Your Customers Are a Mystery. AI Gets Them Anyway

A while back, a friend of mine who runs an e-commerce business vented to me.

He said: I send the same email to a hundred thousand customers and get a 3% open rate. I hired three marketing ops people who spend every day manually building segments and rewriting copy — still 3%. They're nearly worked to death, and we haven't earned an extra cent.

I told him: your problem isn't effort. Your problem is that you're doing a machine's job by hand.

He froze.

What does "doing a machine's job by hand" mean? It means your customers' behavior has already been digitized down to every click, every browse, every abandoned cart — but your judgment of "what should we push to this person" still depends on an operator's gut feeling.

No amount of hard work fills that hole.

Today I want to talk with you about what AI has actually rebuilt marketing into. Three parts: personalization, CRM, and automation. Let's take them one at a time.

First, a sobering number

There was a set of survey data from 2025, and it left me with mixed feelings.

IBM's report says that companies prioritizing AI for customer experience see revenue growth of roughly three times their peers. In HubSpot's research, 94% of marketers say personalization directly drives sales.

Sounds like everyone gets it, right?

But Gartner has another number: 63% of marketing leaders say personalization is a challenge, and only 17% have actually deployed AI at scale.

94% think it matters. 17% actually do it.

That gap, right there, is every marketer's opportunity today. And every marketer's threat.

Whoever turns "we know it matters" into "we're actually doing it" first, leaves the competition behind.

What is AI personalization?

Let's unpack the concept first.

What is AI-driven personalization? Put plainly: machine learning reads through a person's browsing behavior, purchase history, and CRM profile, then makes the call for you — what is this person most likely to pay for, right now?

Netflix and Amazon have been doing this for over a decade. You open Netflix and the homepage is different for every single person. That's not an editor's arrangement — it's a stack of machine learning models computing it. Netflix itself has said the bulk of content consumption on the platform is driven by this recommendation engine.

Back when you had a few thousand customers, the shop clerk could remember every regular's preferences. Now you have hundreds of thousands of customers. Who can remember all that?

AI can.

Here's a fun example. Sephora's Virtual Artist uses augmented reality plus machine learning to let customers try makeup virtually on their phones. You try three lipsticks, and it instantly knows which matching lip liner to recommend. And then, average order value goes up.

That's not magic. That's computing power.

And this has already been productized. Platforms like Dynamic Yield and Adobe Target can serve the same landing page with different images and different promotions to different visitors. Optimizely and Evergage do real-time content swapping. Retail recommendation engines, like Amazon's, push the next item the moment they've read your browsing history.

One small e-commerce case stuck with me: after plugging in an AI recommendation engine, average order value rose 25%, and the customer service chatbot resolved about 60% of inquiries on the spot. There's also a restaurant chain that used machine learning to identify which customers might try a new dish, then sent targeted coupons with photos of the dish — redemption rates 20% higher than a blanket blast.

See, it's not that small companies can't do this. It's that many of them don't know they can.

McKinsey did the math: 76% of customers prefer merchants that offer personalized service. Do it well, and revenue can rise 10% to 15% — up to 25% for some.

Personalization stopped being a nice-to-have a long time ago. It's a variable in the revenue equation.

CRM: From address book to strategist

Second part: CRM.

What's CRM? A customer relationship management system. It used to be a fancy address book: the customer's name, phone number, what you discussed last time.

Now, Salesforce's Einstein and Microsoft Dynamics 365's AI have turned it into a strategist.

What does a strategist do? Three things.

First, rank the troops. Machine learning scores your sales leads — which customer is most likely to close, at a glance. Research suggests lead scoring can lift conversion rates by up to 30%. Sales stops casting nets and goes straight after the hottest prospects.

Second, read the battlefield. AI predicts which deals will close and when — and the predictions get sharper with use. The system also nudges the salesperson: for this customer, here's the next product to pitch, here's the line to say.

Third, sweep the floors. It auto-fills company details and social profiles, deduplicates, cleans data. The form-filling salespeople hate most? The machine takes it.

There's a number from Salesforce's small and medium business survey: among SMBs using AI, 87% achieved scaled growth, and 86% saw margin improvement. Another 91% of SMBs said revenue rose after adopting AI.

75% of SMBs are already experimenting with AI. And fast-growing companies are noticeably more willing to invest in AI than companies that are struggling.

Coincidence? You tell me.

Are growing companies using AI, or are AI-using companies growing? I lean toward the latter carrying more weight.

Ads and automation: you set the direction, the machine hits the gas

Third part: ad buying and automation.

By 2025, the ad platforms of Google, Meta, LinkedIn, and Amazon have gone almost entirely AI. With Google's Performance Max, you give it a goal and it allocates budget across channels on its own. Meta's automated ads have machine learning generating different versions of creative, then pouring more spend into whichever performs best.

What did ad buying used to rely on? Media buyers pulling all-nighters watching dashboards, adjusting bids on gut and experience. Now, platforms like Acquisio and Smartly.io do cross-channel real-time bidding; tools like Copy.ai and Jasper generate ad copy variants outright; AI can run thousands of A/B tests at once, day and night.

A local restaurant's campaign and a multinational brand's campaign might be running on the same underlying logic.

Ad buying is turning from a craft of experience into a discipline of engineering.

Same story in marketing automation. HubSpot, Marketo, ActiveCampaign, Mailchimp — these platforms are all embedding AI. GetResponse has a "Perfect Timing" feature that predicts when each recipient is most likely to open an email, and sends at exactly that moment. ManyChat and Drift chatbots catch leads on landing pages and route them automatically based on the conversation.

One report estimated AI can compress task completion time by about 40%, and cut marketing labor costs by 25% to 30%.

Three people's work, done by two, done better.

Small teams, especially, should be grinning.

But hold off on that shopping cart

By this point you may have already opened a shopping cart. Wait a second.

I've seen too many companies buy a pile of tools that end up gathering dust. The problem isn't the tools. It's the preparation.

Here are the steps I sincerely recommend.

First, set the goal. Not "we need to use AI," but "qualified leads up 30% next quarter." Without a clear goal, whatever you buy is wrong.

Second, connect your data. This is the most fatal step. If your CRM, web analytics, email system, and sales system each live in their own silo, AI is blind. Salesforce surveyed this: growing SMBs are twice as likely to have a well-integrated tech stack — 66% versus 32%.

Third, run small experiments. Pick one customer segment, run a round of AI-generated email sequences, and compare against a manual control group. If it lifts, scale it.

Fourth, don't remove the humans. AI handles the repetitive work: analyzing data, segmenting, drafting. But brand voice and major calls must be made by humans. However fast the machine writes, if it sends the wrong tone to a customer, the loss is yours.

Fifth, transparency and restraint. If you use AI for personalization, let customers know. Keep data compliant. And don't overdo it — personalize past the point of comfort and customers walk. Know one person too intimately and they feel surveilled, not served.

AI handles the scale. Humans handle the warmth. Miss either one, and the business doesn't last.

Looking ahead

So what happens next?

My read: a few directions. AI Agents will shift from "executor" to "manager": planning the marketing calendar themselves, generating a full campaign's creative themselves, renegotiating budgets mid-flight. Generative AI won't stop at copy — images, video ads, interactive experiences all generated on demand. Voice search and visual search will become new battlegrounds. Privacy-preserving personalization will rise, with more computing done on-device.

But remember what HubSpot said: success won't come from stacking more technology — it'll come from streamlining it, so marketers return to creativity and high-value work.

Back to that friend from the beginning.

He eventually did one thing: switched his email system to a platform with AI segmentation and send-time optimization, and ran it for three months as a trial. Last month he told me his open rate had gone from 3% to 11%.

Without hiring a single extra person.

See, the customers didn't change. The market didn't change. What changed is that he finally let machines do the machine's work, and let humans come back to the human's work.

It's a step you should take too.