Can you trust Facebook's AI? 🤖

(Spoiler: it depends...)  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌

Hey, They're pouring concrete next to my house again – which means I'm writing this email from a cafe down the road, sitting in the sunshine. I have a number of different "mobile office" setups now, and when disruption strikes, I'm able to quickly grab what I need and be set up to work – al fresco – in less than 15 minutes. I'm getting so used to all my little mobile devices that sometimes I wonder if I need an office at all – especially since I decided to spring for the Keyboard Folio for my Remarkable 2, which has turned my most used work tool into a God Tier productivity device. (Review coming soon!) Of course, as we come to rely on technology more and more – whether we like it or not – that also raises the stakes in terms of our dependance on all these gadgets and clouds. We're also being asked to cede more control to various AI-flavored or Machine Learning-powered or algorithm-driven "smart" systems – often without being able to pop open the hood or even understand what we are giving away by agreeing to phone-book length terms and conditions to use any little app on the net. Those stakes are especially high with something like Facebook – the pre-eminent social network (resiliently so, despite being declared "over" or "dead" or "cheugy" or whatever every single year for the last decade or more). Facebook was one of the primary instigators of the algo-soup we all swim around in daily, and its ad-platform uses more machine learning than any of the alternatives – far more. And it is increasingly touting how "smart" its automated systems are, desperate to make the increasingly complex ad platform more accessible to the masses. We're not so far away from Facebook having a big green Advertise Me button and letting its AI do the rest. But how smart is Facebook's AI? Which bits of its automation should be trust and which should we deactivate? Where are the machine learning aspects strong and dependable, and when should they be avoided? To put a finer point on it, when should we hand the reins to Facebook, and where should we take command ourselves? I have the answers for you here – a very clear guide on when you can trust Facebook to take control, and when you need to override its suggestions, or you could tank your ads and lose lots of money. ... right after a quick sponsor message. Sponsor This Newsletter Guess what? Sponsoring newsletters, like this one, 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. ​Apply to Be a Sponsor ​ ​ Can You Trust Facebook? Hoo-boy, there's a loaded question. Let me start by saying that I'm a cynical so-and-so and don't particularly trust any company to do much more than maximize profit wherever possible. However, I'm also an opportunistic so-and-so who has no problem dancing with the devil if the tunes are legit. And as long as they keep charging money for things like ice cream profiteroles, I'll keep on hustling too – and my conscience is clear, thank you very much. So, I don't "trust" Facebook in the most basic sense, but I do get how it thinks, and where it pulls the wool over our eyes. After working in Big Tech, you understand exactly how these companies will present information in ways that benefits them, push certain products and features because of various company goals, but also how you can cut through that noise to see the real picture, and make smarter decisions. I guess you could call it a pragmatic approach. I know how to deal with Facebook to get what I want from this Faustian pact – one we all make every time we go online, you could argue. Facebook uses all the buzzy things these days – systems are smart and automated, machine learning and personalization deliver experiences that are tailored and which improve over time, and AI is advanced enough to make lots of advertising decisions for you – sometimes hundreds of micro–decisions in the time it takes you to yawn and reach for your coffee. But here's the thing: not all of Facebook's systems are so smart in practice. Some aspects of this automated ad machine work better than others – with parts which arejust more refined simply because they have been around longer; others are new… and not dialed in yet. Facebook's systems weren't really designed for the weird products we routinely flog. Ask any general marketer who has moved into publishing: books are strange things to sell online, oddball items shoehorned into the binary world of digital advertising. That said, parts of the system work like a dream, or do so once you have made some tweaks to the basic operating parameters. But good luck figuring out which is which from scratch – it takes a fair deal of research, and even more trial-and-error (aka cash), to know when to trust Facebook... and when to channel your inner artisan and go hands–on with your ads. These are the primary areas where Facebook uses some form of AI, machine learning, automation, or some other kind of smart system: 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 There's a lot to get through so we'll split this list over this week and next. Anyway, these are the order you typically encounter them in the ad creation process, although that might not be the order you actually experience them yourself, as Facebook can be quite proactive about suggesting many of these – especially the new (or newly rebranded) Advantage stuff, arguably the weakest of the lot. Course sellers often love all this stuff because it makes their job easier – the more hands-off guys often tell you just to give the reins to Facebook in almost every instance, even though in many cases it's much more profitable to control certain aspects manually. Your interests are different to theirs – and to Facebook's – so let's run through all of them and make some clear suggestions about what to skip, what to tweak, and what to accept wholesale. 1. Dashboard Metrics This isn't an AI system per se, but so much feeds back from the metrics that it's an essential first step that you set these up correctly – that you customize your dashboard so that Facebook is giving you the real picture of your performance, and not the fuzzy, fudged version that is the default. For example, Facebook might be optimizing for clicks, and you might be looking at the cost of those clicks in your dashboard (CPC), but this is a misleading metric which includes all sorts of clicks other than link clicks – e.g. liking, commenting, and sharing your ad, or simply clicking "Read More" if your ad text runs long. I recommend customizing your dashboard so that you get a much more accurate picture than the one Facebook wants to show you – with all sorts of superior metrics like Unique Outbound CPC, which will remove all the extraneous clicks from your calculations. Which is essential to know if your ads are really working or not, and if Facebook's automation is leading you to profit… or loss. My guide to customizing your dashboard so the important metrics are front–and–center is here. 2. Campaign Goal Right at the start of the ad creation process, when you choose your Campaign Goal (i.e., Sales, Leads, Engagement, App Promotion, Traffic, or Awareness), you are making a critical decision which will tell Facebook's AI how to serve your ads. If you choose Engagement, Facebook will look for people who like to engage – like, comment, and share – with things they see on Facebook. Note that this population of people can be quite different to those who have a history of happily clicking on ads. This is why it's important to choose Traffic – you are telling Facebook's AI that you want it to seek out the most likely clickers in your target audience. You can't tell it to focus on the people most likely to buy things, as you aren't generally sending readers to your own site – you are sending them somewhere like Amazon – so choose Traffic. Just keep in mind that you have essentially given Facebook's AI a Standing Order now to Seek Out Likely Clickers – and everything you do from this point onwards will be influenced by that Standing Order. You will see how this plays out in a moment. 3. Advantage Campaign Budget You can skip this. It doesn't really matter if you switch it on or not – Facebook backed down from making this compulsory after a strong reaction from advertisers when it was first proposed a few years ago, but by then the workarounds to render this moot were already widespread. If you have more than one ad set in your campaign, and this is switched on, that grants Facebook's system the power to analyze your performance and dynamically move your budget around between ad sets. I tested this a lot and didn't really like how it operated – it tended to make snap decisions, and never really revised them over time, in my experience. As a result, I prefer handling budget manually and making budgetary decisions myself. But it's all moot because these days I only tend to have one ad set in one campaign – for organizational ease and to maintain total control over serving and budget at all times. And that's what I recommend you do also, unless you know what you are doing and have a clear idea of what you aim to achieve – a perma-caveat throughout my Facebook advice BTW. 4. Audience Facebook will attempt to deliver as many results as possible for your budget, depending on your campaign goal. If you choose Traffic like I recommend, then Facebook will attempt to deliver as many link clicks as possible, inside the Audience you target. If you decide to target Nora Roberts, it will seek to find the most likely clickers in that Audience and deliver you as many of them as possible for your budget. You have the choice of defining the audience even further – for example, you might choose to only target women, or only readers above the age of twenty-five, or only people that speak English, or whatever. But these days, Facebook is pretty good at making those micro-targeting decisions when it comes to demographics. You might have perfectly valid reasons not to serve, say, a romance ad to male readers – I'm not arguing against that logic. Just saying that Facebook generally speaking is quite good at making demographic decisions on the fly. For example, I used to target specific age slices – readers aged 56 to 65, then 46 to 55, and so on. Now I just let the system manage it. If Facebook sees that men under forty are not responding to my ad, it will serve it to them less. This is why Facebook works much better now with bigger and broader audiences – but it's a mistake to assume that you can just target everyone in the USA that owns a Kindle and Facebook will do the rest. That might work for some books – i.e., those in the biggest genres with the most commercial appeal – but it's a real risk outside of that. Remember that Facebook can only see clicks, not your sales on Amazon. It doesn't know who clicks your ad but decides the book isn't for them – it just sees the click. It can only optimize for the click. So, you want the right people clicking or the ad will go off the rails when it comes to conversion especially. So… target Kindle owners, and then something else like your genre, author comps, similar TV shows perhaps – something. The system is smart, but it also needs help too. 5. Advantage Detailed Targeting What Facebook's targeting system doesn't need is even more control - absolutely avoid Advantage Detail Targeting as this will let Facebook run wild with the targeting. This is one system that doesn't work as intended - not for niche products like books. The targeting system is dumb enough that it thinks Kobo owners are the same as Kindle owners, or that a thriller is the same as a romance. Those things are superficially similar in one sense, in a certain light; the system isn't nuanced enough to understand this could completely waste your money. Five More Next Week! That's all we can fit today; you get another five on Friday when we cover the really juicy stuff: like placements, dynamic ads, and split-testing. Dave P.S. Writing music this week is Duff Thompson with Sleight of Hand. Broomfield Business Park, Malahide, Co. Dublin K36 AT04 ​Unsubscribe | Update your profile​