Hey, Sometimes I like to shake up my routine. Wait, that's not true! Let's try again... Every so often life shakes up my routine, and I just decide to roll with it. During my last bout with the plague, I was stuck in Ireland and had no laptop with me - just my trusty Remarkable 2 and the swish, new keyboard case. I had also started walking at half five every morning, which is way too early for my liking. But, ya know, life, lemons, lemonade, et cetera, so I started writing at that ungodly hour instead. And... it's working. I'm getting more daily words down before breakfast than I've writtten in a long while. Now the chain is stretching over two weeks, I really don't want to break it. That's the paradox about habits - so hard to get going, but also so hard to break once you really do get going. It's easy to slip into the habit of checking email and social media first thing in the morning, which is the worst thing if you want to tackle creative work right afterwards. The chances of being pulled into some "urgent" admin or energy-sapping doomscroll is quite high. However, I've found that getting my pages down before I do literally anything else - even before a coffee - kind of tricks my brain. The critical part of my noggin seems to enjoy a lie-in, whereas the creative part is an early riser - and the words doth floweth. This is a routine I've tried before with little success so I'm especially happy to see it working quite randomly this time. Anyway! Today we have the second part of our look at Facebook and AI - where you can trust its systems, and when you should be pulling all the wires out to shut the damn thing down, like my man Dave in 2001: A Space Odyssey. Catch up on Part I here, or dive into Part II below, right after a quick message about sponsoring messages just like this! Sponsor This NewsletterGuess what? Sponsoring newsletters is a great way to reach engaged and targeted audiences. The Decoders newsletter is part of the ConvertKit Sponsor Network. This network connects businesses to audiences of newsletter readers. For example, you can sponsor this newsletter and connect with 17,000 authors. When To Trust Facebook's AIAs a refresher, these are the primary areas where Facebook uses some form of AI, machine learning, automation, or some other kind of smart system. Not an exhaustive list BTW, but the main areas for an author to pay attention to: 1. Dashboard Metrics 2. Campaign Goal 3. Advantage Campaign Budget 4. Audience 5. Advantage Detailed Targeting 6. Advantage Placements 7. Dynamic Ads 8. Split-Testing 9. Ad Serving 10. Performance Suggestions We're doing 6-10 today, but scroll back up if you need a link to 1-5. 6. Advantage PlacementsAn even bigger mistake than anything in Part I - and one that almost everyone makes because Facebook pushes it so hard - is to let Facebook decide where your ads should appear. Do not do this - especially with Traffic campaigns. Facebook has all sorts of Ad Placements, from the News Feed on Facebook and Insta, which you are surely familiar with, to various ad slots running before, during, and after videos on both platforms, to things like its Audience Network - which is a huge array of third-party websites where your ad could potentially appear. I strongly recommend running your ads to News Feed only (I go further personally and do Facebook News Feed only - and I recommend doing that when learning the ropes, and only adding in additional placements later on when you know what you are doing). Facebook will repeatedly pester you about handing it control here - even as your ads are running, trying to tell you that you can reduce costs by activating Automated Placements. I repeat: don't do it. With Traffic campaigns especially, this can be a disaster. Because Facebook deploys so much Machine Learning here, and can make decisions quite aggressively, things can go off the rails in a spectacular way. An example: I had a Traffic campaign with Automated Placements activated, and I spent about $200 on clicks before realizing it wasn't translating into sales at all - I mean at all. This was a Barnes & Noble campaign so the sales data was quite delayed and before I realized something was wrong, I'd burned through way too much cash. When I analyzed the campaign, it was obvious what happened: over 90% of the budget was spent serving ads to the Audience Network of third party-sites - and most of that to mobile devices as well. Know how easy it is to accidentally click an ad on your phone? Yeah, I was getting a lot of that. And Facebook's system couldn't see it wasn't converting - it was just seeing the click, and serving more and more ads to the Audience Network at the expense of News Feed and anything else which might have actually delivered sales. This is one area where the system is simply too smart for its own good and I always always always recommend handling Placements manually. You don't necessarily have to agree with my specific choices of which Placements to use - maybe you think I'm too extreme only choosing News Feed and only choosing Facebook. You might want to add in Instagram, or some other placements like the ads in video. Please feel free to do whatever you like, as long as you test and compare results. And as long as you don't let Facebook choose the Placements. To put it more bluntly, everyone choosing Automatic Placements is subsidizing the ads of those who don't. Yeah, I said that. 7. Dynamic AdsLots of people seem to love dynamic ads and I hate them - so take my views with a pinch of salt, if you like. But here's why I hate them: it's a pretty dumb system which just smooshes combinations of text and images together in a really amateur way. Some people like it because it does the thinking for them and figures out which combinations work best for their audience. I hate them because if you put in a modicum of effort, you can do a far better job yourself - and waste far less money figuring out what works. Try it if you like, but don't say you weren't warned! I really do think that effort is much better placed in learning how to make great ads manually. 8. A/B TestingDuring ad creation, Facebook often suggests running an A/B Test - also known as a "split test" - i.e. where you test variations of your ad against each other. For example, you might want to test alternate ad images, text, or targeting and Facebook has a formalized way of doing that - which will even pick the winner for you and push all the remaining budget on the ad it sees getting a better response. I don't hate this feature, I think it's... fine. The decisions it makes are mostly good, but, as per usual, sometimes it makes that call a little early for my tastes - and doesn't revisit it later on. I prefer running these split tests manually, and setting my own parameters, and when you do this you access all that same machine learning and algo-powered goodness - just with more granular control. I'm starting to think I might have control issues... 9. Ad ServingIf you put two different ads in the same ad set, then Facebook's machine learning will show them both to your target audience and then quickly start favoring one of them - the one getting more clicks, if you chose Traffic as your campaign goal. Of course, if you choose something like Engagement, it will favour the one getting more likes, clicks, and shares. And if you choose something like Conversions - which Facebook can't see with an ad pointing at Amazon, well, who knows what will happen. Again, Facebook makes these decisions a little too hastily for me - I'm old-school and like to see at least 1,000 impressions on an ad before deciding if it's a dud. Facebook's system seems comfortable making that call after just a few hundred. But as I'm not using Facebook's formal split-testing feature, I have the power to override that manually, and I can switch on and off ads to force one to serve over the other. This is handy when I'm doing something like a launch campaign where I'm going to advertise more aggressively. I'll put two ad variations in the same ad set - often a square ad image and a letter box ad image, with everything else identical. I let Facebook decide which is most effective (usually the square ad wins, but not always!). And then when that winning ad eventually starts to tire, Facebook can start serving the fresher "losing" ad in the ad set. And if it doesn't... well I can override that manually when I see the Frequency getting too high. So, in summary, I trust Facebook's system to make good decisions with this, but will periodically monitor performance and intervene manually when required. 10. Performance SuggestionsOn the other hand, I have little faith in Facebook's performance suggestions - often emailed to you directly, or are the subject of annoying prompts and notifications, or appear in various places on your dashboard, sometimes dramatically. They usually say something like "Your cost per click could be reduced by 28% if you turn on Advantage Placements." Or perhaps something like "Combine these ad sets and reduce your cost-per-action by 11%." Or maybe the tantalizing offer to "Expand this audience by turning on Advantage Targeting and reach 6m new customers." I find these suggestions are mostly useless and occasionally disastrous so I just ignore them completely - and recommend you do likewise. Learn to analyze performance yourself. Learn to write better ads yourself. Learn to get hands on with your targeting when necessary, so you know how to reach the right readers. Learn those things and you'll be good with Facebook Ads. Hand over control to the AI, and you might spin the wheel and land on the right place… or you might not. And you won't have learned very much either. Dave P.S. Writing music this week is Richard Bruckner with Tom Merritt. |
Friday, September 8, 2023
5 more ways Facebook is using AI on your ads 🤖
Subscribe to:
Post Comments (Atom)
![[THE LATEST] Inside hell's kitchen 👩🏽🍳](https://lh3.googleusercontent.com/blogger_img_proxy/AEn0k_uZSPj8I_GSBlgBYfDgJ1PxjhTLhO0Fdn8ek5X0BPF66YFk8-oBnP_REWmlsbzUQdOQpLqRMrCcXIjXtTNWGwjWphLcgxYZHyMgoaDKam4biHxBx3i22mmbyRZtHUp7rJFyBJ-bpQl2wXgn_76TLtp5HlHuX20VWMe-jPABv_eNUrUmRQLn9kvnflkVSl6QxFOb1VYNWdiOFZq-V9j0lkcfnhtcApreV9_sAV81zHhCy1lwkXaiW_IZM86gJ1oGv1-dVZupPAZyli0E4fv5sSI=w72-h72-p-k-no-nu)
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.