
Everyone thinks AI made personalization easy
Why one-to-one marketing still isn't simple, and what has to change in your tools, your team, and your process.
In The Godfather, one of my favorite movies, Michael Corleone is the son who was never supposed to join the family business. But after an attempt on his father’s life, he calmly volunteers to commit a murder, and his brother Sonny teases him for taking it personally. Michael’s famous answer marks the start of his transformation into the don he would become: “It’s not personal, Sonny. It’s strictly business.”
But marketers know the best businesses get personal. One-to-one personalization has been the vision for my whole career: a customer’s app, emails, and web pages that all know her, each one picking up where the last left off. The best teams, with the best data and technology, got close. Most of us got as far as a first name in the subject line.
Now AI makes that vision seem achievable. The CEO, the board, and colleagues who have never built a campaign watch a chatbot write an email in ten seconds and reasonably wonder why a campaign still takes a month. So they assume this should be simple now, and take fewer people. Those of us running marketing teams are living in the gap between that expectation and reality, and we need a way to explain why AI has, in some ways, made this harder.
Why is one-to-one personalization so hard?
My own experience running lifecycle programs looked nothing like that vision. More often, I found myself fighting upstream against disconnected tools and data, while my team struggled to segment customers at even the most basic level.
During my time at GitHub, our then-CEO announced one December that we were removing every non-essential cookie from the site. Developers loved it, and for them it was the right call. It also left marketing blind overnight, for longer than I’m willing to admit. At another company, I’m horrified to admit that we couldn’t even tell who was using our product.
And I'm not the only one. In Salesforce's 2026 State of Marketing survey of 4,450 marketers, 98% said they hit barriers to personalization, most often with data. Gartner's 2025 Marketing Technology Survey found marketers use 49% of what their stack can do.
One-to-one personalization does exist. But with tools and data in that state, it's rare across a whole business, and where it does happen, it tends to require a lot of time and human capacity.
Even with good data, someone has to build the message
And when the system did name the next-best message for a customer, someone still had to build it. The personalization execution gap is the distance between knowing the right message for a customer and being able to build, approve, and deliver it.
The gap is built into the math, because personalization work grows exponentially. Every new dimension multiplies everything that came before it. Take Marriott, which reports more than 30 brands in 143 countries and territories. Tailor one offer to each brand in each market, then build it for email, web, and app, and you have nearly 13,000 versions, before anyone mentions language or loyalty tier.
Teams grow by addition, one hire at a time. That is why nobody has hired their way to one-to-one.
And the math is getting worse, because everything upstream of marketing is speeding up. A 2026 National Bureau of Economic Research study of more than 500,000 developers found that coding agents increased software releases by 30%. Every release changes what the next message should say.
The same technology has changed the building part for marketing. AI agents, software that carries out tasks instead of only drafting text, can now assemble the page, the email, and every version of each inside the tools a team already owns. It won't clean up your data. But after all those years of pitches, personalization at scale can be built.
So the question has moved, from whether you can produce this much to whether you can do it without wrecking your brand.
Why AI makes this harder before it makes it easier
Most teams have already run this experiment, and the surveys show how it went:
- 75% of marketers are turning to AI to close their personalized content gap. Yet 78% still can't produce as much as they need, and 84% admit to running generic campaigns (Salesforce, 2026).
- 76% of marketing leaders spend at least three hours a week correcting AI output (Optimizely survey of 2,003 marketing leaders, 2026).
- One in four admit to knowingly publishing off-brand AI content when a deadline is close (same Optimizely survey).
That last number is what brand damage looks like in practice: a steady drip of almost-right, published because the review queue was built for handmade assets and nobody had the time.
And those hours spent correcting are a supervision tax: you hired an assistant (who is prone to hallucination) and became its proofreader.
I've paid that tax. I have spent longer playing whack-a-mole with an AI draft than it would have taken me to write the thing from scratch.
So here is what I'd tell a CEO who expects this to be simple. AI made the drafting cheap. It made everything around the drafting more work: more to check, more to approve, and more to keep current now that the product ships faster too. Until those parts change, more AI just means more to supervise.
What it takes: agents that build, people who decide, review built into the work
Three things have to change together. Change only one and you get the survey results above.
Technology: tools that build, alongside the tools that predict. The useful question for any AI tool is whether it produces the finished page, live where it belongs, or a suggestion someone still has to build. A general-purpose chatbot can write a decent email, but it doesn't know which claims legal approved last week and it can't publish into your systems, so a person carries its draft the rest of the way. An agent that works inside your content management system, email platform, and design system, building from a library of approved components, removes that queue. Forrester's 2027 predictions point the same way: start with agent workflows on the systems you already have, so migration is a later choice and not the price of entry.
People: marketers stop being the glue between tools. A former manager of mine used to tell us that “everyone needs to do a little ditch digging,” a reminder that parts of this job are not glamorous. Early-career marketers used to do most of it. But with taste and judgment being so critical to using AI well, teams are getting more senior. In the second quarter of 2026, entry-level marketing hiring fell 4% from a year earlier while director-level and above roles rose 17%, according to an analysis of more than 86,000 job listings.
The ditches are still there. The people who used to dig them aren't. So your highest earners are now being paid to copy and paste between marketing tools and AI chatbots.
Hand the digging to agents and marketers get back to the real job: understanding the customer, the market, and the business.
I don't think the answer is a set of new job titles. It's three places where human judgment stays, however much the agents take on:
- What each customer should hear, and why. That takes knowing the customer.
- What the brand will stand behind. Which claims, offers, and proof points are approved, and which have gone stale.
- What's ready for a customer to experience. To a customer, the pages and messages that reach them are the brand.
Today those calls sit with different people in different queues. Over time I expect them to collapse into one: a marketer who owns an audience end to end, the way a product manager owns a product, with agents doing the building and a shared system holding the brand's rules. Designing that system, and teaching the agents how your team works, is new work. For marketing operations professionals who got into the field to make great marketing possible, it's closer to the job they signed up for.
Process: approve the parts once, then check what gets assembled. This is the change that protects the brand. Today, a new email version for one region waits in the same legal queue as a product launch. In the new model, legal cleared that claim once, in March, and every message that uses it is cleared the moment it's assembled. Each assembled message then gets automated checks while it can still be fixed, and people keep the decision to release.
What it looks like when the building gets handed off
I run customer and community marketing at Gradial, an Agentic Marketing Platform. I'm trying to figure out how we make this sea change, alongside a lot of other marketers on the same journey. T-Mobile's digital operations team counted the hours on one large campaign with personalized experiences across web and app. It had historically taken 1,000 to 1,200 cumulative hours over about a month, most of it training, coordination, review cycles, and production queues.
In a pilot with us, agents did the assembly inside the stack T-Mobile already had. The same campaign took about 80 hours, with people concentrated on quality assurance. Nick Pappas, senior director of digital and web operations, said the cut in time to market was "opening up valuable capacity."
Notice what T-Mobile had before it changed anything: a number. The team knew what one campaign cost in hours. Most teams don't, and it's the easiest place to start.
Clock one campaign before you change anything
You can find out where you stand this week, with no pilot and no vendor, whatever state your data and tools are in. Pick one targeted campaign your team ran last quarter, even a simple one for a single segment, and trace it. You're looking for three numbers:
- Days from decision to live. From the day someone decided that audience should get that message to the day a customer saw it.
- Planned versus built. How many versions you wanted, and how many went live.
- The correction load. The hours spent fixing drafts, AI-written or not, before anything could go out.
If the first is measured in weeks, the gap in the second is wide, and the third is growing, a smarter model was never going to help. The work around it is the constraint.
Then pick the one workflow where those numbers look worst and change all three things there first. Put an agent on the assembly. Name who makes those three calls. Move review onto the parts. One workflow is enough to learn what the rest will take.
If you sign the budget, count the finished, approved experiences that reached customers this quarter. The number of AI tools, agents, or prompts in use tells you nothing.
So when does it get personal?
Here is the honest timeline.
- Now: personalized campaigns for every segment you serve, built in a fraction of the time, with a person approving what goes live.
- Later: every customer gets an experience built for them and nobody has to touch it. It won't arrive as a launch. It gets earned one workflow at a time, as each kind of work proves it can run without someone watching every step.
None of this fixes your data. That half is still yours to do, and the better it gets, the more of this you can use.
One of my other favorite movies happens to quote that Godfather line. In You've Got Mail, Tom Hanks's big-chain bookseller leans on it after putting Meg Ryan's little bookshop under. She asks him what's so wrong with being personal. He has no good answer, and neither does Michael Corleone. I've always believed personal is good business. For the first time, we can afford to build it.
So clock one campaign this week. Then make it personal.
Sources
- Salesforce, State of Marketing 2026, survey of 4,450 marketing professionals, fielded October to November 2025
- Optimizely research conducted by Savanta, 2,003 marketing leaders in seven countries, as reported by IT Brief (June 2026)
- National Bureau of Economic Research, Writing Code vs. Shipping Code (Working Paper 35275, 2026)
- Gradial customer story: T-Mobile
- Gartner 2025 Marketing Technology Survey, as reported by CMSWire (October 2026)
- Marriott International, 2025 growth announcement (brand and market counts)
- Forrester, Predictions 2027: Humans Reclaim Content Creation, Machines Dominate Content Consumption (October 30, 2026)
Taligence and Aspen Technology Labs, US marketing hiring report, second quarter 2026

Virginia Bryant
VP, Customer and Community Marketing, GradialVirginia Bryant is VP, Customer and Community Marketing at Gradial. She brings a practitioner’s perspective on lifecycle marketing, personalization, and the people, tools, and processes needed to deliver customer experiences.













