I suppose it helps disambiguate that from the content around which ads appear. Beyond that, I suppose language is no less subject to fads and trends than any other realm of human endeavor.
Referring to ad content as creative has been around since at least the days of Madison avenue ad agencies in the 1950s. I wouldn't be surprised to learn that it's usage arose concurrently with advertising.
Thank you for asking. I am sure the term is as old as the industry, but in every domain I am irked by the use of adjectives, specifically, as nouns. This article is a particularly irritating example.
My earliest memory of struggling with this kind of language was in high school economics class. I was forever put off from the material by the confusing truncation of noun-phrases into adjectives.
And it's a shame, because there's a lot to discuss here: vast swathes of the technology industry hinge on the subjects discussed. Weinberg and Lombardo seem to be suggesting that an emperor has no clothes.
I think the authors are overlooking that a Disney movie is 1.5 hrs long and has the opportunity to cater to multiple audiences (eg, insert a joke that parents get but flies over the kids’ head), whereas an ad is viewed for a split second. Also, there is certainly specialization in content (BET vs MTV vs Disney Channel etc). The key to advertising is in going broad with the audiences to reach, but specific in the message for each segment.
(I’ve worked for one of the top TV and cinema content companies in the world, and currently working for one of the top DSPs (advertising)).
If you’re including Marvel in the scope of Disney movies, you might need to bump that runtime up to 120 minutes minimum, and all the way up to 150 minutes in many cases.
There was a post a little while back arguing that Avengers isn’t even a kids franchise anyway, so your observation still stands.
Disney targets everyone because they have 1 product they can show to everyone and it isn't technically possible to customize a movie per interest. If personalization didn't work then advertisers wouldn't see ROI on their ad spend on FB/IG/Google. Clearly the fact that these companies make hundreds of billions says otherwise. It doesn't need to be perfect, just good enough.
A more apt point would be that personalization doesn't scale. If you need to target a hyper specific audience, then you by definition, will have scaling problems.
Now if you need to target 100k hyper specific audiences, then it's a different story.
Does it really make more than creating ads and targeting them appropriately at the right interest groups (i.e. advertising footballs on a sport fan site)? Aside requiring snooping to be effective in the first place.
Are there good studies on this to validate it?
Companies spend a mint on this stuff because they are told it works and is necessary, I'm not sure that's true, especially when that comes from the sales/marketing arm of those ad companies...
If creating interest group ads was more effective and cheaper, you'd see it used more by marketers. Marketers might not be able to measure whether targeted ads are effective but they can definitely compare the ROI between interest group ads and targeted ads.
This just makes me wonder if it would be possible for a modern streaming platform to make movies that personalize to a user. There's no fundamental reason you couldn't shoot a movie that plays different versions of a scene to different users.
Games generally don't let you make narrative decisions though - it is still rare and troubled that this implementation is attempted and executed. When they do, those decisions are still one of the more complex pieces of media humans ever create: an interaction of interpretation between the writers, artists and player interpretation as well as the technical limitations of the system.
Story branch decisions in gaming are to some extent just an evaluation of what the writer's think of your world view - at least when it comes to consequences. Arguably they can only really be constructed as a trial between two opposing narrative voices - resolve questions of character via the wisdom of the crowd.
Sometimes I think Netflix is basically doing this, in a fashion, with their originals. Make 20 Ryan Reynolds movies, suggest them dynamically to different audiences...
I think different versions of the same movie would completely break word of mouth, though.
I had a long discussion on how ML/AI could be used in games to provide more personalized and realistic interactions. Which was countered by a convincing argument that this is not what players and the industry want. And how largely games have been focused on tightly controlled scripted gameplay that feels a lot like a movie. While an AI powered game would feel unfocused and likely nonsensical without any director oversight for the story.
Similarly for a movie, perhaps we could personalize them, but would the result actually be desirable.
Isn't the point of a movie to provide studios with outsized economic rewards? "Shared cultural experiences" are just one way of economizing. Procedural generation might be another. Perhaps we'll see Web 4.0, ie the "procedural web," obliterate feature-length video content just as it's creeping over text and short videos.
For me it is, I would find it deeply depressing to watch movies and be unable to discuss the experience with others because they got different characters and plot lines.
Do people really talk about movies? It seems you do. Do you talk with your friends? Online? On some specific forum? With your significant other?
I'm so surprised that there are memes from big hits, but no real talk about them.
I mean there's relatively a lot of talk about how Harry Potter feels good but doesn't make sense, and of course lots of critiques, reviews, "reaction videos", but somehow it seems actual people stopped talking about movies with each other (beyond the very shallow have you seen it? yes/no. was it good? yes/no.)
maybe I just miss when people talked too much about the Matrix? :D
also I find it strange that it's still so hard to find good movies/books/series/games that I would like.... despite all the data, metadata, taxonomy, folksonomy (tags! unstructured data! big data! bad data, bad!)
for example, IMDB knows what I like, Netflix too, yet the recommendations are pretty bad.
let's take Mindhunter. "CSI done well", right? but I still haven't found something that is similar to it. (not that I looked very much.)
To the request to name a single brand based on personalization, music brands like Pandora and Spotify are built on personalization:
Similarly, clothing brands like stitch-fix and trunk club are personalization specific.
Regarding Disney, we just signed up for Disney+ and the first step was to decide if you wanted mature content and the second was to decide on your avatar. I think Disney+ will allow the company to cater to personal tastes more than they ever could through theaters and we will see that in the future.
Finally, for the longest time marketers have looked at demographic groups to target with their products, this is a weaker form of personalization. Obviously that won’t be abandoned, so I wonder why the author thinks companies should only target mass markets?
While it is true that Spotify does recommendations, I'm sceptical that the majority of users value this part most. I'm happy to be corrected, but I'd guess it is simply the ability to stream music on demand that remains its main selling point. Put another way, the value of Spotify is its expansive collection, not its personalisation.
I know that some people find value in this, but for me, I wish I could turn off the recommendation engines on Netflix, Amazon, Spotify, and wherever else they are in services that I use. Whatever notions of similarity that they use to generate recommendations have nothing to do with how I judge things, and for me this is just noise that makes those services slower to navigate.
I feel Spotify's recommendation engine works kinda OK since music is many small songs rather than long and few movies.
I mean if you are into a movie there might be only three ones like it which you allready have watched and surely are not on Netflix.
However, as soon as you use your Spotify account for kids or parties the engine gets messed up forever. It is a shame since it helped me find alot of artists.
I assume, perhaps incorrectly, that everyone has the same catalog (at least for my fairly boring taste in music) and actively choose Spotify’s personalization over Apple Music’s tighter integration with my phone, car, etc. Just a single data point though.
"No" seems to suggest that you're disagreeing with your parent's post, but your parent is saying why they chose it, not why you did:
> I assume, perhaps incorrectly, that everyone has the same catalog (at least for my fairly boring taste in music) and actively choose Spotify’s personalization over Apple Music’s tighter integration with my phone, car, etc. Just a single data point though.
>but your parent is saying why they chose it, not why you did:
And I am saying why I chose it. You don't need to read too much into the "no", nor have to find some perfect formal consistency in a quick response. It's just "no" as in, "no, that's not my case".
Anecdotally, I love the personalized station Apple Music generates for me. I have pretty eclectic, cross-genre taste and the quality of the personalized station - the mix of stuff I already love and the new stuff that I don't know yet but that I end up loving - is by far the stickiest part of the service for me. It also seems to be improving over time? I know, anecdotal, but that has been my experience..
> While it is true that Spotify does recommendations, I'm sceptical that the majority of users value this part most
Any time Spotify is brought up (which is really any time music is brought up) it's a good bet the quality of discovery of recommendations are mentioned. The Discovery Weekly and Daily Mixes are loved. My anecdotal evidence is the dozens of people I've talked to about this over the years. This HN thread is the first time I've seen anyone suggest otherwise.
I know for myself, if it wasn't for the discovery that Spotify offers, I wouldn't use it.
There AI is amazing. So I make sure to tell my Douyin (Chinese TikTok)watching other to always also spend time watching totally random things on the platform.
Yeah anyone who doesn't believe that Disney is doing personalisation should get in contact with me. I have a bridge to sell them in the Nullarbor.
In fact, they're actively building it and selling it to advertisers. Disney+ and similar platforms are the tools that media companies have every intention of using to drive the accuracy numbers much higher.
E: I should note that I work for Snowflake, we're marketing this capability pretty hard. I'm not directly involved though.
You would be the first person I've heard say that. Everyone I ever spoken to about this loves Spotify's recommendations. And I'm talking about non-technical people. They are excellent recommendations. No, they are NOT successful for the reasons you state. Apple Music does that. Now, that's a recommendation that is indeed bad.
You can add a second to your list. Spotify doesn't have a lot of my favorite music, and the recommendations have always been bland and boring or downright bad to me.
Add a 3rd. Stopped using Spotify because the recs downright annoying.
I prefer to use YT Music s it seems thoroughly confused by having each person in family use from my service account to the point the suggestions are general in nature. They still do not get played though. I refuse to click on suggestions in any product as that reinforces their data on me/us. I’m quite ok if any music engine thinks I’m a polyglot toddler with penchant for death metal, Thai ballads, and ancient Chinese orchestral
Music. Such a profile makes me soooo much less likely to get other music or junk marketed to me ;-)
Fourth, I've never found a recommendation engine I like. Maybe I have weird taste in music, but Spotify only ever recommended one song that I ended up loving (oddly, the very first one it ever did).
At this point adding numbers to the people agreeing seems pointless, but I'm also in this cat3egory.
Sure, I've had some good recommendations, but I'd expect any recommendation engine to do that just by chance.
What I can say is that Spotify has never proposed something truly new and interesting to me. If I listen to a lot of metal, it'll recommend me the most bland stuff in the same genre. What's even worse is that it keeps playing the same songs over and over again. It's like the recommendation engine just gives up and starts repeating its suggestions again.
Your last paragraph is why I got rid of Spotify. I like music to move me. It was amazing how I could start a run of upbeat, pumping, specifically chosen songs and then have the recommendation engine suggest 30 minutes of cruisy blandness..
I simply gave up on it. Why have access to everything when I only got the crap. Going back to my personal collection increased my good/bad song ratio considerably.
That's fine. It doesn't change my opinion, nor my sentiment. If Spotify's recommendation engine wasn't as good as it was, it would not have the users it has, and would probably have been eaten up by now. Nothing anyone has said has shaken that belief, nor offered up anything worthwhile.
>If Spotify's recommendation engine wasn't as good as it was, it would not have the users it has
Now that's just confirmation bias + circular logic.
A streaming service can have users regardless of how good its recommendation engine is. YouTube has crappy recommendation (and has had worse for most of the time it existed) but tons of users.
For Pandora and Spotify, merely offering a convenient way to stream music, a free tier, and a big catalog, was enough.
You seem convinced of some bizarro idea that a media/streaming service can only succeed based on its recommendation engine.
I've found Spotify's recommendations to be brilliant. Easily the best music-related thing I've been introduced to in 10-20 years. I'll often open Discover Weekly and end up favouriting more than half of the suggestions and only not favouriting the rest because I'm feeling pickier given the success rate. It's helped me discover loads of new artists and tracks.
Apple Music (well, iTunes at the time) was pretty good. But that was in the era where you bought music, so they had a full profile of what you liked. More importantly, with your iPod, they knew what you actually played, no matter how obscure. Now, with streaming, this is much harder problem because the only signals you get are likes and play time, but this signal is a lot more noisy than music you would spend real money on.
It may vary well work best by genre, or not cater to people who have gone through spaghettification (insist on subdividing genres infinitely) with their music.
For example, take the "Phonk" genre, which I only was introduced to due to the Ukrainian war. To mean these recommendations give me a lot of new music to listen to. But then you have people lamenting that "Phonk" has really been overtaken by "Drift Phonk".
https://www.youtube.com/watch?v=UAV7hnCB_ZE&ab_channel=yokai
On the other hand, I personally enjoy albums and songs major artist after they achieve critical success, when they establish their signature sound, like David Bowie(1970-1983), Stevie Wonder (1972-1976), or Peter Gabriel(1986-1992), but if I "like" any of these songs on Spotify, it means I get their entire catalog mixed in my daily mixes, and if they ever release a live album, those tracks show up. This is not what I want.
Music genres change, that's nothing new at all. Just look at the type of trance from the early 2000s and compare it to today's trance. It's a completely different sound. Would the author also complain about "progressive trance" taking over "trance"? It's natural that some sub genres might become more popular, while others lose listeners relative to the newcomers.
Same happened in techno, rap etc.
While Spotify might accelerate this process through a positive feedback loop, this video is just another form of gatekeeping and saying "I knew XYZ before it was cool".
>No, they are NOT successful for the reasons you state. Apple Music does that.
Which is neither here nor there. Apple music came later, and Spotify already had a headstart, a good UI, a good selection, and a good free plan.
Spotify, Pandora caught on because they were the first good streaming solutions, at the time bandwidth, mobile phones, etc, were in place and ripe for streaming. Not because of their recommendations...
Last.fm's recommendations were better even 15 years ago. I was spoiled by them, and that's the reason I don't pay for or use Spotify despite wanting to like it. The recommendations are bad compared to other options.
Spotify would not stop recommending me the same NFT avatar trash even though I did ever negative signal I could do it, reported the artist, whatever, no matter what I had no choice in a third of my playlists containing this ninny.
Canceled and would not consider a recommendation based service that does not let me explicitly remove something from being recommended.
Spotify's recommendations have been my bread-and-butter for music discovery for a long while. The radio feature and the Discovery Weekly playlists are very tailor made for my taste (which is broad and pretty underground in general), which probably has to do with the amount of data I generated by being an user since 2010.
Disclaimer: I work for Spotify nowadays but have been an avid user for more than a decade before joining, my opinion is based on my personal experience with the service and not as an employee.
My gut is that I'm looking at two sides arguing past each other.
Most major value creation happens well before Disney+ or Pandora. They just have somewhat broadly defined parameters that will get you more of the same, from what has been created.
As easy examples, there are a lot of good kids shows that adults could enjoy, but my guess is that they are not recommended to most adults. Just as really good folk music is likely to get suggested to someone that hasn't listened to folk music. Or music/movies from another nation.
Which is all too say that the personalization is ultimately a customer fitting themselves to what they want, from what they know. Dropping someone in fresh with no prefit is probably a lost cause. Even though that is how most personalisation talks brand themselves.
The post is about marketing and ad targeting. For Spotify to satisfy their question of a "brand built through personalization", Spotify would have had to do something like sign up for google/facebook ads, target the demographic of "music listeners" and convert enough users to end up with a successful business. That's not how spotify built their business.
You're talking about their recommendation engine for users they already have, which is completely different.
We also have to take into account that there's a lot of "misses" in those recommendation engines - which is normal. But when they nail it, people have positive emotions towards it, so it's a good feedback loop.
Personalization would be to have Spotify recommend you only songs about expensive watches after you've searched for a Rolex on Google, with ads in the middle about Rolex ahaha.
Personalization would be to have Spotify give you ads for a Rolex right after you purchased a Rolex. At least the way that kind of thing usually works for me is that I always get the recommendations after the purchase (definitely not a Rolex for me).
I agree Spotify is definitely built on personalization. The only time I subscribe to premium is when they offer me 3 months for $9.99, and they know enough to offer it to me in May when I start using premium for running or cycling. Then I cancel it in August, and they offer it to me again soon after where I subscribe again from October - December. Then it goes dark again until May hits and I'm offered 3 months for $9.99 again.
The problem is, everybody hates all the ways in which spotify doesn't act sensibly and predictably. Just google for "spotify random broken" to spiral down a depressing rabbithole of a decade of forum posts about a feature so simple and still not implemented, that it's clearly being done on purpose; either to drive engagement or save on money paid out to labels and artists, your guess is as good as mine.
"To the request to name a single brand based on personalisation, music brands like Pandora and Spotify are built on personalisation."
As I read this article, this meaning of "personalisation" is not the one used by the authors.
Tech companies like Spotify and Pandora are just intermediaries. They may help to "deliver" commercial product, or advertising, however they do not create the product referred to in this article. The article uses the term "personalised creative" to refer to product. Tech companies operate as middlemen and produce no content. They are dependent on others to produce it. This is the bait for computer users. Tech companies sell advertising services to companies that produce content.
Tech companies gather data about individual computer users, e.g., web browsing histories. This is "personal" data. Tech companies allege this makes it especially effective and therefore valuable. The studies cited in the article suggest this is claim is false.
Disney produces content. There is no shortage of personal data being collected and sold by tech companies which is available to Disney. The article highlights that, despite the availability of personal data collected by tech companies, Disney generally does not produce "personalised creative".
In sum, the article is not about what "tech" companies do, it is about what content producers do. More specifically, it is about whether, based on available research, producers of creative should or should not attempt to use personal data about computer users collected by third parties in order to produce "personalised creative".
> Tech companies like Spotify and Pandora are just intermediaries. They may help to "deliver" commercial product, or advertising, however they do not create the product.
They do create a "product", the delivery system. This is a creative product as much as any of the music. A lot of software was written to allow this and all of that is the "product".
We know marketing works, under specific constraints. Sophisticated marketing does A/B testing to confirm the efficacy. Data collection is part of it. This article is making sweeping inaccurate statements, muddling the long-standing debates about marketing methodology.
> Personalisation assumes that marketers have perfect data on every individual customer.
No individual believes this, so it's hard to believe an extrapolation into organizational commitment.
> Marketers infer who customers are based on their browsing behavior.
Not in isolation. This is mischaracterization.
> So, how accurate is gender targeting? It’s accurate 42.3% of the time.
Probably not. There is no access to the datasets (https://pubsonline.informs.org/doi/epdf/10.1287/mksc.2019.11...).
"Gender accuracy ranges from25.7% to 62.7% with an overall average of 42.3%" without explaining how "no data" is handled. Having the actual data 42% of the time and not having it 58% of the time isn't the same as measuring how accurate it is. This looks like using bad statistics paired with confirmation bias.
On and on and on.
This is an article built on a quick summary of a questionable paper to push some anti-marketing prattle. Not to say that DSPs are reliable or that everyone is handling legitimate data, but bad data is filtered out pretty quickly in AdTech where there are means-tested costs...although it sometimes takes time to present.
Personalisation as a concept needs to die. It’s sold as a shiny cutting-edge thing and sounds like it should be a good idea, but no one seems to realise that beyond the hype the tech is surprisingly shitty, like Web3 or voice apps. And the ones that do know aren’t going to let on because they’re the ones trying to sell personalisation software.
Personalised ads result in vast amounts of corporate surveillance and still don’t work. I’ve never understood why contextual ads aren’t seen as good enough. If I’m looking at reviews for baby products, for example, then probably going to be interested in ads for similar baby products.
And with personalised content or product recommendations, very few companies have the breadth and depth of content that even begin to make it worthwhile, and even those that do can’t seem to do more than show you more of what you’ve encountered before. YouTube and Instagram’s personalisation algos can be downright annoying ‘You watched that one video randomly? Okay, here’s hundreds more like it, crowding out all the things you do watch regularly.’
Good search and discovery tools are always better than a stupid recommendation engine, and they’re all stupid.
Personalisation is effective. I've tested enough personalisation campaigns to be convinced.
And as a counter argument to Youtube/Instas algos. The reason TikTok has become so popular is that it simply better at personalising content, than say instagram reels.
Does this mean all the billions spent on targeted ads, all the gigatons of CO2 emitted running that crap JS in browsers worldwide, all the people annoyed, all the hours wasted, all the malware pumped and all the privacy violated, ... it was all for nothing? Suck it, marketers.
Yeah sure. Regarding advertising, i'm sure the clothing labels that target young female demographics on instagram would thrive by changing their branding to target all genders of all ages.
The argument is if you specialize in young women's clothing, instead of trying to micro target messaging to different personas, have a single message that can work with everyone you are trying to sell to.
Suppose it's my job to market Viagra and I'm paying for each impression. Doesn't it make sense that I would want to avoid paying for a 25 year old to see the ad? Clearly my customer base is not 25. If I'm selling bras, I don't want to pay for a male to see the ad. The desire for relatively low-level personalization is so obvious I don't see how the author can ignore it, accuracy of the data notwithstanding.
> Recently, Professor Nico Neumann partnered with the brilliant marketing team at HP to replicate this research for B2B. The results were unsurprising – but horrifying. Many enterprise technology companies spend millions of dollars ‘hyper-targeting’ IT decision makers (ITDMs) using third-party data. But if we get gender wrong more often than 50% of the time, what percentage of ITDMs do you think are actually ITDMs, according to the research?
> Do you want to guess? It’s 14.3%. And for ‘senior ITDMs’, that number drops to 7.5%.
> Super impressive! That’s about as precise as… a drunk monkey throwing darts?
That... seems... like a great result? How many ITDMs are there in a random sample of 10,000 people? According to the Bureau of Labor Statistics [0], there are 715,000 people working as IT managers in the USA. So the incidence in a random population of 10,000 is 715,000/300,000,000 = 24 people. Seems like the ads are working really well.
I think it's interesting to see when you get a targeted ad you benefit from. I attended HackMIT for the first time after seeing a Facebook ad for it--I had an amazing experience at the event and that ad is possibly the reason I am a software engineer today. I doubt I would have received that ad if targeting weren't possible.
If I had to put my imagination to work where ads and humans can coexist peacefully, I could imagine a world where ads are required to be hyper targeted, and you would be shown no ads unless the ad was specifically targeted to you. If the ad wasn't relevant to you, you could click an x on it to make it disappear, and the advertiser would have to pay a penalty to the host of the ad (google, facebook, whatever) for inconveniencing their users. I suppose marketers at Disney and P&G would suffer, but I am really sick of seeing liberty mutual ads when I don't even have a car.
I’m pretty ignorant of the marketing industry, but it seems like maybe you’re lumping demographic targeting in with personalization - broad categories like age, geographic location and gender are probably good ways to target advertising.
Hyper-individualized ads are just plain creepy. I’m glad to hear they’re generally ineffective at reaching their intended audience, and I’d hope that when they are effective, that the sense of invasion people feel causes them to turn the other way.
That's targeting, not personalization. Personalization is that it shows you a different ad based on who they think you are. eg: Show a "Buy Viagra for yourself" ad for men and "Buy Viagra for your spouse" ad for women.
The article is arguing that instead, you should just craft a "Buy Viagra for you or your spouse because of these broad based benefits of Viagra" ad that is for everyone.
Expanding on that a bit more, the article isn’t advocating that marketers senselessly put up Viagra ads everywhere, but instead to “pivot to more contextually relevant and attentive channels, while working with partners who had permissioned, first-party relationships with [target customers]”.
Which in this case would be something like putting ads up on a website or magazine that’s know to target an older male demographic. Or getting an older men’s health influencer on Instagram/TikTok to make a sponsored post. So still directed marketing, just not personalize on a user by user basis based on data from 3rd party brokers.
I’m puzzled: The authors say it’s impossible to accurately guess who the consumer is, and that the idea that you can show them the ‘right’ ad and that this quasi magical targeting will make broad-reach, low-impact ads (also called banner ads) work is BS; and people here on HN go on arguing about Spotify’s recommendation engine and if it’s any good or not.
The difference you're alluding to between music targeting and ad targeting isn't obvious. Can you explain more? By default I would think they're pretty similar. Trying to find a unit that will resonate with you specifically. And if they're similar then it's entirely fair to talk about both as different angles on the same problem.
Unless you're maybe saying "the problem isn't the targeting, it's the type of ad"? But I don't see anything in the article that makes it specific to banners.
> Proponents believe we are entering a new era in marketing, where every creative message will be tailored to the specific needs of individual buyers.
They are talking about ads, not about which songs are suggested to us.
> Most personalisation efforts are powered by third-party data. Marketers infer who customers are based on their browsing behavior.
In order to deliver personalised ads, they use cookies — third-party cookies, by companies that invade our privacy and track our every move.
Spotify does not rely exclusively on poor quality third-party data: they also have first-party data about us: the songs we listen to, like etc; how often we listen to them; if we put them in our playlists etc. This is one reason why their guesses as to what we may like are better.
Another reason is that they are ‘only’ trying to suggest another song to a person who, apparently, likes music; they are not trying to guess my gender, age, job, wealth, type of person I am, mood, interests etc and find a match between these tracts and the fact that I may be interested in… an ad I never requested from a company I don’t know…
> “We determined there was simply too much waste in the old model of activating this third-party cookie-based data across high-reach, low-impact placements.
My guess is that the type of ads they are talking about here are general rotation banner ads, i.e. garbage worth nothing which gets sold for good money under the (false) promise that companies have ways to show your ads to the ‘right’ people. Problem is, they don’t.
> They are talking about ads, not about which songs are suggested to us.
It has a big list of media and it's trying to pick one that will be effective on you. That part is quite similar.
> Spotify dies not rely exclusively on poor quality third-party data: they also have first-party data about us
Is that why they're saying it doesn't work? Or is it a different reason? Because I see people citing all sorts of possible reasons and intuition can only go so far in knowing which ones are important.
> My guess is that the type of ads they are talking about here are general rotation banner ads
Maybe. But in that case how much is the data the problem and how much is the type of ad the problem?
Hi, let me try to answer your last point, which imho is the most important.
The problem is the type of ad, which is worth nothing. The data is just BS used in order to dress up the selling of a worthless ad placement for good money; and yes, that data is also largely flawed.
And even if it were not flawed, ‘guessing’ that someone who is supposed to be a certain age, gender, in a certain job, in a certain relationship, with or without kids, etc could be interested in a certain product is much harder than simply guessing that if you like these 10 or 100 songs, you may like this other song.
And how is anybody going to ‘prove’ that all this quasi-magical targeting worked, when click rates are ridiculously low even for highly targeted ads?
Imho targeting and ‘personalisation’ serve just one simple purpose: convince the marketer that he/she is doing the right thing, or what is considered right or best practice at the moment, and so that they are safe and are not going to be fired because, you know, nobody ever got fired for choosing IBM.
> Is that why they're saying it doesn't work? Or is it a different reason? Because I see people citing all sorts of possible reasons and intuition can only go so far in knowing which ones are important.
The Spotify data about like/dislike is rather accurate (like 80-100%), as long as someone doesn't click under substance influence.
Third party data (according to article) has accuracy around 4%-44%.
The stats in this don't pass the smell test and seem made-up. For example gender targeting at 42%? That means they somehow did 8% worse than random chance.
Their data somehow resulted in a statistically significant anti-correlation. This is a pretty unlikely result from a stats POV.
> For example gender targeting at 42%? That means they somehow did 8% worse than random chance.
This is a lot more plausible than you might think. Simply build a machine-learning model that is accurate 58% of the time (maybe because it takes the "gender" field people manually entered on their online profiles and adds noise) and accidentally invert its output.
You are badly confusing sex and gender (and wrong about sex, too, but that's not my job to educate you past your ignorance). Advertisers do care about the gender differences between "people who menstruate", "people who have facial hair", "people who are post-menopausal", "people who like masculine aesthetic products", "people who like feminine aesthetic products", "people who would prefer not to answer any such questions on a survey" and several more. They overlap more than a lot of people would like (there are "people who menstruate" that are also "people who have facial hair", etc), but that doesn't mean that they aren't "real" genders that real people experience. Advertisers continue to mistake sex for gender themselves, often thinking that sex plus age range equals gender. It is a first approximation and it sort of accounts for the distinctions between "people who like feminine aesthetic products" who are also "people who menstruate" versus the ones that are "people who are post-menopausal", but they are wrong about that too and it is still only a bad first approximation.
Those are all real categories (the word gender is just one of several English borrowed from other languages for the word "category", along with genre), whether or not you believe people should be allowed to choose the category they best identify with and that category should be distinct from assigned sex at birth. It's all real, whether you choose to believe in it or not.
I mostly agree, most of the companies try to personalize marketing on bad data, but I think the main power for personalization is not acquisition but retention. In retention, you already have a lot of data bout the customer you collected yourself.
One of the most interesting usecases of my company (videobolt.net) is personalized videos at scale (via API or one time video creation on a bunch of data). We powered some very interesting campaigns this way.
Since this comment seems popular, this is an example of a campaign we powered: https://www.youtube.com/watch?v=7ufTiBRMnxo (we built ~750k personalized videos for this one, all the data in the video is user specific). If you want to discuss something like this, drop an email at aleksandar@videobolt.net (or just chat here).
Rendering some stats you have about your users in a "video" (could just as-well be a PowerPoint presentation turned into a GIF) is a pretty weak form of "personalisation".
This perfectly demonstrates the point of this article - it's a canned and mass-produced product.
As far as I understand the article, two main points are that data on which you think you're doing personalization is often inaccurate, and that even if it is accurate you cannot hit the person well enough because that data is not enough.
As said, I pretty much agree on acquisition level, but don't on retention level.
On retention level, data should be good, since you are the one that is collection the data through your app / service usage. You should collect important data and trust your data.
If you are sure you have that data, you may very much provide your customer with interesting personalized marketing experience.
For given example, people use that platform every day, enjoy it, and are proud of their achievements there. They very much liked and shared these videos, as they are telling story of their journey on a platform that is important for them.
PS. I agree this is not for everyone. You should have immersive platform users are spending a lot of time at, with a bunch of interesting data. These would maybe even work on remarketing level, if you collected enough data in your system for given customer.
PPS. And difference between good quality video and PPT / GIF is... just huge.
Personalization work but companies tend to overfit which reduces the benefits of personalization.
Let’s forget the obvious demographics correlation with products and just start with an example.
Simply knowing that someone is active on HN, I wills bet ads on tech product and services would be more effective on them compared to a random person on the internet. The problem comes when the personalization gets too personal and stuck in the past, like how Amazon recommends another of the same product category after you bought one.
As someone who works in that industry my overwhelming feeling after reading that article is that it’s largely junk for a couple of reasons.
Without going super deep into the specifics they do make one good point that I agree with which is using 3rd party data brokers and sources to drive a personalisation strategy is going to give you terrible results precisely because that data is junk to begin with and is probably only going to get worse over time.
That is however probably where the good points in that article end.
In short, they take this ultra narrow view of what personalisation actually is and just straw man it to death.
There is an underlying principle in the world of sales / marketing etc that basically boils down to show the right people the right message at the right time. The idea that this approach doesn’t work is not in anyway backed up by any evidence whatsoever and the second half of the article that just says “make something great for everyone because… movies” is just incredibly stupid advice.
Like yes that is a great baseline to start from and is a goal in and of itself but what about when I start to know a bit more about what you’re actually looking for? Why would I not intentionally try to tweak my approach? A human salesperson who couldn’t adapt their approach would be out of a job within a short amount of time.
The problem ultimately comes down to how can I understand
1. Who is the RIGHT audience
2. What is the RIGHT message for that audience
3. What is the RIGHT time to show that message?
So it might come as no surprise to anyone that if you just outsourced that question to some 3rd party data source and had some incredibly broad and useless audience group like “women 18-24” then your results are going to suck for a multitude of reasons.
Just for the sake of contrast let me demonstrate a simple but otherwise realistic example about how you may want to think about personalisation using a topic you’re all familiar with.
Let’s say you’re working for GoDaddy and you’re trying to improve the revenue of users who land on your homepage for example.
You are going to run a lot of A/B tests to figure out that first part I mentioned before regarding what’s the best possible version of this page I can show to people in general and tweak all kinds of things like the kinds of products / offers I push (domains vs hosting for example), the order I show them in, the way I talk about them etc.
Cool? Concurrently to that I’m also going to start thinking about what are the kinds of customers we have.
For example, GoDaddy have a pretty wide variety of customers when it comes to technical sophistication to just pick a random audience attribute that is actually relevant.
There are a lot of HN crowd types who just want a cheap domain but want absolutely nothing to do with GoDaddy’s otherwise not great hosting plans. There are people who want very specific kinds of hosting like Wordpress, there are lots of non technical small business owners who are just trying to get setup with something basic and aren’t sure where to start among many other groups.
So I’m going to start thinking about where could I find those specific groups of people both at a targeting level (ads, sponsorships etc) and also from an identification level (I.e. maybe the referrer string on the HTTP header shows HN for example, maybe I see they are coming in via one of the ad campaigns I set up earlier specifically targeting Wordpress hosting, maybe the user agent shows a windows vista user running IE which says maybe they aren’t likely to be technically savvy etc)
Now I have some additional contextual information that I can use when I need to make a decision about what offers do I want to show in the hero section of the homepage for example. Do I push domains? Do I push how we specialise in working with non technical users etc.
That kind of thing works, consistently. That’s the right message to the right audience and at the right time (to be fair I’ve not really...
> There is an underlying principle in the world of sales / marketing etc that basically boils down to show the right people the right message at the right time.
That's kind of the industry mantra for us, marketers. The reality is much more complex than that, goes beyond that to be quite honest.
You just have to ask the question: "What's the right time to reach the right someone, with the right message?"
And you'll find out the answer to that exercise is a game of Darts, with hits and misses, and as a consequence a lot of wastage. But that's the game we chose to play.
Now, of course if you're reaching people who left your eCommerce shop and you retarget them with the items they have on their basket, it's a personalized ad that has a great probability to get results because you are reaching people who expressed a very specific behavior. But this, as you know, it's a rather small slice of your potential customers.
But for example, when you say:
>useless audience group like “women 18-24” then your results are going to suck for a multitude of reasons.
If your product offer is a pill that helps to relief period pain and cramps, why isn't this a valid audience? Or the pill needs to be a "a pill for early teen women that like skateboards and scratch off tickets"?
In my point of view the problem isn't the lack of personalization, or the need for it, but more a matter of saturation of some media channels.
Agree with you that IRL it’s more complicated than just 3 points but am writing this for an audience of developers and tech folk so figured simplicity was more important than exhaustiveness but is otherwise generally accurate as a high level mental model.
As for the straight up demographics example I called out I think that would work ok in theory but suffers from the major problem that you can not accurately identify that audience group to begin with and often doing so gets into the sleezier side of marketing that I personally want nothing to do with if I can avoid it.
On top of that, I think there are better approaches, women 18-24 are not one homogeneous group with the same wants and needs and treating them that way might work better than the idea of let’s just show the same message to everyone all the time but I hardly think it’s the pinnacle of tailoring your content either.
It sounds like the insider's version of "content", or perhaps it's just a more recent term. As a non-native speaker, it was really grating and hard to read, but obviously you're not supposed to understand other industry's jargon.
Creative is a very old term in the advertising industry.
"Content" is used more for long form/non advertorial type of content, like blog posts, video expanding on a subject, etc.
> As a non-native speaker, it was really grating and hard to read, but obviously you're not supposed to understand other industry's jargon.
Yes these are industry terms, for example: the term "copy" is also a very old term used to address the "written" part of ads (of course is more than this).
In the end it's all content and media, but it would be very hard to specify what type of content and media using these broad terms in this context.
It is quite obvious that it doesn't work, not even close. Anyone using some big tech software can see it with his own eyes - ADs are atrociously tailored to the viewer. Facebook shows me total bullcrap for as long as I'm using it. And using it regularly I can see how AD cycles come and go, sponsored posts unified by some theme change every several months - it can be an avalanche of t-shirts, then creepto, then some political party, then creepto again, then meme groups, then some China sponsored posts masquerading as personal blogs, etc. etc. All of them are a complete miss, then can see that there are clickthroughs, no related searches, no views.
Same with Google, MS, anyone. Trillion dollar corps specializing at harvesting and selling personal data can't figure it out. Personalized ADs are a myth.
Yes! It's just bots targeting bots in a giant chain of grifts where each player thinks they're the grifter. Authenticity on the internet is dead, read about the Dead Internet Theory. AdTech is a con and only serves to do damage to the world and enrich itself.
171 comments
[ 2.9 ms ] story [ 226 ms ] threadI suppose it helps disambiguate that from the content around which ads appear. Beyond that, I suppose language is no less subject to fads and trends than any other realm of human endeavor.
My earliest memory of struggling with this kind of language was in high school economics class. I was forever put off from the material by the confusing truncation of noun-phrases into adjectives.
And it's a shame, because there's a lot to discuss here: vast swathes of the technology industry hinge on the subjects discussed. Weinberg and Lombardo seem to be suggesting that an emperor has no clothes.
(I’ve worked for one of the top TV and cinema content companies in the world, and currently working for one of the top DSPs (advertising)).
There was a post a little while back arguing that Avengers isn’t even a kids franchise anyway, so your observation still stands.
Now if you need to target 100k hyper specific audiences, then it's a different story.
Are there good studies on this to validate it?
Companies spend a mint on this stuff because they are told it works and is necessary, I'm not sure that's true, especially when that comes from the sales/marketing arm of those ad companies...
Story branch decisions in gaming are to some extent just an evaluation of what the writer's think of your world view - at least when it comes to consequences. Arguably they can only really be constructed as a trial between two opposing narrative voices - resolve questions of character via the wisdom of the crowd.
I think different versions of the same movie would completely break word of mouth, though.
Discussion and memes based on movies don't work quite well if everybody is watching a different movie.
Similarly for a movie, perhaps we could personalize them, but would the result actually be desirable.
I'm so surprised that there are memes from big hits, but no real talk about them.
I mean there's relatively a lot of talk about how Harry Potter feels good but doesn't make sense, and of course lots of critiques, reviews, "reaction videos", but somehow it seems actual people stopped talking about movies with each other (beyond the very shallow have you seen it? yes/no. was it good? yes/no.)
maybe I just miss when people talked too much about the Matrix? :D
also I find it strange that it's still so hard to find good movies/books/series/games that I would like.... despite all the data, metadata, taxonomy, folksonomy (tags! unstructured data! big data! bad data, bad!)
for example, IMDB knows what I like, Netflix too, yet the recommendations are pretty bad.
let's take Mindhunter. "CSI done well", right? but I still haven't found something that is similar to it. (not that I looked very much.)
Similarly, clothing brands like stitch-fix and trunk club are personalization specific.
Regarding Disney, we just signed up for Disney+ and the first step was to decide if you wanted mature content and the second was to decide on your avatar. I think Disney+ will allow the company to cater to personal tastes more than they ever could through theaters and we will see that in the future.
Finally, for the longest time marketers have looked at demographic groups to target with their products, this is a weaker form of personalization. Obviously that won’t be abandoned, so I wonder why the author thinks companies should only target mass markets?
(Or so it seems to me)
I mean if you are into a movie there might be only three ones like it which you allready have watched and surely are not on Netflix.
However, as soon as you use your Spotify account for kids or parties the engine gets messed up forever. It is a shame since it helped me find alot of artists.
I also prefer Spotify user-run playlists better, which I pick myself, searching for genres and songs explicitly.
Nothing to do with preferring the personalization.
"No" seems to suggest that you're disagreeing with your parent's post, but your parent is saying why they chose it, not why you did:
> I assume, perhaps incorrectly, that everyone has the same catalog (at least for my fairly boring taste in music) and actively choose Spotify’s personalization over Apple Music’s tighter integration with my phone, car, etc. Just a single data point though.
And I am saying why I chose it. You don't need to read too much into the "no", nor have to find some perfect formal consistency in a quick response. It's just "no" as in, "no, that's not my case".
Any time Spotify is brought up (which is really any time music is brought up) it's a good bet the quality of discovery of recommendations are mentioned. The Discovery Weekly and Daily Mixes are loved. My anecdotal evidence is the dozens of people I've talked to about this over the years. This HN thread is the first time I've seen anyone suggest otherwise.
I know for myself, if it wasn't for the discovery that Spotify offers, I wouldn't use it.
Just remember that when you think this, the reality is that you are just that predictable based on half a dozen meaningless datapoints.
In fact, they're actively building it and selling it to advertisers. Disney+ and similar platforms are the tools that media companies have every intention of using to drive the accuracy numbers much higher.
E: I should note that I work for Snowflake, we're marketing this capability pretty hard. I'm not directly involved though.
You mean the recommendation engine? Hardly ever it worked for me.
What Spotify is built on is "play albums/songs you select or follow playlists on genres you like". The recommendations are tacked on, and bad.
I prefer to use YT Music s it seems thoroughly confused by having each person in family use from my service account to the point the suggestions are general in nature. They still do not get played though. I refuse to click on suggestions in any product as that reinforces their data on me/us. I’m quite ok if any music engine thinks I’m a polyglot toddler with penchant for death metal, Thai ballads, and ancient Chinese orchestral Music. Such a profile makes me soooo much less likely to get other music or junk marketed to me ;-)
Sure, I've had some good recommendations, but I'd expect any recommendation engine to do that just by chance.
What I can say is that Spotify has never proposed something truly new and interesting to me. If I listen to a lot of metal, it'll recommend me the most bland stuff in the same genre. What's even worse is that it keeps playing the same songs over and over again. It's like the recommendation engine just gives up and starts repeating its suggestions again.
I simply gave up on it. Why have access to everything when I only got the crap. Going back to my personal collection increased my good/bad song ratio considerably.
"It" being reality?
>If Spotify's recommendation engine wasn't as good as it was, it would not have the users it has
Now that's just confirmation bias + circular logic.
A streaming service can have users regardless of how good its recommendation engine is. YouTube has crappy recommendation (and has had worse for most of the time it existed) but tons of users.
For Pandora and Spotify, merely offering a convenient way to stream music, a free tier, and a big catalog, was enough.
You seem convinced of some bizarro idea that a media/streaming service can only succeed based on its recommendation engine.
Where did you get that from?
Maybe it varies dramatically by genre?
For example, take the "Phonk" genre, which I only was introduced to due to the Ukrainian war. To mean these recommendations give me a lot of new music to listen to. But then you have people lamenting that "Phonk" has really been overtaken by "Drift Phonk". https://www.youtube.com/watch?v=UAV7hnCB_ZE&ab_channel=yokai
On the other hand, I personally enjoy albums and songs major artist after they achieve critical success, when they establish their signature sound, like David Bowie(1970-1983), Stevie Wonder (1972-1976), or Peter Gabriel(1986-1992), but if I "like" any of these songs on Spotify, it means I get their entire catalog mixed in my daily mixes, and if they ever release a live album, those tracks show up. This is not what I want.
Music genres change, that's nothing new at all. Just look at the type of trance from the early 2000s and compare it to today's trance. It's a completely different sound. Would the author also complain about "progressive trance" taking over "trance"? It's natural that some sub genres might become more popular, while others lose listeners relative to the newcomers. Same happened in techno, rap etc.
While Spotify might accelerate this process through a positive feedback loop, this video is just another form of gatekeeping and saying "I knew XYZ before it was cool".
Which is neither here nor there. Apple music came later, and Spotify already had a headstart, a good UI, a good selection, and a good free plan.
Spotify, Pandora caught on because they were the first good streaming solutions, at the time bandwidth, mobile phones, etc, were in place and ripe for streaming. Not because of their recommendations...
That's not an argument
Canceled and would not consider a recommendation based service that does not let me explicitly remove something from being recommended.
Disclaimer: I work for Spotify nowadays but have been an avid user for more than a decade before joining, my opinion is based on my personal experience with the service and not as an employee.
Most major value creation happens well before Disney+ or Pandora. They just have somewhat broadly defined parameters that will get you more of the same, from what has been created.
As easy examples, there are a lot of good kids shows that adults could enjoy, but my guess is that they are not recommended to most adults. Just as really good folk music is likely to get suggested to someone that hasn't listened to folk music. Or music/movies from another nation.
Which is all too say that the personalization is ultimately a customer fitting themselves to what they want, from what they know. Dropping someone in fresh with no prefit is probably a lost cause. Even though that is how most personalisation talks brand themselves.
You're talking about their recommendation engine for users they already have, which is completely different.
We also have to take into account that there's a lot of "misses" in those recommendation engines - which is normal. But when they nail it, people have positive emotions towards it, so it's a good feedback loop.
Personalization would be to have Spotify recommend you only songs about expensive watches after you've searched for a Rolex on Google, with ads in the middle about Rolex ahaha.
As I read this article, this meaning of "personalisation" is not the one used by the authors.
Tech companies like Spotify and Pandora are just intermediaries. They may help to "deliver" commercial product, or advertising, however they do not create the product referred to in this article. The article uses the term "personalised creative" to refer to product. Tech companies operate as middlemen and produce no content. They are dependent on others to produce it. This is the bait for computer users. Tech companies sell advertising services to companies that produce content.
Tech companies gather data about individual computer users, e.g., web browsing histories. This is "personal" data. Tech companies allege this makes it especially effective and therefore valuable. The studies cited in the article suggest this is claim is false.
Disney produces content. There is no shortage of personal data being collected and sold by tech companies which is available to Disney. The article highlights that, despite the availability of personal data collected by tech companies, Disney generally does not produce "personalised creative".
In sum, the article is not about what "tech" companies do, it is about what content producers do. More specifically, it is about whether, based on available research, producers of creative should or should not attempt to use personal data about computer users collected by third parties in order to produce "personalised creative".
They do create a "product", the delivery system. This is a creative product as much as any of the music. A lot of software was written to allow this and all of that is the "product".
And to run counter to this article - people have been going after niches for centuries. To great success.
> Personalisation assumes that marketers have perfect data on every individual customer.
No individual believes this, so it's hard to believe an extrapolation into organizational commitment.
> Marketers infer who customers are based on their browsing behavior.
Not in isolation. This is mischaracterization.
> So, how accurate is gender targeting? It’s accurate 42.3% of the time.
Probably not. There is no access to the datasets (https://pubsonline.informs.org/doi/epdf/10.1287/mksc.2019.11...). "Gender accuracy ranges from25.7% to 62.7% with an overall average of 42.3%" without explaining how "no data" is handled. Having the actual data 42% of the time and not having it 58% of the time isn't the same as measuring how accurate it is. This looks like using bad statistics paired with confirmation bias.
On and on and on.
This is an article built on a quick summary of a questionable paper to push some anti-marketing prattle. Not to say that DSPs are reliable or that everyone is handling legitimate data, but bad data is filtered out pretty quickly in AdTech where there are means-tested costs...although it sometimes takes time to present.
Personalised ads result in vast amounts of corporate surveillance and still don’t work. I’ve never understood why contextual ads aren’t seen as good enough. If I’m looking at reviews for baby products, for example, then probably going to be interested in ads for similar baby products.
And with personalised content or product recommendations, very few companies have the breadth and depth of content that even begin to make it worthwhile, and even those that do can’t seem to do more than show you more of what you’ve encountered before. YouTube and Instagram’s personalisation algos can be downright annoying ‘You watched that one video randomly? Okay, here’s hundreds more like it, crowding out all the things you do watch regularly.’
Good search and discovery tools are always better than a stupid recommendation engine, and they’re all stupid.
Personalisation is effective. I've tested enough personalisation campaigns to be convinced.
And as a counter argument to Youtube/Instas algos. The reason TikTok has become so popular is that it simply better at personalising content, than say instagram reels.
Literally no such thing.
If you're actually involved in creative work, knowing your niche is extremely important.
> Recently, Professor Nico Neumann partnered with the brilliant marketing team at HP to replicate this research for B2B. The results were unsurprising – but horrifying. Many enterprise technology companies spend millions of dollars ‘hyper-targeting’ IT decision makers (ITDMs) using third-party data. But if we get gender wrong more often than 50% of the time, what percentage of ITDMs do you think are actually ITDMs, according to the research?
> Do you want to guess? It’s 14.3%. And for ‘senior ITDMs’, that number drops to 7.5%.
> Super impressive! That’s about as precise as… a drunk monkey throwing darts?
That... seems... like a great result? How many ITDMs are there in a random sample of 10,000 people? According to the Bureau of Labor Statistics [0], there are 715,000 people working as IT managers in the USA. So the incidence in a random population of 10,000 is 715,000/300,000,000 = 24 people. Seems like the ads are working really well.
I think it's interesting to see when you get a targeted ad you benefit from. I attended HackMIT for the first time after seeing a Facebook ad for it--I had an amazing experience at the event and that ad is possibly the reason I am a software engineer today. I doubt I would have received that ad if targeting weren't possible.
If I had to put my imagination to work where ads and humans can coexist peacefully, I could imagine a world where ads are required to be hyper targeted, and you would be shown no ads unless the ad was specifically targeted to you. If the ad wasn't relevant to you, you could click an x on it to make it disappear, and the advertiser would have to pay a penalty to the host of the ad (google, facebook, whatever) for inconveniencing their users. I suppose marketers at Disney and P&G would suffer, but I am really sick of seeing liberty mutual ads when I don't even have a car.
[0] https://www.bls.gov/cps/cpsaat11.htm
Hyper-individualized ads are just plain creepy. I’m glad to hear they’re generally ineffective at reaching their intended audience, and I’d hope that when they are effective, that the sense of invasion people feel causes them to turn the other way.
The article is arguing that instead, you should just craft a "Buy Viagra for you or your spouse because of these broad based benefits of Viagra" ad that is for everyone.
Which in this case would be something like putting ads up on a website or magazine that’s know to target an older male demographic. Or getting an older men’s health influencer on Instagram/TikTok to make a sponsored post. So still directed marketing, just not personalize on a user by user basis based on data from 3rd party brokers.
That's the Victoria's Secret business model, or was, so it's not necessarily a bad idea.
Please read the article again.
Unless you're maybe saying "the problem isn't the targeting, it's the type of ad"? But I don't see anything in the article that makes it specific to banners.
Ads deal with real life, does your AI engine know that because of a deadly heatwave sales will change?
Guessing which songs I may like based on 10 or 100 songs I like, while not trivial, is possible.
Guessing who I am and if I am the right target for a kind of ad (broad-reach, low-impacts ads) nobody pays any attention to anyway is delusional.
They are talking about ads, not about which songs are suggested to us.
> Most personalisation efforts are powered by third-party data. Marketers infer who customers are based on their browsing behavior.
In order to deliver personalised ads, they use cookies — third-party cookies, by companies that invade our privacy and track our every move.
Spotify does not rely exclusively on poor quality third-party data: they also have first-party data about us: the songs we listen to, like etc; how often we listen to them; if we put them in our playlists etc. This is one reason why their guesses as to what we may like are better.
Another reason is that they are ‘only’ trying to suggest another song to a person who, apparently, likes music; they are not trying to guess my gender, age, job, wealth, type of person I am, mood, interests etc and find a match between these tracts and the fact that I may be interested in… an ad I never requested from a company I don’t know…
> “We determined there was simply too much waste in the old model of activating this third-party cookie-based data across high-reach, low-impact placements.
My guess is that the type of ads they are talking about here are general rotation banner ads, i.e. garbage worth nothing which gets sold for good money under the (false) promise that companies have ways to show your ads to the ‘right’ people. Problem is, they don’t.
It has a big list of media and it's trying to pick one that will be effective on you. That part is quite similar.
> Spotify dies not rely exclusively on poor quality third-party data: they also have first-party data about us
Is that why they're saying it doesn't work? Or is it a different reason? Because I see people citing all sorts of possible reasons and intuition can only go so far in knowing which ones are important.
> My guess is that the type of ads they are talking about here are general rotation banner ads
Maybe. But in that case how much is the data the problem and how much is the type of ad the problem?
The problem is the type of ad, which is worth nothing. The data is just BS used in order to dress up the selling of a worthless ad placement for good money; and yes, that data is also largely flawed.
And even if it were not flawed, ‘guessing’ that someone who is supposed to be a certain age, gender, in a certain job, in a certain relationship, with or without kids, etc could be interested in a certain product is much harder than simply guessing that if you like these 10 or 100 songs, you may like this other song.
And how is anybody going to ‘prove’ that all this quasi-magical targeting worked, when click rates are ridiculously low even for highly targeted ads?
Imho targeting and ‘personalisation’ serve just one simple purpose: convince the marketer that he/she is doing the right thing, or what is considered right or best practice at the moment, and so that they are safe and are not going to be fired because, you know, nobody ever got fired for choosing IBM.
The Spotify data about like/dislike is rather accurate (like 80-100%), as long as someone doesn't click under substance influence.
Third party data (according to article) has accuracy around 4%-44%.
Their data somehow resulted in a statistically significant anti-correlation. This is a pretty unlikely result from a stats POV.
This is a lot more plausible than you might think. Simply build a machine-learning model that is accurate 58% of the time (maybe because it takes the "gender" field people manually entered on their online profiles and adds noise) and accidentally invert its output.
Those are all real categories (the word gender is just one of several English borrowed from other languages for the word "category", along with genre), whether or not you believe people should be allowed to choose the category they best identify with and that category should be distinct from assigned sex at birth. It's all real, whether you choose to believe in it or not.
One of the most interesting usecases of my company (videobolt.net) is personalized videos at scale (via API or one time video creation on a bunch of data). We powered some very interesting campaigns this way.
This perfectly demonstrates the point of this article - it's a canned and mass-produced product.
As far as I understand the article, two main points are that data on which you think you're doing personalization is often inaccurate, and that even if it is accurate you cannot hit the person well enough because that data is not enough.
As said, I pretty much agree on acquisition level, but don't on retention level.
On retention level, data should be good, since you are the one that is collection the data through your app / service usage. You should collect important data and trust your data.
If you are sure you have that data, you may very much provide your customer with interesting personalized marketing experience.
For given example, people use that platform every day, enjoy it, and are proud of their achievements there. They very much liked and shared these videos, as they are telling story of their journey on a platform that is important for them.
PS. I agree this is not for everyone. You should have immersive platform users are spending a lot of time at, with a bunch of interesting data. These would maybe even work on remarketing level, if you collected enough data in your system for given customer.
PPS. And difference between good quality video and PPT / GIF is... just huge.
Let’s forget the obvious demographics correlation with products and just start with an example.
Simply knowing that someone is active on HN, I wills bet ads on tech product and services would be more effective on them compared to a random person on the internet. The problem comes when the personalization gets too personal and stuck in the past, like how Amazon recommends another of the same product category after you bought one.
Without going super deep into the specifics they do make one good point that I agree with which is using 3rd party data brokers and sources to drive a personalisation strategy is going to give you terrible results precisely because that data is junk to begin with and is probably only going to get worse over time.
That is however probably where the good points in that article end.
In short, they take this ultra narrow view of what personalisation actually is and just straw man it to death.
There is an underlying principle in the world of sales / marketing etc that basically boils down to show the right people the right message at the right time. The idea that this approach doesn’t work is not in anyway backed up by any evidence whatsoever and the second half of the article that just says “make something great for everyone because… movies” is just incredibly stupid advice.
Like yes that is a great baseline to start from and is a goal in and of itself but what about when I start to know a bit more about what you’re actually looking for? Why would I not intentionally try to tweak my approach? A human salesperson who couldn’t adapt their approach would be out of a job within a short amount of time.
The problem ultimately comes down to how can I understand
1. Who is the RIGHT audience
2. What is the RIGHT message for that audience
3. What is the RIGHT time to show that message?
So it might come as no surprise to anyone that if you just outsourced that question to some 3rd party data source and had some incredibly broad and useless audience group like “women 18-24” then your results are going to suck for a multitude of reasons.
Just for the sake of contrast let me demonstrate a simple but otherwise realistic example about how you may want to think about personalisation using a topic you’re all familiar with.
Let’s say you’re working for GoDaddy and you’re trying to improve the revenue of users who land on your homepage for example.
You are going to run a lot of A/B tests to figure out that first part I mentioned before regarding what’s the best possible version of this page I can show to people in general and tweak all kinds of things like the kinds of products / offers I push (domains vs hosting for example), the order I show them in, the way I talk about them etc.
Cool? Concurrently to that I’m also going to start thinking about what are the kinds of customers we have.
For example, GoDaddy have a pretty wide variety of customers when it comes to technical sophistication to just pick a random audience attribute that is actually relevant.
There are a lot of HN crowd types who just want a cheap domain but want absolutely nothing to do with GoDaddy’s otherwise not great hosting plans. There are people who want very specific kinds of hosting like Wordpress, there are lots of non technical small business owners who are just trying to get setup with something basic and aren’t sure where to start among many other groups.
So I’m going to start thinking about where could I find those specific groups of people both at a targeting level (ads, sponsorships etc) and also from an identification level (I.e. maybe the referrer string on the HTTP header shows HN for example, maybe I see they are coming in via one of the ad campaigns I set up earlier specifically targeting Wordpress hosting, maybe the user agent shows a windows vista user running IE which says maybe they aren’t likely to be technically savvy etc)
Now I have some additional contextual information that I can use when I need to make a decision about what offers do I want to show in the hero section of the homepage for example. Do I push domains? Do I push how we specialise in working with non technical users etc.
That kind of thing works, consistently. That’s the right message to the right audience and at the right time (to be fair I’ve not really...
That's kind of the industry mantra for us, marketers. The reality is much more complex than that, goes beyond that to be quite honest.
You just have to ask the question: "What's the right time to reach the right someone, with the right message?"
And you'll find out the answer to that exercise is a game of Darts, with hits and misses, and as a consequence a lot of wastage. But that's the game we chose to play.
Now, of course if you're reaching people who left your eCommerce shop and you retarget them with the items they have on their basket, it's a personalized ad that has a great probability to get results because you are reaching people who expressed a very specific behavior. But this, as you know, it's a rather small slice of your potential customers.
But for example, when you say:
>useless audience group like “women 18-24” then your results are going to suck for a multitude of reasons.
If your product offer is a pill that helps to relief period pain and cramps, why isn't this a valid audience? Or the pill needs to be a "a pill for early teen women that like skateboards and scratch off tickets"?
In my point of view the problem isn't the lack of personalization, or the need for it, but more a matter of saturation of some media channels.
As for the straight up demographics example I called out I think that would work ok in theory but suffers from the major problem that you can not accurately identify that audience group to begin with and often doing so gets into the sleezier side of marketing that I personally want nothing to do with if I can avoid it.
On top of that, I think there are better approaches, women 18-24 are not one homogeneous group with the same wants and needs and treating them that way might work better than the idea of let’s just show the same message to everyone all the time but I hardly think it’s the pinnacle of tailoring your content either.
> Disney only invests in creative that works across all segments – angsty superheroes, lost animals, magical princesses.
Are these quotes typos, or is "creative" as a noun some sort of jargon? I find it really grating; what was wrong with "media?"
>I find it really grating; what was wrong with "media?"
We use "media/medium" to address the distribution channels of the creatives.
"Content" is used more for long form/non advertorial type of content, like blog posts, video expanding on a subject, etc.
> As a non-native speaker, it was really grating and hard to read, but obviously you're not supposed to understand other industry's jargon.
Yes these are industry terms, for example: the term "copy" is also a very old term used to address the "written" part of ads (of course is more than this).
In the end it's all content and media, but it would be very hard to specify what type of content and media using these broad terms in this context.