Content Personalization: A Reality Check
I’ve had a front-row seat to the rise of personalization in CMS. This is the idea of customizing content to the visitor in real-time – managing individual content elements and tossing a page together at the last second, based on information you’ve gleaned about what the visitor is doing and what they might want.
I remember “business rules” getting built into Ektron more than 10 years ago. Seven or eight years ago, we implemented ad hoc personalization in a CMS using language tools (“Cincinnati Customer” was its own pseudo language translation, for example). Five years or so later, the product manager of Episerver explained his crazy idea of building-in a framework to custom-tailor content based on user behavior.
Today almost every major commercial CMS vendor is doing this. We have lots of experience with Episerver’s implementation, we’ve played with Sitecore’s, and I saw it in a Magnolia demo the other day.
But here’s the problem: almost no one uses this feature – in any CMS.
Vendors will argue this, of course. They have case studies holstered on quick-draw and they’ll shoot from the hip if you ask for one. But if they’re being honest and you phrase the question right – not “does anyone use this?” but “what percentage of your customers use this?” – they might admit the truth: usage of personalization is in the low single-digit percentage if they’re lucky.
At Blend, we’ve never seen a customer use these tools beyond some very basic things, like offering a coupon in certain geographic areas or making a banner display if the user lands due to a specific marketing campaign. And even then, this usage rarely sticks. The customers stop doing it after a while.
I thought it was just us, so I asked around. I asked multiple companies that implement multiple platforms if anyone was using the personalization tools available. The answer, again and again, was “[long pause]…no.” And my queries were deep and wide – these firms ranged from smaller companies like Blend to mid-range firms with headcounts in the hundreds to massive global ad agencies that regularly sell CMS licenses in the seven figures.
Some of the answers:
“Nobody uses it to the level [Vendor X] sells on.”
“Minimal adoption at all.”
“You can find lots of great demos online that always look perfect, but it’s never as good as in the real world.”
“Well, our clients tell us that they will. But they never actually do.”
It’s to the point we’re this is the elephant in the room. It hardly ever gets used, and everyone knows it. Prospects swear they’ll use it – and we absolutely hope they do – but in the back of our heads, we’re know they won’t. The possibilities are intoxicating, but if pressed on actually implementing them, the average CMS customer usually doesn’t even know where to start.
One of my favorite analogies is that it’s like sex in high school:
Everyone talks about it. No one knows how to do it. Still, everyone is convinced everyone else is doing it. So everyone wants to do it themselves. But in the end, no one is actually doing it.
Why is this?
Problem #1: It’s not a technology problem, it’s a content problem
The technology for personalization is relatively solved. It still hasn’t filtered down to lower-end systems, but most commercial CMS from the mid-market upwards have implemented personalization as a matter of survival. It demos incredibly well. In a competitive sales situation, you simply have to have it, because your sales prospects will see it from your competitors.
But the problem is not technology, it’s content and process. To personalize an experience from a revolving bank of alternative content assets, you have to have those assets.
Do you have them? Do you have a bunch of unused content assets sitting around? If not, can you create them?
The sad truth: most organizations can barely manage a single effective version of their content, much less multiple versions. No client I know has multiple versions of their content sitting around, tailored to different audiences. This is something you see in case studies, but that few organizations actually have the resources to do.
Creating content is always the thing organizations sort of take for granted. We’ve seen finished implementations sit for years (not a typo – literally years) waiting for content questions to be resolved and content to be created. For a lot of organizations, just getting the content together to launch can be difficult.
If a single version of content is so hard, how are you convinced that you’re going to be able to slice-and-dice multiple versions of content into small, manageable chunks, then be able to organize them into a coherent order, on-demand? Throwing content together ad hoc is a lot harder than you think.
If you plan on a deep dive into personalization, here’s the first step: double your content creation budget. (I know, I know, you probably don’t even have a separate content creation budget, because content just kind of happens, right? Well, create a budget. Make it big. Then double it.)
Problem #2: Many organizations haven’t identified their individual audiences and might not have enough audiences to effectively differentiate
The idea of personalization is that you can identify personas – individually addressable demographic groups – and then deliver them combinations of content that are meant to be effective for them, specifically.
For instance, if you have a university, who are your audiences? Clearly, potential students are important, but you can subdivide them as well –
And each one of these groups might be combined with other groups to form composite groups with multiple characteristics and corresponding goals.
Have you done this audience analysis? Do you have a list sitting around of all the individual audiences who might visit? (I hope so, because if you don’t, personalization is not your biggest issue right now…)
Assuming this list exists, how many audiences do you actually have? The university example is an outlier (we know this because we’ve actually studied those audiences). Some organizations with limited product lines might only have one audience: potential customers, full stop.
The critical question is this: can modifying content for different audiences actually provide you with an advantage? Are your audiences far enough differentiated that you’ll be able to assemble content combinations that speak to one audience more effectively than another?
Yes, it would be great if you had a massive library of case studies to apply to subtle demographic shifts, but you probably don’t, and at the end of the day, your audiences might be clustered so tightly that one of them probably won’t respond any more effectively to Case Study A over Case Study B. You might be able to convince yourself that Case Study B is better, but only analytics will let you actually know (see below).
I think it’s true that you can make an individual argument to any one specific person that is more effective than a generic argument, but how large is this degree of differentiation and will you get ROI for trying to make it? Are the needs and points of pain for your customers that different? And even if you can find some combination that will be more effective than another, is it worth the complication and expense to develop it and maintain it over time?
Problem #3: Identifying audiences based on technical markers might be harder than you think
Where the rubber hits the road in personalization is in actually bucketing audiences. You can’t show them different content unless you can put them in different buckets. And can you? This is not magic – a CMS will give you a finite toolset of detection devices to try and figure out what audience bucket a particular visitor fits into.
Assuming you’ve differentiated your otherwise-anonymous audiences, how can you tell one from the other? In general, you have two options:
For the request, you can figure out things like what site they linked through from, or where they are geographically (though geolocation is sometimes vague). For the behavior, you can analyze things they’ve done on your site during this session, or perhaps in other sessions (assuming they don’t block cookies or doing anything else to disrupt tracking).
Can you differentiate audiences based on this? Some of your audiences might be extremely finely-grained, especially when it comes to purchasing behavior or objections. For example, consider these two audiences:
Value shopper: someone who wants the best price
Features shopper: someone who wants the most features
That’s a pretty standard dichotomy and one that any marketer should be prepared to handle. But, assuming you have the content prepared and ready to go to address each different situation, how can you tell them apart? What actual, technical parameters can we put in place to decide if Visitor X belongs to one audience or the other?
A couple ideas off the top of my head:
Check inbound search parameters. Are they searching for “cheapest,” or some feature name? Did they enter the site through a new article that talk about how cost-effective your product is? Are they browsing on a 10-year old flip-phone? (Ha! I kid, I kid…)
Check behavior on the site. Do you have a separate pricing page? Do you have separate pages for all your features? Maybe we could check what content they’re browsing. But how accurate would this be? Even someone who is feature-driven might check the pricing page.
Even if we bucket based on this, how rock-solid is our thinking here, and how many of our eggs do we want to put in that basket? If we think they’re a value shopper, do we banish all talk of features and just concentrate on price, price, price? What if we got it wrong? What if a more balanced (read: non-personalized) approach is safer? Does our degree of commitment to a particular audience bucket vary based on the technical confidence of our identification?
When we run out of request and behavior markers, we probably need to start seeding content – setting “content traps” for people to fall into. Perhaps, write some article about “Why Product X is the Most Cost Effective,” and hope people click on it, and thereby giving you a chance to identify them was a “value shopper.” So now you have to create more content and run a social media campaign just to get the visitor to self-identify and allow you to personalize.
Identifying audiences is both (1) marketing strategy, and (2) hacking. The former is abstract, theoretical, and something you do in a conference room with your team. The latter is specific, exacting, and something one person does from code or configuration based on a finite array of tools.
There are a few layers to how this might play out:
Your CMS administrator looks at your audience descriptions and tells you to go back and be more specific about how they might be identified. Remember, they don’t want descriptions, they want methods and algorithms.
You come back with more specific methods and your CMS administrator responds, “This is not technically possible. I have no way to find out if they have ever read Infinite Jest.” Joking aside, learning that your identification method isn’t feasible is a very real possibility for much more boring reasons.
You can implement your identification method. However, after a few days you’re not getting any users in that audience bucket. Is this a problem of marketing strategy (that audience does not actually exist) or technical identification (you are just not identifying them correctly)?
You actually implement your method, and you appear to get enough users in the bucket that you’re pretty sure this is a valid identification method. Note that you don’t know if the personalization is actually improving anything (hell, it might be making things worse), but you’re halfway there, at least.
Know that result #4 is pretty rare on the first attempt. Getting to that point will probably take considerable trial and error, and it will usually always be more complicated than you think.
Remember: just saying “Audience X is a specific group” isn’t enough. You have to translate this is into the technical parameters that your CMS can identify to know that Visitor X is indeed part of that audience. Are you prepared to do this?
Problem #4: You’ll need non-trivial analytic talent to really know if it’s working
Personalization doesn’t live in a vacuum. It’s fundamentally about coercing user behavior through a considerable investment of time and money. The thing that’s going to make you want to continue making that investment is proof that it’s working, which means analytics.
Do you have the analytic foundation to know whether or not your personalization is worth the trouble? This can be hard on websites where you have no “hard” conversion. If you have an ecommerce site where people checkout, or a subscription site where people can sign up for a free trial, questions of conversion tracking and ROI are simpler.
But if your conversion is that people are just more impressed with your brand, how are you going to measure if personalization is having any effect? Clearly, this is a more general analytics problem, but personalization consumes real budget in excess of the baseline required to run a website, and that budget will need to be justified.
To measure this effectively, can you:
Isolate audiences who have been exposed to personalization?
Identify the exact combination of personalized elements that caused them to convert?
Determine if they converted any differently than a control group?
Determine if they directed interacted with (clicked on) a personalized element that constituted (or initiated) the conversion?
Yes, this can be done. I"m not casting doubt on our friends over in the analytics group, but just understand that you will have to embark on personalization hand-in-hand with them if you want proof that it’s working and worth the additional expense. Plan on that.
Personalization without accompanying analytics will produce an expense with no corresponding proof that it’s worth the investment, and that is hard to sustain over time.
Problem #5: Your options for delivering personalized content might be more limited than you think
Assuming everything above is going swimmingly, what do we actually do with our personalization tools? What personalized content do we deliver? And let’s not talk in theoretical terms here – let’s get specific: a person has made an HTTP request to your server and you have the opportunity to change the HTML in the response based on what you know about this person.
What do you actually do here?
Remember that you’re somewhat limited by what the user is asking for. When they click on a link and make a request back to your server, they are expecting a certain content payload, and you need to deliver that. If they clicked on “Show Information About Product X,” then they want to see information about Product X, and you can’t just swap in Product Y because you think it will convert better, no matter how much you’re convinced you know about this visitor.
This means that you have three options for actually personalizing content:
Modify the main content of the page (the thing they requested)
Swap in supporting elements around the main content (associated promotional content on the page)
Change more systemic elements of the site (navigation, color palette, etc.)
For #1, it means we might swap out an image, or change some wording, or somehow “flavor” the content to make it more palatable for this visitor, while still delivering the same general content they requested. But this can get very finely-grained. You might find yourself trying to manage content down to the sentence and adjective level, chunking narrative content up into paragraphs (again, this is hard), or just duplicating entire blocks of content so you can change a single word.
For #2, you have secondary content areas like sidebars, footers, banners, etc. What you’re trying to do here is display some content that might catch their eye if the main content doesn’t work. Or perhaps keep driving home some content so they might acknowledge it at some point through sheer repetition.
And #3 is more subtle, but you might run into usability issues if you start re-arranging large-scale site structures (like navigation) mid-session.
When I discussed personalization with a client once, I kept asking them, “Okay, now that we know who the visitor is, what do you want to do about it?” They kept responding: “Well, we want to change the value proposition.”
This isn’t enough. There is no button in your CMS that says “Change the value proposition.” You are going to have specifically state something like, “For Audience A, when they visit Page B, we are going to place Promotional Item C in Position D.”
And this is the entire problem of personalization in microcosm.
This lofty idea is a continual process of dragging the theoretical down to the concrete – the progressive reification of an ideal. This is always hard because we love big ideas and we’re subconsciously afraid that reality won’t live up to them. We often never make concerted attempts at it because (1) even superficial consideration causes us to realize it’s not nearly as glamorous as we expected it would be, or (2) we immediately run into problems – both intractable and mundane – which make us think we’re clearly not doing this right so perhaps it’s not for us.
In my experience, this is the enduring story of content personalization: boundless potential, messy specifics, and untapped reality.
In the end…
Please know that I don’t mean this post to be overly cynical, and I don’t want to condemn personalization as completely unworkable. There’s nothing inherently bad in the theory of personalization, and many different platforms have personalization tools that they have made as easy to work with as possible.
But this is not a technology problem. This is a problem of:
Personalization is just a new backyard that those disciplines can run around in. You simply can’t buy a CMS that has personalization tools and think that they’re going to make things all better out of the box.
If you’re still convinced personalization is for you, then you need to answer these questions before you invest:
Have you identified your different audiences?
Can these audiences be identified through behavior or request-based markers? Specifically, how?
Do you have value propositions for all of them? Do you know what alternate content would influence them?
Do you have this alternative content for each audience to do this modification?
Will you be able to continue this increased level of content creation and management over time, post-launch?
What are your options for actually modifying a delivered page of content? Does your design support (or can it be modified to support) personalized elements?
Will you have the analytics in place to determine if any of this is working?
If you have answered yes to all those questions, then you’re in pretty good shape and you can probably make this work.
Just know that when all costs are factored in, a deep and effective personalization strategy will likely be the largest digital investment you can make. Over time, it will easily eclipse the cost of the CMS and the implementation combined.
There is no master switch we just flip and make this work. Personalization tools do not eliminate work. They create new work.
Hopefully, you can make that work pay off.
Postscript: Episerver, Peerius, and “Third-Wave” Personalization
A few weeks ago, Episerver purchased Peerius, which is software to analyze visitor behavior and personalize automatically based on algorithms.
If we look back at personalization in waves, we can clearly define two.
First Wave was known personalization. I know who this person is, so I can adjust content based on that.
Second Wave was anonymous personalization. I don’t know who this person is, but I know information about them, so I can adjust content based on rules.
Is something like Peerius the Third Wave? Is the Third Wave going to be algorithmic personalization in the mid-market? More cynically, is this indicative of a trend of CMS vendors admitting their customers can’t use the Second Wave tools and attempting to automate it for them?
Given my own long history with Episerver, I reached out to CMO James Norwood about the Peerius acquisition and what it means for Second Wave Personalization (“rules-based personalization,” or “RBP,” as he abbreviates below) – a technology that Episerver has hung its hat on for the last five years.
James simply doesn’t think it’s a “zero-sum game” where the decision to do Second or Third Wave is binary. Here was his response (quoted with permission):
[…] it’s not an either/or situation. The whole position is that you can’t do it all with rules, it’s just overwhelming, but rules and triggers do have their place. Episerver has one of the best RBP frameworks on the market but everyone oversells what these RBP can do.
Now with what we call autonomous personalization we can do so much more via the machine (it learns it improves it delivers), but you can always then overlay that with rules and triggers for what we call assisted personalization, making sure you get the best of both and maintaining control for marketers and merchandisers over their strategies.
[…] I think Episerver is solving a very real problem with Peerius, and that is the problem of actually putting personalization to work in a way that doesn’t overwhelm users, take up all their time but leaves them in control of merchandising and marketing strategy.
It’ll be interesting to see how the Peerius acquisition plays out, and if this is indeed the vanguard of Third Wave Personalization. Stay tuned.