Marketers, Someone Just Walked Off With Your Holy Grail

Recently, I scrolled past a news item.

There's a company that sells crafts — Michaels Stores. Until recently, only 20% of their email marketing was "personalized." The other 80%? Mass-blast. A thousand people receiving the same email that opened with "Dear Customer."

Then they did one thing.

A short while later, the share of personalized emails went from 20% to 95%. SMS click-through rates climbed 41%. Email click-through rates climbed 25%.

Same team. Same budget. They'd just started using generative AI to write copy and make segmentation decisions.

My first thought: this is bigger than I realized.

Let's Run the Numbers

McKinsey put out a report estimating that generative AI could contribute $4.4 trillion in productivity to the global economy every year. 4.4 trillion.

Three quarters of that value is concentrated in four functions: software engineering, customer operations, product R&D, and marketing and sales.

Marketing alone, thanks to the generative AI boost, could save 5% to 15% of total marketing spend. That works out to roughly $463 billion a year.

463 billion.

Think about that. Money either saved, or freed up for more creative work. Either way, whoever gets there first, wins.

What Does "Actually Using It" Look Like?

A lot of companies say they're "using AI." But look closely and it's a few people on the marketing team who occasionally open ChatGPT, type a prompt or two, generate a handful of images. Is that "using it"?

Genuinely using it comes in three tiers. Each one deeper than the last.

Tier 1: Off-the-shelf adoption. Take publicly available models, plug them into your existing workflow. Let AI help write copy, generate images, handle segmentation, reply to customer emails. The hallmark of this tier: you see results immediately — no waiting years for payoff.

One example. Designers at Mattel working on Hot Wheels toy cars used to spend days grinding out a single product concept visual. Now, with AI, the same time produces four times as many visuals. Kellogg's scans the web for trending recipes featuring breakfast cereal, then uses that data to drive social campaigns. L'Oréal is analyzing millions of online reviews, images, and videos to spot the direction of the next breakout product.

Or take a direct-to-consumer retailer that handed its customer service tickets to AI. The AI looks up information in the background, modifies orders, and replies to customers in the brand's own voice. The result? First-response time dropped by more than 80%, and the average ticket took 4 fewer minutes to resolve. The support team finally had time for the genuinely complex issues that require a human conversation.

This tier is open to anyone. Any company can play. But here's the catch: if you can do it, so can everyone else. Very quickly this stops being an advantage and becomes table stakes.

Tier 2: Custom-built. This is where the real gap opens up.

What does "custom" mean? You take public open-source models and train them on your own data. Your brand guidelines, the last ten years of your marketing creative, your customer profiles — all fed in. What comes out is a model that knows nothing but your company.

Two stories stuck with me.

A European telecom company. They used to message customers in only four broad categories — the entire country sorted into four versions of copy. Tight budget, small team, couldn't produce more. Worse, their country has several dialects, and messages sent to customers who didn't speak the dominant dialect converted terribly.

Then they built a generative AI engine that carved customers into 150 micro-segments. Copy for each segment was generated based on region, dialect, and demographics. Messages and images went through GPT-4 and Dall-E, finished assets were piped straight into the email system via API, and a machine learning model decided: when should this customer receive it, through which channel, and what product should we lead with?

The result: response rates up 40%, deployment costs down 25%.

The second: an Asian beverage company that wanted to break into the European market. The traditional approach — just brainstorming a new product aimed at European consumers — could take a full year.

Here's how they did it.

First, they fed ChatGPT a batch of European consumer behavior data (anonymized), and had it analyze what flavors Europeans actually like. What used to take a week was done in a day. Then they layered on traditional interviews and diary studies to deepen the insights.

Product concept visuals? A single industrial designer used to spend 7 to 10 days on one high-fidelity concept. Using text-to-image tools, they knocked out 30 in a day. They took those visuals to consumers for testing, and the feedback was remarkably candid — because the images looked like real, market-ready products.

In the end: a year's worth of work, done in a month.

When I read that, my stomach dropped. This kind of efficiency gap isn't a 10% or 20% optimization. It's an order-of-magnitude leap.

Tier 3: Start over. The first two tiers are still patching the existing workflow. Tier 3 means redesigning the entire marketing function around AI. From writing copy to doing research, from customer outreach to campaign design — every single task starts with one question: "If AI is my starting point, how should this even be done?"

This tier is still early days. Nobody can claim they've made it all the way through. But the direction is clear: the truly distinctive, signature customer experiences will grow out of here. A cosmetics brand builds a chatbot that asks about your skin condition today and your goals, then customizes an entire skincare routine. A grocery delivery service generates next week's menu and shopping list automatically, based on what the family ate this week and what they're avoiding.

These things weren't impossible before. They were just unscalable. Now they are.

But This Isn't All Upside

Anything that can create enormous value carries enormous risk.

Generative AI has a known flaw called "hallucination." It will, with total confidence, fabricate an answer that sounds perfectly plausible and is completely made up. It struggles with tasks requiring rigorous numerical reasoning, and it's not suited for high-stakes decisions or tightly regulated environments.

So before you deploy AI, build guardrails.

First, assign clear ownership. Designate a person, establish a technology oversight committee — don't leave AI in a vacuum where "everyone uses it, nobody actually owns it."

Second, add a human review step for anything going directly to customers. Don't let AI make public statements on its own.

Third, define the scope of what AI is allowed to discuss. Don't let it engage with anything and everything. Draw the boundaries.

The European telecom company's approach is worth learning from. They had humans watching the entire pipeline. They deliberately limited the number of copy variations and the degree of personalization. They got risk, ethics, and privacy under control first — then talked about scale.

So, How Do You Start?

Don't try to boil the ocean. Trying everything at once is the most common mistake in the AI era. This gets piloted, that gets piloted, and in the end nothing goes deep. Money spent, lessons scattered.

Focus your firepower on two or three use cases. Pick ones where off-the-shelf tools can deliver visible results immediately. Build something you can point to. Then learn from the process: what kind of people, what capabilities, what operating model do you need to scale this up?

Thinking further ahead, move in three phases:

• First six weeks: Map a pilot roadmap. Pick use cases, audit your tech stack, find the right people, identify risks.
• First 90 days: Stand up an "AI war room." Lock down priority use cases, pipe in the data, put guardrails in place, and run a few audits to confirm the AI is being used responsibly.
• First six months: Start building the long-term transformation strategy. Measure impact, manage change, tune the models, and integrate your AI efforts with your existing marketing technology.

After a few months, you should have a handful of use cases you can show off — the kind that make people lean in and want to learn from you.

A Final Word

The "Holy Grail" marketers have always chased — the right message, at the right time, delivered the right way, to the right person, as one seamless whole — used to be out of reach. Not because nobody wanted it, but because it couldn't be done at scale. Limited by creative capacity. Limited by data analysis. Limited by time.

Now, those limits are loosening.

After reading through the McKinsey report, my strongest takeaway comes down to one line: the change is already underway, and sitting still means getting left behind. Those 40% response-rate jumps, those four-fold concept visuals, those year-long roadmaps compressed into a month — behind every one of them are real companies, real people, already using new tools to redefine what "marketing" means.

AI won't do your marketing for you. But the people who know how to use AI will replace the people who don't.

On this road, whoever can bind the tools deeply to their own data, processes, and judgment — that's who gets to walk away with the Grail.

May you hold on tight.