In order to stay true to the simplified approach, testing can be extremely challenging. We believe if you take simplification as a foundation of what you are doing on Facebook, you should 1) launch tests that supplement your existing efforts that don’t interfere with removing redundancies and/or 2) launch tests with the mindset that it will have an impact on overall performance but if you value the learning, it will be worth it in the long run.

What we like to do here at Swipe Right is to utilize Facebook’s reporting breakdown to see where the bulk of your conversions are coming in. Because you want to launch any test at the peak time where volume is much higher and there will be enough conversions that will mitigate the impact on overall performance. That way, you are more strategic about launching these tests – otherwise, you have to keep in mind that there might be auction overlaps and a potential impact.

Let’s now dive into the mechanics of the auction:

‍

Objective 2: 👉 Machine learning-driven Scaling

These notes are based on an ad system that’s been recently put together by a good friend of mine, Dim Niko – an Australian eCommerce extraordinaire.

Note that you’ll actually already want to have a fair bit of data from traffic before running these strategies. You can see that even in Stage 1, we’re using 1% LLAs.

So, ideally you want ideally 40-50 conversions per day for the Facebook machine learning to run at its full cap, but – because a lot of people can’t achieve that – you should aim for at least a minimum of 10 conversions per day.

So the formula goes:

             Historical CPA x 10 conversions per day = minimum budget per day

             Ex. you run a jewelry store that has an average order value of $150 with a minimum CPA of $50

             $50 x 10 = $500 per day is the recommended budget

‍

Stage 1: Initial Test Campaign

At this stage, we don’t have historical spend.

This $500 budget is recommended based on the $50 baseline CPA (i.e. we’re hoping to get at least 10 conversions a day). We set up 5 different ad sets as above.

1. Ad set 1: Lookalike 1% Purchasers. These should almost always be profitable.

2. Ad set 2: Lookalike 1% Purchasers x 5% Time Spent on Site (TSOS).

3. Ad set 3: Lookalike 1% x Events > n. For example, you create a custom audience with users who have landed on 5 different pages on your website. These are some of the highest engaged users whose behavior is expressing high intent.

Besides the list of potential Lookalike Audiences (LLA), youa ton here – but LLAs of what, exactly? And here is where testing can get a bit crazy. You can try these:

If you have the budget and resources to further split these individual audiences, you will also want to test the time frames. More often than not, those customers who purchased within the last 30 days are far more valuable than a bigger list of customers who purchased within 180 days. This is because one of the factors that the auction system takes into account when measuring the estimated action rate of an audience is their recency of that advertiser’ desired action.

1 Day
3 Day
7 Day
14 Day
30 Day
60 Day
90 Day
120 Day
180 Day
365 Day (only available of some)

Also you want to make sure you’re including all different types of creatives. That would be a 1-2 link ads, video ad, carousel ad, a canvas ad – basically one of each. The reason is that Facebook likes to break down its inventory like below. Basically, over half of the inventory available on Facebook are for video. So if you’re not running video ads, you absolutely need to. If you’re only running link ads, you’re only getting 22% of the auction and that may lead to spend inefficiencies as you’re only targeting a small piece of the auction.

‍

Stage 2: Lookalike Narrowing

At this stage, we’re narrowing down on our winning lookalike audiences. We’re taking our successful lookalike and plugging it into the new campaign.

Set up a new CBO campaign with e.g. $500 (if your CPA is $50, aiming for at least 10 conversions).

Target 1%, 2%, 3%, 4%, 5%. Note that this method is heavily reliant on a large amount of data, so e.g. that initial $500 from Stage 1 Campaign may not be enough to make the LLA high quality.

Again, we include more or less the same dynamic creatives. However, we turn off the dynamic creatives that are underperforming within that dynamic creative.

‍

Stage 3.1: Scaling Winners

Let’s say that from Stage 2, our ad sets with LLA 1% Customers and LLA 3% Customers are winning.

You would then start an entirely new campaign with 5 duplicates of the LLA 3% Customers ad set and run that at $500. You’ll also start another entirely campaign with 5 duplicates of the LLA 1% Customers ad set.

What’s happening here is that we’re targeting the exact same audience within a campaign. In the USA, a 1% LLA will be about 2 million people large. So this scaling phase will hit different pockets of that 2 million people. Of that, some will perform and some will not.

Again, note that stage 1 traffic should continue to feed into the later stages. You should always be testing the different kinds of audiences listed.

Stage 3.2: Scaling Winners out of the Winners

When Stage 3.1 yields a winning ad set, scale that by duplicating that particular winning ad set into yet another fresh CBO campaign with another $500.

Forcing Facebook to test better

Thus far, we’ve been on auto bid and auto CBO all this time. One thing to note here – Facebook often makes the mistake of prematurely shifting budget away from an ad set before enough data has been gathered to determine if that ad set is working well or not (i.e. not testing it enough).

So you should e.g. use a CBO campaign with $500, and then for your 5 ad sets make sure that you set an ad set minimum spend of $100 each.

Target splitting (scaling)

Out of all these successful ad sets, what we should aim to do is to target split our demographic so we can scale more. This step would allow you to find your most valuable customer demographic and dive deeper into that segment while giving you that opportunity to scale horizontally.

1. M/F
2. Age
3. Device
4. Region

David Liu from Futuremake.io recommends that, once you find that sweet spot using Facebook Ads Reporting tool, you should aim to broad match to them and combine them with an ‘Online Shopping’ interest (e.g. create a specific campaign that is aimed at “AU 45-54 | Online Shopping”).

When seeking to horizontally scale, could further combine this with other interests by using Audience Interest tool to find out other pages that people like, and targeting 45-54 year olds who – for example – like Tennis brands.

‍

Stage 4.1: CPA Reduction Using Manual Bidding

This method initially comes from a marketer called Alex Still, who talks about a target cost strategy. It can, in Dimitri’s business for hair extensions, drop CPA from $150-200 —> $30 (for a $250 AOV product) – which is incredible. Many people report 10x ROAs with this.

We start a new CBO of $5000 (daily spent max $500) with our absolute best winners and creatives from previous campaigns. We duplicate that winner into 10 ad sets.

Previously to this point, we’ve been using cost cap bidding i.e. just using Lowest Cost + bid cap, for our bids. What that means is that we’ve been telling Facebook: “hey, go and find me the cheapest conversions, but I will spend no more than $100 per conversion.”

However, for these ad sets in Stage 4.1, we will set a target cost bid. This is where we say to Facebook:

“Hey, I want all my conversions to come at around $100 per conversion (e.g. cost per conversion of $90-110). ” It won’t look for drastically cheaper conversions, but it also won’t spend much more than your target cost.

This is why we’re putting the target cost of $5000/day, but we’re not actually spending that much. We need to set our budget high so we can force Facebook to spend. A good figure to go by would be to spend 50x your CPA. You’ll have to play around with it to get it to start spending.

Our Manual Bidding (MB) percentages:

MB10% = Manual Bid @ 10% of Target Cost

MB20% = Manual Bid @ 10% of Target Cost

etc.

Up to MB80%.

Ad set minimum

We’ll also want to set ad set minimum budgets to about 3x our CPA. Again, it won’t spend the full $150 (in this case), it’s only going to spend if it’s profitable.

How to prevent overspending

Use a Rule to actually automatically stop the spend e.g. if you spend $500, so you don’t wake up to some crazy bills.

Principle

You want to see traffic trickle in. This is aiming to get the most valuable traffic at the cheapest price. The whole aim is that we’re wanting to get conversions for lower costs than what we were doing before.

Then

Determine the winners out of your ad sets from Stage 4.1. Then plug it into Stage 4.2.

Stage 4.2: CPA Reduction Part 2

Take the winner out of Stage 4.1, then plug it into a new CBO campaign whereby we have ad sets that match Manual Bidding of +-5% and +-10% of the one that won.

So for example, if your winning ad set from 4.1 was the MB50% ad set, then you’ll create a new campaign with MB50%, MB45% and MB55%, and MB40% and MB 60%.

Again, you’ll set a huge amount of budget.

‍

AND DON’T FORGET

• Retargeting: all of this has essentially been scaling up budgets for top or middle of funnel traffic. But don’t forget that running retargeting campaigns are also going to be an extremely profitable action to take.
• Rules: make sure you have a stop loss for traffic that cuts off unprofitable ad sets.

‍

And then scaling from $10k to $100k

Set up a rule where the campaign budget is doubling every 2-3 days. So for example, you will be doubling $10k -> $20k -> $40k /day etc. You’ll need a hell of a lot of creative to refresh with if you plan to do this.

‍