AI Personalization: Why 60% of Consumers Want Shared Control

AI personalization uses customer data and machine learning to choose what customers see online, including a brand’s products, offers, the team’s lovingly crafted content, and what channel and time they arrive in. The more of those choices AI personalization is allowed to make, the more important it is for customers to be able to correct things when the AI gets it wrong.

SAP’s Customer Loyalty Index 2026 asked consumers how much control they want over AI. Just 4% want AI to act fully autonomously, and 60% want control shared between themselves and AI. Most consumers “still want visibility, oversight, and the ability to jump in when AI gets it ‘wrong.'”

The potential of AI personalization is naturally shadowed by its potential pitfalls. 

For example, most online grocers’ apps build a customer’s provisional weekly order based on what they’ve previously bought. Let’s say that one week the grocer’s AI adds a chocolate cake mix a customer bought once for her son’s birthday. She deletes it from the order on Tuesday without giving it much thought – however, on Thursday she opens a delightfully chirpy email from the grocer with a huge image of a moist chocolate fudge cake and her cake mix listed right under it. 

Her reply to that email: “I literally took this out.”

The app would have recorded her correction, but the email’s recommendations came from a customer profile that the correction wouldn’t have been able to reach to edit.

Applied to AI personalization, the standard I’d hold any AI-led customer journey to would be to give the shared control people want by letting AI make everyday choices for them, but giving them options to correct the AI – and for those corrections to make it back, cross-channel, to the central customer profile. Let’s explore how to do that.

AI personalization is a branch of personalized marketing in which a model makes the per-customer choices a marketer would otherwise write as rules, and predictive segmentation is one of its most common forms.

Rule-based personalization has traditionally worked from instructions a marketer writes for a segment: for example, customers who bought running shoes get an email about socks two weeks later, which applies whether they’re training for a marathon or they bought the shoes to walk the dog. 

With AI personalization, every customer gets a score for each product, offer and channel, so for the dog walker who has also looked at a waterproof jacket three times since buying the shoes, the AI has access to that data, cross-channel, so it can predict that a special offer on the jacket will beat the more generic socks offer. 

That’s how AI personalization improves customer engagement: each customer gets messages that match what they’ve bought, browsed and told you, at the moment they’re most likely to respond.

One-to-one personalization at scale needs AI, because a team can write rules for twenty segments, but not for two million customers. Your team sets the goals and the guardrails – which products to promote, how often to contact each customer, who to leave alone – and the AI makes the per-customer choices inside them. 

Some campaigns should stay rule-based, of course: a consent rule or a welcome series should work the same way for every customer.

AI-powered personalization turns what you know about each customer into a choice for that customer. It runs on seven building blocks, from the data you collect to the journeys that deliver each choice:

• Customer profiles: who the customer is and what they’ve told you and agreed to, including consent and channel preferences
• Behavioral data: browsing, searches and engagement with your emails and app
• Transactional data: orders, returns, loyalty activity and service history
• Real-time interactions: what the customer is doing now, such as viewing a product or leaving a cart
• Predictive analytics: a score for what each customer is likely to do next, such as buy again or stop buying
• AI decisioning: the product, content, offer, channel and send time chosen for each customer, within the limits your team sets
• Marketing automation: the journeys that deliver each choice in every channel

If you run a CRM program, you probably already hold customer profiles, behavioral data and most of your transactional data. The usual gaps are real-time interactions from the app and the store, service history, and consent preferences held in the profile the AI works from.

In SAP Engagement Cloud, Personalization brings demographic, sales, support and behavioral first-party data into one customer profile. Using product affinity, channel and purchase predictions, the AI recommends products, content and offers for each customer and keeps them consistent across channels.

Most consumers want to share control with AI. Some customers would rather check every substitution in their grocery order themselves, and a small minority are happy for the app to build the order and place it without them, but in SAP’s Customer Loyalty Index 2026, most consumers want something in between. 

On what we call the AI control spectrum, 60% of consumers want shared control between themselves and AI, and 16% want AI to lead, with human oversight. At the ends, 20% want full human control and 4% want AI to act fully autonomously.

SAP’s report sums it up: “Loyalty lives in the shared-control zone.”

Customers don’t want to step in all the time, either. Our research shows half of consumers would rather a brand “just gets it right” without them needing to engage. For AI personalization, I’d read the two findings together: let the AI handle the everyday choices without asking, and give customers a clear way to step in when it gets one wrong.

A parent’s monthly diaper subscription is due to ship on Friday, still at size 3 because that’s what the last six orders were. The baby moved up a size last week. Wednesday’s reminder email has a button to change the size before the order ships, so the parent fixes it in ten seconds instead of sending back a box they can’t use.

In a journey with shared control, the AI makes the everyday choices and the customer can change any of them. The AI recommends products and the customer chooses what goes in the cart, the AI picks content within the preferences the customer has set, and the AI times each send within the channels the customer has agreed to. When a fashion customer taps “not for me” on a pair of wide-leg jeans in the app, the jeans should be gone from next week’s email too.

Customers want two safeguards in particular. In SAP’s research, 56% of consumers want to be able to reverse AI decisions, and 48% want human intervention to be available.

Reversal is something you design into each journey, and it matters more as AI agents take on more of the work, from building a cart to placing a reorder. I’d build the undo on automated orders first, because a wrong automated order is the mistake customers notice soonest.

An electronics retailer’s AI builds an accessory bundle for a customer who’s just bought a new phone, and the charger in it has the wrong connector. She opens the chat, and the service agent who picks it up can see why the bundle was built: it was matched to the phone she bought two years ago, which is still on her profile. The agent swaps the charger before she’s finished typing and updates the phone on her profile, so her next recommendation fits the phone she owns now.

What she’ll remember is how fast a person fixed the AI’s mistake. Consumers rank that kind of help first: in the Customer Loyalty Index 2026, human support if something goes wrong tops what builds trust in AI, at 42%, ahead of the ability to review or override decisions, at 36%.

These are the changes I’d make to an AI-led journey for each of the five trust factors in SAP’s research:

• Human support: Give a real person in the team responsibility for the channel the customer is already using, and give that person the AI’s recommendation and the order behind it, so the customer doesn’t have to explain everything from the start.
• Review and override: Make an override update the customer’s profile as well as the order in front of you. If the agent had swapped the charger and left the old phone on the profile, the next bundle would have had the wrong connector too.
• Privacy assurances: Tell customers which of their data the AI uses to personalize what they see.
• Clear explanations: Give a one-line reason the customer would understand, such as “Because you bought this phone.”
• Proven accuracy: Track which AI recommendations customers reverse most often, and fix the causes.

The Customer Loyalty Index 2026 advises brands to “connect AI assistants across your operational and customer data.” At the electronics retailer, that connection is what let the agent fix the profile as well as the bundle, because the order, the phone on the profile and the reason for the recommendation were all in front of her.

Personalization runs on signals such as: 

• Purchase history
• Browsing
• Engagement
• Loyalty activity
• Product preferences
• Service interactions 
• Delivery preferences

Most of those signals are operational data, held in order, loyalty, delivery and service systems that the marketing team doesn’t usually have access to.

Past purchases and order history top the list of what shoppers want an AI assistant to remember, at 32%.

Customers happily hand those signals over as part of a value exchange that makes sense for them. Asked what they expect from brands in return for their loyalty, consumers in the Customer Loyalty Index 2026 name better deals, great service and seamless experiences, and 23% name “more personalized experiences.” 

How much data customers are willing to share depends on trust. Separately, 23% of consumers say they’re more loyal to brands they trust with their data, and 30% name irresponsible data use as something that breaks their loyalty.

Consent and service records belong in the same profile as the order history. An open complaint about a product should stop the AI recommending that product, and your team can only make that happen when the service record is held in the profile the AI works from.

SAP Engagement Cloud’s Customer Data connects customer data across CX and operational systems, such as your ERP, supply chain or BI tools, and brings consent preferences, purchase history and loyalty status into the customer profile alongside web and channel engagement. With the order history and the profile in one place, your team can set recommendations to work from the corrected profile, instead of fixing each one as it goes out.

These are the eight checks I’d run on every AI-led journey, to find out whether customers can step in when the AI gets something wrong:

• Customer value: Can you name what the customer gets from the journey, such as a faster reorder or fewer emails they’ll ignore?
• Meaningful choices: If a customer opts out of SMS, is SMS removed from the channels the AI can use for them?
• Reversibility: Can a customer undo an automated reorder before it ships?
• Human access: When an AI-led journey goes wrong, can the customer reach a person in the same channel, and can that person see what the AI recommended and why?
• Explanation: Can a customer find out why they got a recommendation, in one line they’d understand?
• Trusted data: Does your preference center write to the same profile the AI works from?
• Learning from behavior: If a customer rejects a recommendation in the app, will the next email leave it out, and who on your team reviews the AI recommendations customers reject most often?
• Outcomes beyond clicks: Do you report retention and repeat purchase for AI-led journeys alongside opens and clicks?

Each “no” points to a journey your team needs to change. Most of those fixes need data work as well, because the app, the email, the service team and the preference center all have to work from the same customer profile.

Over the next few years, I expect AI to do more of the anticipating and acting for customers, from building the weekly order to timing the reorder before the customer asks. I also expect customers to keep wanting what they want now: to see what the AI did, and to step in when it’s wrong.

In SAP Engagement Cloud, AI Marketing brings predictive segments, product recommendations and send-time optimization to email, web and mobile, and Customer Data keeps consent preferences in the customer profile with purchase history and loyalty status.

Back at the online grocer, the cake mix comes out of the app order on Tuesday, and the correction lands in the same profile the email’s recommendations come from. Thursday’s recipe email opens on the lasagne she’s bought the ingredients for three weeks running, and the grocer’s reply inbox stays peacefully empty, for now.