AI Personalization at Scale: Retail Playbook A design for relevant, consented, identity-clean recommendations. Not a promise of “one-to-one personalization,” not an invented lift, and not a product bake-off. Edition: November 2026. 20 pages. Written for: Retail, e-commerce, marketing, data and technology leaders, and the sponsors who have to fund a personalization case. Canonical URL: https://asymilinkmeta.com/resources/retail-personalization-guide Written for the search query: retail AI personalization playbook Key takeaways - Fourteen design tests for relevant, consented, identity-clean recommendations. - Consent is a design input, not a later preference. Clean identity is a design, not a slogan. - This playbook will not promise “one-to-one personalization at scale,” and it will not invent a lift, a conversion gain, or a sector benchmark. - The illustrative retailer in the document is a picture, not a client case study, and not a sample of the sector. What's in this playbook This 20-page playbook is a design method for retail, e-commerce, marketing, data and technology leaders who have to fund a personalization case. The library has sometimes described it as a promise of “one-to-one personalization at scale,” with an invented lift. It will not make that promise, and it will not invent a conversion gain, a revenue lift, or a sector benchmark. What it is, instead, is fourteen design tests to apply before a personalization case is funded: a way to treat consent as a design input, identity as a design (clean, honest, and no broader than the work), and a recommendation as something a person can stand behind, refuse, or unwind. A relevant recommendation, on a consented and identity-clean foundation, is the job. A slogan is not. It is written for the people who own the customer relationship and the systems it runs on, and for the sponsors who have to fund the work. The illustrative retailer in the document is a picture of how the tests would read, today. It is labeled illustrative. It is not a client case study, and it is not a sample of the sector. A companion workbook holds the fourteen tests and a register you can apply. A completed workbook is not a finding that a program is “personalization at scale,” and it is not a certificate. Consent, identity, and an honest first case Consent is a design input, not a later preference, and not a banner that is assumed to cover every use. What was consented to, for which use, and what happens when it is withdrawn, are written down before a recommendation is funded. This playbook is not legal advice, and it will not recite a date by which a rule applies. Counsel is asked in time, with a written answer. A missing answer is a reason to pause, not permission. Clean identity is a design. Who the recommendation is about, how that identity was joined, and what is allowed to disagree with it, are named. A join that is broader than the work, or that silently merges people who should stay apart, is a failed test. Permissions travel with the profile: a system cannot use, for a recommendation, what the person could not be shown, or what they have asked to be withdrawn. The honest first case is bounded: a named channel, a named catalog or category, a frozen way to examine whether a recommendation is relevant, a person who can stand behind what is shown, and a way to unwind a bad day. “One-to-one at scale, across every channel, unsupervised,” is not a first case this playbook will recommend. The data-strategy guide, the CRM checklist and the workflow playbook in this library are companions. This playbook will not stand in for them, and it will not invent a lift to close a business case. Fourteen design tests (the shape of the inquiry) The full wording of each test is in the PDF and the workbook. The list below is the shape of the inquiry, drawn from the document, not a substitute for opening the guide. Consent. Consent is a design input. What was consented to, for which use, and what happens when it is withdrawn, are written down before a recommendation is funded. Counsel is asked in time. Identity. Clean identity is a design. Who the recommendation is about, how that identity was joined, and what is allowed to disagree with it, are named. A silent merge of people who should stay apart is a failed test. Permissions. Permissions travel with the profile. A system cannot use, for a recommendation, what the person could not be shown, or what they have asked to be withdrawn. Relevance, measured. A frozen way to examine whether a recommendation is relevant, with a person who can stand behind what is shown. This playbook will not invent a lift. A first case that is bounded. A named channel, a named catalog or category, a person on the action that matters, and a way to unwind a bad day. “One-to-one at scale, unsupervised,” is not a first case this playbook will recommend. An operator, and a switch-off. A named operator who is not only the builder, a trail that can reconstruct a day, and a way to stop the system without a project. The complete fourteen tests are in the guide and the workbook. A completed workbook is not a finding that a program is “personalization at scale.” What this playbook will not do The cover and the opening are explicit about the limits. They are part of the method. Not a promise. This playbook will not promise “one-to-one personalization at scale.” A relevant, consented, identity-clean recommendation is the job. A slogan is not. Not an invented lift. It will not invent a conversion gain, a revenue lift, or a sector benchmark to close a business case. The financial method, where a case needs one, is the ROI guide in this library. Not a product bake-off. It will not rank personalization platforms by name, and it will not declare a vendor “best.” Frequently asked questions Q: Does this playbook promise one-to-one personalization at scale? A: No. It is a design for relevant, consented, identity-clean recommendations. It will not make a “one-to-one at scale” promise, and it will not invent a lift. Q: Is this legal advice about consent? A: No. Consent is a design input. Counsel is asked in time, with a written answer. This playbook will not close that question, and it will not recite a date. Q: What is an honest first case? A: A named channel, a named catalog or category, a frozen way to examine relevance, a person who can stand behind what is shown, and a way to unwind a bad day. “One-to-one at scale, across every channel, unsupervised,” is not a first case this playbook will recommend. Q: What is in the companion workbook? A: The fourteen tests as a register you can apply to a case of your own. A completed workbook is not a finding that a program is “personalization at scale.” This page is the public summary. The full PDF and companion workbook are available through the request form on the page. There is no public file URL that bypasses that form.