I know I'm living on the blogger's boulevard of broken promises, but it's another short issue without a review. If you live *checks news* anywhere... it's likely been a hard week for you too. Maybe this will make you smile.
Hoping you, your friends and families are safe.
This newsletter is for both free and paid members of The Content Technologist. Want to receive the final paid-only installment of the personalization series?
Upgrade your membershipCan content personalization algorithms satisfy the multitudes of personal taste?
This is second in a three-part series about personalization. Read last week's definition of personalization tactics.
Our personalities have as much variety as our fingerprints, and while we see common patterns among large groups within cultures, we're assured: no one group is exactly like another.
Perhaps because I'm properly steeped in American individualism, I expect brands that claim they're personalizing content to accurately reflect my preferences at some sort of individual level. It's the promise of digital delivery over mass communication: you don't have to blanket broadcast content to all people and can adjust to their preferences, niches and tastes. If all works as tech marketers have promised, right time, right channel, right place should be in our grasp for delightfully personalized experiences.
When actually engineering and optimizing content for personalization, two major challenges arise. The first, familiar to all production and ad sales personnel: the inventory problem. The second, my personalized take on human taste that probably has a better name in some econ textbook: what I call the Interpol problem.
The inventory problem: More creation = more content investment
Marketers and publishers have been banging the segmentation drum for years, and segmentation technology is everywhere. Even open source little engine that could email/CMS software like Ghost offers methods to segment content through tagging. Audience segmentation, we are promised, enables us to reach all sorts of specialized groups as long as they're tagged correctly.
If you're a digital editor with experience in segmentation, I assume your stomach is churning at the thought of what I'm about to say: for every new segment your software can create, you've saddled yourself with an entirely separate experience to develop and populate with richly resonant content.
When users in one segment see completely different content from another segment, someone on the creative end has to create that wholly separate content experience. In my experience, for creative content teams that are already strapped for resources, segmentation is rarely accompanied by an increase in budget or staff.* Segmentation tech companies usually fail to mention how much content inventory their clients need to develop when selling their software to willing publishers and marketing teams.
Serving an individual even one daily piece of relevant, quality, personalized content is a more gargantuan task than 99.9% of media companies can handle. Even Netflix, with more data, content and engineering prowess than most platforms, now resorts to the same mass marketing tactics that network TV has used for years: Just show everyone the same ads for the same six programs because all the people seem to like them so much.**
For many companies, web or email content personalization correlates to a decline in quality content and customer experience for all but the most prioritized segments. Personalization through segmentation can be done well, but the costs of creating custom content for multiple segments need to be considered before the technology is implemented.
Creating enough inventory for multiple segments on a weekly or daily basis means that content is fast, voiceless and shallow—the exact opposite of the web's strengths. Machine learning creation tech like GPT-3 still requires a massive amount of editing to be readable, let alone creative or useful to an expert audience.
User-generated content can absolutely feed personalization preferences—but unless you have the bandwidth of Amazon, Google, TikTok or Facebook, you're unlikely to attract the quality and reach that personalization promises.
If you're looking at segmenting content for personalization, I recommend the following to address the inventory problem:
- Only invest in segmentation if you have 10,000+ people in your audience. ROI's not worth it otherwise.
- Work with three segments, tops, until you really get the hang of the segmented content workflow.
- Give your audience the option to opt into segments, rather than automatically sorting them.
- Audit your segmentation user experience with analytics. Is one segment getting a better experience than others? Are folks in one segment opting out more than others?
- Track your ROI from content segmentation. Do your customers care? Would you get the same results without investing so much in filling your inventory with content?
Let's say you've engineered a really advanced content segmentation workflow. That's awesome. Your audience gets content personalized to their preferences and behaviors. They're super happy with you. So what happens when your super advanced computer starts recommending the same old thing?
**I mean, unless you work with a consultant like me to adequately plan for personalized content before software is selected.
**I don't know which segment the "Bob Ross's sordid life" documentary was created for, but every time it's thrown at me I get a little more turned off. I really wish that after 25 without clicking on it or showing any engagement, the whole thing would just disappear from my options.
The Interpol problem: Can predictive analytics really meet the demands of taste?
The second massive problem with personalization tech: the better it gets, the more your audience expects. The bigger it gets, the harder it is to actually personalize to your audience's preferences. In practice, personalization at scale can be too much of a mediocre thing.
Again: people=large, contain multitudes. In my experience, individuals don't always love being grouped into segments in which they had no choice in defining. Particularly when segments are based on demographics or audience likeness, serving content based on perceived preferences can change from a delightful brand experience to a frustrating customer experience in a split second.
Disclaimer: The example I'm about to provide seems petty and benign, particularly in the context of climate collapse and the gutting of women's rights from the past couple of weeks, but stay with me. I promise it's relevant.
I have loved pop music my whole life. As a tween I devoured rock rags intended for readers way older than me, obsessively listened to Philly's indie radio stations and attended concerts as soon as I could argue my way to permission. I adored Sleater-Kinney and rolled my eyes at the Spice Girls. My taste in music is admittedly described as either "discerning" or "snobby," depending on whom you ask. When I saw High Fidelity and Ghost World as a Y2K teen, I felt seen. Finally, proof: there were other people like me in the world!
I'm comfortable with liking what I like and avoiding what I dislike, and I know that my music tastes put me in certain cohorts, like assigning me a lunch table in the high school cafeteria. If you look at all the data points of my life, you'd find that I'm extremely likely to be an Interpol fan. For example, I:
- Enjoy all of Interpol's influences
- Have long been a huge fan of other bands in Interpol's cohort, like the Strokes, the Yeah Yeah Yeahs, LCD Soundsystem and Les Savy Fav
- Went to the same college as many members of Interpol
- Saw Interpol live before their self-titled EP and first record were released
- Brag about how I saw Interpol live before they were cool
- Used to hang out at the same club that Interpol did, according to the book Meet Me in the Bathroom
But:
I do not care for Interpol. I actively turn off Interpol songs if I have the choice.
Record companies, dudes at parties, ex-boyfriends, and celestial jukebox algorithms have been trying to sell me Interpol for most of my life. Have you heard of Interpol, they ask? Yes. Yes I have heard of them. At this point I've been sold Interpol more times than I've actually listened to Interpol songs. I just would rather be listening to Joy Division. Or, I dunno, a band that's not all dudes with the same exact set of artistic references as everyone else I've ever met.
I've written before about how much I love Spotify's personalization algorithm, which nails my taste, more or less. Although I've actively gone into the mobile app and turned off Interpol, I think, Spotify still throws an Interpol song at me every couple of years in one of my Daily Mixes, like "How about this time???"
No. No no no. No Interpol. Skip. Next song.
In true personalization, people should be able to actively dismiss the content that they're not into, and the algorithm should remember that, no matter how many new datasets are deposited into the data lake. Algorithmic memory can and should accommodate that.
So that's the Interpol problem: Voracious content consumers often know more about why they like the content they consume than an algorithm does, and content that's predictable is just that. Predictable content annoys the many individuals seeking something artful, novel, fun—the people often referred to as tastemakers and gatekeepers. The more you recommend content that annoys your customer, the more actively they reject your technology and your brand.
If Spotify can't figure out that I don't like Interpol, none besides the best of my friends (and the readers of this newsletter) ever will. The promise of personalization—everything you want on demand, nothing you don't—never actually meets reality, especially if there's not enough content inventory to meet your audience's fickle, inconstant, unlikely demands that belie their most obvious data points.
When individuals say they dislike a particular type of content in a personalization system, they tend to mean that they don't want to see that type of content again. Not because it's damaging or harmful, but just because it's annoying and they're tired of it, no matter how many times people like them have viewed that content. Individual content consumers have less tolerance for annoying or distasteful content when they think it's personalized. If you don't get me, I can move on.
If you're claiming true personalization... you'd better be right most of the time. Remember your user's preferences for longer than two years in a row. Otherwise, maybe spend your money and time creating one really good piece of content, rather than 30 hastily conceived "personalized" pieces. There are too many other good bands to discover to be spending my time on Interpol. I'm happy to change the channel to something fresh.
Stay tuned for next week's finale: the good-better-best of personalization tech as it exists in 2021. Because I truly enjoy personalization tech and see its promise—as long as it's not selling me Interpol.
Content tech links of the week
- Do our music moods predict market behavior? Recapped for The Conversation, this research says: maybe. I don't know whether I'm sold on the sentiment analysis here, but it's a curious idea.
- Why are hyperlinks blue? on the Mozilla blog, via HeyDesigner, an excellent curated newsletter.
- On the TextTok, aka I have to read TikToks now? In Embedded.
The Content Technologist is written by Deborah Carver, an independent consultant based in Minneapolis and a member of Indee Marketing Co-op.
Want to work together parsing out content analytics, SEO, design and strategy or any of the other topics covered in this newsletter? Get in touch or just reply to this email.