Howdy Haveaclues!
I hope you had a fun and/or relaxing great weekend! I had a wild, lovely, jam-packed Brooklyn Saturday.
Before we get into it, I’d love to hear from more of you! If you haven’t already or you’re new here, please quickly smash the reply button and tell me what you want to hear more of from me. I’ve gotten some really helpful and supportive feedback so far. Also replying to my emails helps with its deliverability!
Stick around until the deep end of this email, but don’t skip ahead, I have things to give away!
First, I wanted to clue you in to my current “process” for making this newsletter you’re reading right now:
I see a tweet I hate, someone asks me a difficult question on a consulting call, or I get frustrated by something a client or FB rep said.
I get angry 😡 or inspired
I tweet about it (or is it X post? 😅)
I procrastinate, get distracted, or focus on other stuff
I write this newsletter
Allllrighty, let’s dig in to what made me angry this week:
Some marketers don’t trust Facebook’s system and other marketers trust it entirely without questioning it.
I keep hearing fellow marketers talk about how they don’t trust Meta’s system. Here’s how the story goes: “I used to trust the system to optimize my ads/budgets, but then [the system did something wrong] and now I don’t trust it and I prefer to control everything manually.”
On the other hand, too many marketers trust the data and optimizations they see in Ads Manager and don’t question them or dig into them to learn more about them because they don’t understand how the system can be cheating.
Facebook ads requires a balance of trust and skepticism.
1. Trust in Optimization: Facebook's machine learning can optimize better and faster than humans, making predictions that align with your ad goals.
2. Understanding its Limitations: The system makes predictions and predictions aren't always right. Mistakes can happen, leading to suboptimal results. Also, the system can learn to cheat or be lazy without the proper guardrails.
Btw, if you want a safe place where you can talk with other professionals and experts about complex nuanced topics like this, it sounds like you need my sponsor: Foxwell Founders.
It’s the ideal community for business and agency owners/leaders, freelancers, consultants, media buyers, creatives, and CRO specialists who want a private place to talk about any and all of these topics with others.
There’s no spam, no selling, and no anonymous trolls (unless you count me as a troll, sorry!).
There is tons of information/experience sharing, intimate cohorts, group calls, supportive feedback, useful resources, and community events.
I owe Andrew and the Foxwell Founders community a huge thank you for helping me make some amazing friendships and for helping me launch my account audit template.
If this sounds like it might valuable for you and you want to join me and the rest of the incredible members, click here to get started. You can always try it for a month or two to see how it feels. I’m sure you’ll get a ton of value out of it!
Back to today’s topic, trusting Meta’s ads optimizations and machine learning system.
I've advertised on all the major social ad platforms and every single one of them has reporting that makes their own platform performance look good.
They don't care if they take credit for action driven by other efforts.
Every platform is incentivized to count the most conversions possible so it looks like their platform is efficient and effective to advertisers so they can win (steal) more ad dollars from them (that’s you!).
The problem is that if you’re running ads on Meta, TikTok, Snapchat, YouTube, Pinterest, X(?), etc at the same time, all of those systems already have data on users who have shown signals they might be more likely to buy than others. The systems know who your recent visitors and engagers are, even if you don’t manually choose to target them, the system can and will target them if you let them.
For example, if someone clicked your TikTok ad (or searched for your brand after seeing the ad), they can then easily be retargeted by Meta and both can take credit for the sale.
The systems are not trying or incentivized to get you incremental conversions, they’re just trying to get as much attribution credit as you allow them.
If you took the number of conversions each of them drove in one day and added them up, it would be waaaaay higher than the actual number of sales you had that day because those users are using multiple platforms and there’s overlapping efforts. Yes, attribution platforms like Northbeam and Triple Whale help with this, but they can’t be 100% accurate.
Even worse, the ad platforms are using machine learning, which can learn to cheat or take shortcuts. Not only is it bad that the system can and will go after your warmest users when you let it, but if you let it, the machine learns to do that easy thing more.
If the machine learns to track conversions from low hanging fruit (warm users) via 1-day view impressions, it can learn to allocate more spend towards more users on cheaper placements and just hope that they’ll convert so I can get credit for it, even if those placements are less likely to have a tangible impact or get users to click. Is that helpful for your business at scale? Probably not.
This isn’t a major issue for smaller advertisers, but as your business gets larger and you spend more (and on more platforms), the system can get lazier until it eventually becomes a completely non-incremental money furnace.
Ok, so now that we know we can’t trust the system entirely, that means we can’t trust it at all and should manually control everything, right?
Nope. Not at all! Even with these flaws, the systems can still do better at optimizing for sales than we could.
We need to trust the system because:
We don’t have access to all of the data the system does.
Even if we did have the data, our human brains wouldn’t have the capacity to optimize all of the dimensions of available data.
The breakdown effect makes it difficult to properly analyze individual ads or breakdowns because the system is making predictive optimizations on how to best deliver the ads.
You can’t evenly compare the cost per conversion of ads that received different amounts of spend, but you should trust that the system is trying to get the lowest possible for your desired action.
For example, turning off a top spending ad because it has high costs or low ROAS can be extremely detrimental if that ad is getting the most spend because it’s able to scale more than your other ads.
(Note to self: I need to do a much deeper dive into this topic in the future. Please reply to this email with “breakdown effect” to let me know you’d to hear more about this)
Multiple overlapping audiences make it impossible for us to know which ads in which ad sets or campaigns were seen by which users. A user can be shown multiple ads from a variety of ad sets and there’s no way of knowing which ads/ad sets had any material/tangible impact on conversions, except for the last ad they viewed/clicked before converting.
Our limited human blind spots and biases make us likely to inadvertently skew things towards our existing beliefs.
There are more valuable things we can do with our time to help scale our businesses.
So, what can we do to make the system work for us?
Give the system good data. Make sure all of your desired conversions are getting sent to the system.
Be mindful of how the system could cheat or be lazy and set guardrails that prevent it.
For example, if you run a broad targeted ad set without any exclusions, the system can and will go after your existing customers to get the easiest conversions.
Consider adding visitor exclusion
Compare your in-platform data to external data sources to get more perspectives on how your ads might be performing.
Remember that none of these are actually a source of truth.
Study the available data breakdowns.
Look for shifts in age or gender budget allocations for different ads to better understand why/how the system is shifting delivery based on relevance and performance.
Pay attention to the placement breakdown in particular as it can help you better understand where your ads are being delivered, which can help you try to unpack what about the ad is working and why.
Keep feeding the system great ads and keep iterating and improving them.
Be thoughtful about the ads you’re putting in the system and how different users will respond to them.
If you need an additional ongoing source of performance creative, Adcrate.co is a fantastic creative partner that’s like having an entire additional creative team producing high-performance ad creative in your account without you needing to think much about it. I advise and own a piece of Adcrate, please reply to this email “Adcrate” if you’d like to know more or if you’d like an intro!
Speaking of feeding the system great ads:
If you’re at a brand or an agency that wants to quickly and easily make a bunch of ugly ads using my favorite templates, you need to get my Creative OS Expert Volume.
You’ll be able to use these use these ugly ad templates immediately and get them live in your account in minutes.
Oooooh, ugly ads 🤤
If you want more than just my ugly ads, you can also check out templates from other experts like Dara Denney, Ashvin Melwani, Sarah Levinger, Rabah Rahil, and Nick Shackelford.
Back to the trust thing, let’s finish this up:
TL;DR: Trust the system, but not blindly.
Give the system the best data you can give it and set the proper guardrails for it to operate so it can go to work for you more effectively.
Apply your human brain and experience (or book a call with me to learn from my experience!) to how the system is set up and optimizing.
Always keep these questions in mind:
Based on the rules the system is given, how could it cheat?
If I could cheat to make this look like it’s performing better than it is so I could get more money, how would I do that? What would I do? Would anyone notice? Is that something the system could be doing?
How are the optimization system’s incentives different from my goals?
Is this data accurate or is it just what I was hoping/expecting to see?
I think that’s everything on this topic for now.
Alrighty, welcome to the “deep end” as Colin & Samir call it. If you’ve made it down here, you deserve something. Yes, even if you simply skipped all the way down here.
I want to give a few of you some free stuff.
If you want a chance to get something I offer for free, I need you to reply to this email with two things:
Where did you find me or how do you know me?
What would you want for free? (only choose one!)
45 minute consulting call with me
Make ugly ads stickers
[Something else you want] with “make ugly ads” on it!
You can just reply with “Twitter, Hatt” and that’ll be enough for me. I’ll be giving away at least 1 thing to 1 person who replies to this email who I choose.
I’m also going to be making some improvements to my website soon, huge thanks to my dude Nico at Bottomless for the help (you’ll see it soon). If you have any suggestions/feedback for how to improve my site, please send it my way and I’d love to return the favor somehow!
That’s it for this week! I covered a lot here! I hope you found this valuable! If you have any feedback, don’t hesitate to reply, I’d love to hear from you. I usually respond to every email!
Lastly, I’d appreciate if you could share this newsletter with anyone you think might enjoy it.
Hott regards,
Barry Hott
P.S. I apologize for the very late email. My goal was to send these on Thursday evenings or Friday, but obviously it’s not Thursday or Friday. Did you notice? Do you care about what day/time these emails come? Let me know!
