Customers Say They Hate Being Tracked — But Their Wallets Say Otherwise

Let me start with a set of contradictory numbers.

Segment (the company Twilio acquired) ran a survey:

• 83% of consumers say they're willing to share their data with brands in exchange for more personalized experiences.
• But at the same time, 48% of consumers say they worry brands will do things with that data they don't know about.

See that?

Customers shout "stop tracking me" with one hand and hand you their data with the other.

Both of these things are true at the same time.

Why?

Because what customers want isn't "no recommendations" — it's "recommend accurately, without making me feel like you're spying on me."

This is a delicate balance. Brands that nail it rake it in; brands that don't lose both the data and the customer.

1. Personalization Is No Longer a "Bonus" — It's the Baseline

A lot of companies are still agonizing over: "Should we even do AI personalization?"

Honestly, in 2026 that's not even a question anymore.

Accenture has a number: 91% of customers are more likely to buy from brands that "remember their preferences."

91%.

Do you know what that means? It means if you don't personalize, you can't even hold onto the remaining 9%.

McKinsey has another number: companies that use AI for personalization see sales rise by an average of more than 20%, and in some industries by 30%.

This isn't "icing on the cake." This is "make-or-break."

So why are so many companies still not doing it?

Because they're stuck on two problems: "I don't know where to start" and "I'm afraid of screwing it up."

Today I'm going to walk you through both, in depth.

2. The Three "Playbooks" of Personalization

AI personalization isn't one thing — it's three things. Mix them together and you'll screw it up for sure.

Playbook One: Predictive Personalization

Using past data to predict what the customer will do next.

The classic example is Netflix. They use a customer's viewing history to predict which customers are at risk of canceling their subscription, then push those people new shows they'll probably like. This single move keeps Netflix's annual churn down by 5% — which translates into $1 billion in revenue.

$1 billion, from a single churn prediction model.

The telecom industry uses it heavily too. Vodafone and AT&T use call records, payment behavior, and customer service interactions to predict which customers are about to leave, then precision-retain them, cutting churn by 40%.

The key to predictive personalization isn't how complex the model is — it's whether you have enough data dimensions. Purchase history alone isn't enough. You need to layer in customer service interactions, browsing behavior, social signals — the richer the dimensions, the more accurate the prediction.

Playbook Two: Dynamic Personalization

Adjusting content in real time based on what the customer is doing right now.

Spotify is the best example. You listened to jazz for 30 minutes this morning — by lunchtime it's pushing you a "jazz for afternoon focus" playlist. You listened to rock last night — the next day it's pushing you "5 new bands with a similar vibe."

433 million users, every single one with a different homepage.

Social media advertising runs on the same logic. The ad systems at Facebook and TikTok (Douyin) take the micro-actions you make — likes, comments, shares, how long you linger — and adjust the ad they serve you the next second, in real time.

The core of dynamic personalization is speed. The customer acts this second, your system reacts the next. Miss a beat, and personalization becomes "stale recommendation."

Playbook Three: Recommendation Engines

Recommending things a user might like, based on their preferences.

There are three core techniques behind recommendation engines:

1. Collaborative filtering — "people similar to you liked X, so you'll probably like X too." 80% of Netflix's viewing hours come from this mechanism.
2. Content-based filtering — "you liked the features of A, so A2 and A3 are probably up your alley too." Spotify uses this, recommending new songs based on a track's tempo and style.
3. Hybrid models — combining the first two. Amazon is the pioneer of hybrid models — 35% of its sales come from its recommendation engine.

Note that these three aren't "the newer, the better." Collaborative filtering has been around for 20 years and is still the mainstream — because it's simple, effective, and explainable.

Don't get seduced by the "deep-learning large model" hype — sometimes a plain-vanilla collaborative filter is all you need.

3. Four Industry Case Studies That'll Make You Sit Up

Let me walk you through four real cases, and you'll see how much money is on the table.

Case 1: Nike Fit

Nike built an AI foot-scanning feature — you snap a photo of your foot with your phone, the AI measures your exact size, and recommends the best-fitting shoes.

Conversion on digital channels jumped 40%.

Why so effective? Because the single biggest pain point in buying shoes is "the size isn't right." Once AI solves that, conversion takes off on its own.

Case 2: JPMorgan Chase's COiN System

JPMorgan Chase uses AI to analyze legal documents and customer transaction data, giving customers personalized investment advice and loan options.

McKinsey predicts that this kind of AI-driven financial personalization will add $170 billion in profit growth to the banking industry over the next four years.

$170 billion — an entire industry's revolution.

Case 3: Hilton's Connected Room

Hilton lets guests use their phones to control the room's temperature, lighting, and TV, while AI quietly records every guest's preferences in the background. The next time you check into any Hilton, the room's parameters are already dialed in.

This is the highest form of "cross-channel personalization" — your preferences travel with you, not the other way around.

Case 4: Bank of America's Virtual Assistant, Erica

Erica is an AI customer service agent. In 2023, customers invoked Erica 673 million times, up 28% year over year. It gives customers spending summaries, payment reminders, financial advice — all of it personalized.

673 million times means customers are genuinely willing to deal with AI, as long as the AI is actually useful.

4. The "Personalization Appetite" of Different Generations

Here's something counterintuitive — not everyone is sold on "personalization."

The reactions across generations are night and day:

Gen Z (born 1997–2012): 74% expect personalized products, far higher than any other group. They grew up on TikTok and Instagram, and have a natural acceptance of algorithms "getting them."

Millennials (born 1981–1996): 70% are willing to let retailers track their browsing behavior in exchange for a better experience. This group is the biggest revenue segment for AI personalization.

Baby Boomers (born 1946–1964): They have low expectations for personalization overall, but they're especially receptive to "personalized financial advice" — retirement planning and investment portfolios from AI are things they actually need.

In other words, you can't win them all with one personalization playbook. Push short videos and gamified content to Gen Z; push conservative financial planning to Baby Boomers — personalization itself has to be personalized.

5. The Real Ceiling on Personalization: Trust

Having covered playbooks, cases, and audiences, let's go back to that contradiction from the opening —

Customers are willing to give you data, but they're also terrified of it being abused.

This is the real ceiling on personalization.

Segment's survey shows: 60% of consumers become repeat customers after having "one good personalization experience." That's up 16 percentage points from 44% the year before.

But flip it around — a single incident of "data being abused" and the customer will never trust you again.

What counts as "data being abused"?

• You sold their data to a third party without their explicit consent.
• You used their data to make a recommendation that felt "offensive" (like pushing maternity clothes to a woman who just had a miscarriage).
• Your data got breached, and hackers walked off with it.

Any one of these three happens, and every bit of personalization work you've done up to that point zeroes out.

So the bedrock of personalization isn't AI — it's trust.

AI is the engine, trust is the steering wheel. No matter how powerful the engine, one wrong turn of the wheel and the car flips.

6. Three Judgments for You

Finally, three plain judgments to leave you with:

1. Personalization isn't "should we do it" — it's "we need to do it now."

91% of customers have already voted with their feet. Hesitate a year, and you lose 91% of the opportunity.

2. Do "small and accurate" first, then "big and comprehensive."

Pick one concrete scenario (say, cart-abandonment recovery emails), wire the data together, tune the model, validate the results — getting one small closed loop working is a hundred times better than bolting on ten channels all at once.

3. Make "customers can take their data back at any time" a product standard.

Regulations like GDPR (General Data Protection Regulation) and PIPL (Personal Information Protection Law, China's data-protection statute) are all reinforcing "data portability and the right to deletion." Doing this proactively, ahead of the curve, isn't a compliance burden — it's a trust asset.

When customers see "this company actually lets me export my data with one click and delete my account with one click," they'll be more willing to hand their data over to you.

Transparency is the most powerful form of personalization in 2026.

Back to that contradiction from the opening.

83% of customers are willing to share their data. 48% worry about it being abused.

Whoever can turn that 48% worry into 0, wins the 83% of the market.

That's the ultimate competition in personalization.