Back to blog

Counterintuitive marketing to lower your CPA

February 11, 2020
In this blog post, you’ll read how to drop CPA for your mobile app advertising — even when you are spending a couple of thousand dollars a month. It’s a real case study, based on our longstanding existing clients. Let’s start!
Follow the yellow line – this is what a 57% drop in CPA looks like.

Paid advertising specialists like myself over at Agency have the solutions you’ve been looking for. I’ve molded our results-based strategies out of a crucial question American investor, and entrepreneur Tim Ferriss posed in his book Tools of Titans, “What if I did the opposite for 48 hours?” Let me walk you through how acting counterintuitively can dramatically lower your CPA.

Part 1 — You need to be organized

There’s a reason my friends call me “the Excel man,” and it’s not because I’m constantly excelling. Instead, I’m obsessed with storing info about everything (right down to my lunch and dinner) in a Google Spreadsheet. I mention this because keeping things logged and organized is a crucial first step in lowering CPA. Let’s start with some basic organizational rules.

  1. Keep track of every change you make in your Facebook Advertising account with an ‘Activity Log.’
  2. Run your experiments like a scientist — write down a hypothesis, observations, results, and learning. Even if the test is a failure, drawing conclusions is key.
  3. Have a place where all your future marketing ideas live — an Ice-box to keep them fresh and findable in one convenient spot.

Airtable is a fantastic tool to have all this information in one place.

Activity log

Imagine you’ve made some minor changes to your ad account. Alongside these, you’ve made dozens of other minute adjustments, and after a few days, you see a decrease in CPA. The problem is you don’t remember all the changes you’ve made so you don’t know what has had an impact on decreasing CPA. That’s where the activity log comes in; it keeps track of everything and provides a central source of truth you can use to analyze what actions have impacted your low CPA. As a bonus you can send all activities from Activity Log to Slack channel where you communicate with your client — this way, your client will be updated will all actions you are taking on their ad account.

This is how we log everything

Experiment Log

This is one of the most important logs you have. As I mentioned before, it’s essential to run every experiment very methodically. You have to have a hypothesis, observations, and most importantly, learnings. This process will prevent you from continuously making the same errors.

This is how we log our experiments


This is where all your random ideas for testing live. It’s the place where you determine what to test first based on impact, ability to execute, and gut instinct (ICE score). At Agency, we use the Ice-box to log our customer’s ideas, ensuring they get tested alongside our own. This creates a collaborative dialogue between company and agency where unconventional ideas are encouraged and given due process.

This is how we store our ideas

Part 2 — the work begins

Finally, in this paragraph, I want to share the knowledge we’ve gained from spending a seven-figure amount of British pounds over the past 10 months on different campaigns. This is a list of tactics based on experiments that gave us our best results yet and laid the foundation for scaling our campaigns, eventually leading to a ridiculously low CPA our clients CEO thought he could only dream of.

1. Micro-targeting

Your app has its ideal customers. You, as a CEO or marketer, should be able to list at least 3 of them in your sleep.

Action points:

  1. Create a persona sheet
  2. Create a copy, creative, and an app landing page that provides solutions for your personas’ needs and fears
  3. Align each persona with interest targeting on Facebook
  4. Run an experiment to determine which persona works best for you

Description & Learnings

For this recent campaign, we prepared over 60 personas, one of which was “gamer.” The copy and creative were tailored to people who like playing video games, mentioning the universally popular racing game Mario Kart in this pursuit. Then on Facebook, we targeted people interested in gaming, gaming consoles, and of course, Mario. Thanks to this experiment, we discovered that this creative resonated exceptionally well with our target audience, we got really low CPA and were able to push the ad to broader audiences. The ad, again, performed well, and CPA continues to decrease.

2. Audience cycling

No, we don’t ride a bike to find new audiences. Audience cycling is a constant change of audiences in your campaign to find the best ones. Our setup is such that in a campaign, each ad set contains only one audience.

Action points:

  1. Prepare a list of audiences you want to create lookalikes of.
  2. Start testing from 1%, but don’t put a limit on it. We have seen great results from 3%, 7%, and even 10% lookalikes.
  3. Determine your KPI’s, so you know when an audience is exhausted, and you need to test a new one.
  4. Have a file where you keep all info on what Lookalikes you’ve tested and with what results (hello Airtable!)
  5. Cycle continuously whenever you see a drop in performance (but first check your Activity Log to ensure the fall isn’t the result of other recent changes).
  6. You can follow this rule — if CPA increases in the last 7 & 3 days to more than 10% above desired KPI turn off this audience.

Description & learnings

Our audience cycling wasn’t perfect at the beginning. This is likely because before we started implementing it at Agency, we hadn’t heard of another agency doing it. After a few iterations and perfections, we’ve found that changing audiences on particular days has significantly worse results than changing them when performance drops. Also, we have found that high Lookalikes (like 7% or 10%) work really well if you combine them with good creativity (remember Mario Kart?). Incorporate these learnings, and you’ve mastered the art of audience cycling.

3. Interest targeting

For years we’ve heard that Lookalike works even better than interest targeting. This is conventional wisdom we at Agency also followed until we started thinking counterintuitively. By diving deep into interest targeting, we’ve been able to find gold, with our campaign garnering approximately 30% lower CPA than equivalent Lookalike campaigns.

Action points:

  1. You need a persona sheet (remember this from micro-targeting?).
  2. Prepare a list of interests you could target. If your fan page is big enough, use Facebook’s Audience Insights tool to generate these. If you’re starting out, make a competition analysis.
  3. Don’t limit yourself to only the interests which first come to mind. Go wild, try to use something different from everybody else (remember when I said counterintuitive?).
  4. Keep your interest targeting in one file to track individual performance and determine the category or audience sources that work best for you.

Description & learnings

Our most significant learning was that we could lower CPA by about 30% when targeting interest audiences instead of just Lookalikes. We did a lot of research and tested audiences that at first look had nothing in common with our target audience. However, please keep in mind that first, you need to have good creative assets, for this to work.

4. Manual bidding

I was told once that if you’re doing any bidding other than auto Facebook thinks you’re pro and puts you in a different class of advertisers where you can get better results. Like Santa Claus, this is probably an urban legend, but what’s true is that you can get great results using non-auto bidding. For now, Facebook offers manual bidding with bid cap, target cost, and cost cap.

Action points:

  1. To test manual bidding, you need to have tried-and-tested creatives and audiences. I don’t recommend testing manual bidding with untested creative or audiences.
  2. As always — prepare a testing sheet in Airtable.
  3. Populate with info about audiences and bidding types to keep track of results.
  4. Run campaign. Test one type of bidding each time, don’t overcomplicate it.
  5. If you want to determine perfect bidding try with your average CPA.
  6. Think counterintuitively — test bidding as high as ten times your CPA to bully your competition. It’s working.

Description & Learnings

Thanks to manual bidding tests, we were able to lower CPA another 20% on average. If you’re on a big budget, you need many audiences and creatives for this as campaigns can stop delivering results very fast. Luckily small tweaks in creatives or new audiences can give things a fresh flow. Now you’re ready to use manual bidding to outbid your competition.

Part 3 — take action to lower CPA

That’s all, my friend. Now go and implement these tactics to reduce your CPA. To be honest, each of these tactics provides enough material to be its own article (which might be forthcoming if you like this one) as there are so many variables you can test.


Limitless creativity in automated production process

Last year I shared this blog post on how Paintgun sets order in creative production “chaos” explaining how our weekly sprints work. And since then we have been receiving a lot more questions about our production, mainly asking about how we stay creative in such a structured and always repeating process. The hack that we use is simple but requires a very well-planned action plan. And here is how we do it:

Want to reach users that love your app?

Let’s build the future.

Coming June
Join now for early access.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
By signing up you agree to our Terms & Conditions