Good morning, まさきん here.
I recently reread some notes I’d taken on an article from overseas I’d read a while back. It’s about what it actually takes to pull off 1:1 marketing — what we usually call personalization.
It’s an old article, but the framework itself still holds up today, I think, so I wanted to organize my thoughts on it again here.
That old survey number is just an old number
The article cited a survey result: fewer than 40% of companies had properly implemented personalization.
But this survey is from around 2016. It doesn’t describe where things stand today — I think the right way to read it is as a historical reference point, a “that’s roughly where things were back then” kind of number. Adoption is surely much higher by now, and generative AI has probably shifted things even further since.
That said, the reasoning behind why implementation is hard still felt like a solid breakdown to me. The article named three elements as essential.
The three elements: data, content, and logic
Data is the foundation for accurately understanding each individual customer. Purchase history, on-site behavior, membership attributes, and the like need to be organized to a level where you can actually act on them. On an e-commerce site, for example, that means the products someone viewed, added to their cart, and actually bought are all properly tied together under one customer profile. On top of that, stale information doesn’t help much. Only once what someone viewed yesterday and what they bought today are connected in something close to real time does the data become useful for the next recommendation.
Content is the material you vary from customer to customer. Even for the same recommendation email, someone shopping for baby food and someone stocking up on daily essentials for a one-person household need to see different products and read different copy. Even classic email-marketing scenarios like cart-abandonment or win-back campaigns boil down to the same question once you dig into them: what content goes to whom. With too few variations, you end up sending roughly the same thing to everyone.
Logic is the mechanism that decides which content goes to whom. Sometimes that’s rule-based branching; sometimes it’s a machine-learning-driven recommendation. Even a simple rule like “if baby food shows up in someone’s browsing history, prioritize parenting-category recommendations” counts as a perfectly legitimate piece of logic. Even the original article singled this out as the hardest part. Even with data and content both in place, personalization doesn’t actually get moving without a brain deciding how to combine them.
Missing even one of these three leaves personalization half-finished. You end up with plenty of data but only one version of content, or plenty of content but no logic to decide who sees what. It’s only once all three are in place that you can actually deliver a different experience to each customer.
Generative AI has lowered the content barrier considerably
Of the three, I think content is where the biggest change has happened.
It used to take a fair amount of manpower and time just to prepare copy and images for each customer segment. Even if you wanted to expand from a handful of segments to ten, the effort of producing all that material meant you’d often settle for just a few in the end — that was a pretty common story. Now, feed a set of conditions to generative AI and you can churn out similar copy in bulk, in a fraction of the time. Product-description variations and differentiated taglines have both gotten a lot easier than before. It’s not too far-fetched anymore, cost-wise, to write copy for every single customer segment you have.
Logic feels more within reach than it used to be, too. Machine-learning-based recommendation engines are now available as packaged services, so there’s much less need to build a statistical model from scratch than there used to be.
Phone plan recommendations probably run on similar logic. I figure I’ll start by running a rate simulation to check my own case.
Clicking this opens the Rakuten login page. Once you log in, you’ll see the campaign details.
The real bottleneck that still remains
Still, now that the barriers around content and logic have dropped, something else has come into sharper focus.
First, there’s data. No matter how much copy generative AI can churn out, if the underlying data is messy or fragmented across systems for the same customer, you end up delivering content that misses the mark. If membership info lives in System A, purchase history in System B, and in-app behavior logs in System C, you can’t accurately grasp who each individual customer even is in the first place. If anything, having a solid, high-precision data foundation is becoming more of a differentiator than ever.
As for logic, now that the technical bar has dropped, a different kind of difficulty has become more visible: the judgment call of deciding what actually counts as good logic for your own customers. Logic that maximizes revenue doesn’t necessarily lead to customer satisfaction. Logic that keeps recommending discounted items might grow short-term sales, but over the long run it risks training customers to always wait for a markdown. That connects with the idea that CRM is fundamentally about not creating “bad revenue”. This feels like a piece you shouldn’t hand entirely over to machine learning — it needs a person making the call and designing it.
Even as the tools keep evolving, deciding who you’re delivering something to, and why, still comes down to a person in the end. Now that data is well-organized and content can be mass-produced, I suspect the thinking behind how you design your logic is going to become an even more direct mirror of a company’s actual values.
The same thinking shows up at home, too
This “who, what, and why” framework doesn’t feel like something that only applies to work.
Deciding how to divide up phone plans and contracts among family members turns out to be a surprisingly similar judgment call to personalization. The family member who uses their phone a lot gets the bigger-capacity plan; the one who barely uses it gets the cheap plan. Data — usage patterns — and logic — who gets assigned what — seem to spin around naturally inside a household too.
I’ve actually gone through and reviewed my own family’s phone plans with that lens. Take a look if it’s useful.
Clicking this opens the Rakuten login page. Once you log in, you’ll see the campaign details.
This article contains affiliate advertising. If you sign up for a product or service through a link on this site, we may receive compensation from the partner company. The site operator is also an employee of Rakuten Group and may receive compensation through an employee referral program. The content and opinions here are based on the operator’s own experience and research regardless of any advertising relationship, but please read with the above in mind. See the Disclaimer & Affiliate Disclosure for details. Disclaimer & Affiliate Disclosure
This article includes machine-translated content. Please check the official Rakuten Mobile multilingual page for exact terms.
If you have concerns about Rakuten Mobile coverage or signal quality, you can consult through the official signal improvement request form .
