The handbook · Working draft
The AI Personalization Handbook
A practitioner's reference for AI personalization, from the first decision to production.
Every leader knows personalization pays. So why does most of it still start with “Hi {first_name}”?
The case is familiar: it lifts conversion, lowers acquisition cost, heads off churn, opens upsell and earns loyalty and referrals. Customers ask for it too. In McKinsey's 2021 survey, 71% of consumers said they expect it, and 76% said they are frustrated when they don't get it.
Executives know the value, and the data is already there. Companies know who is renewing, who opened three tickets this week and who just hired a new ops lead, then send all of them the same email. Most companies are not short of data. They are short of ways to use it for the customer.
This is a working draft, written for you and for your AI. I'm revising every chapter in the open, so expect changes.
Why this handbook
Written from building, not advising
I build products from scratch and lead the teams that ship them. Personalization runs through many of them, and I kept hitting the same three gaps. Between what a company knows about a customer and what it says to them. Between a good idea on a slide and a system that runs every day. And between executives who know what personalization is worth and anyone who owns it: nobody whose job is to take it from a line in the plan to personalization pipelines running in production. The market has not closed those gaps yet. Whoever does will find growth sitting untapped in the customer base they already have.
So this handbook goes past the case for personalization. It covers how to think about it, design it, build it and operate it: the concepts, the practices that hold up in production, the options and where each one fits, and the resources to build with. It is a practitioner's view, not a deck of survey charts. Read a chapter, then act on it.
It has two readers: you and your AI. Every chapter ends with a structured block your assistant can reason with, and I'm adding code and docs so it can help you apply the handbook to your own company, not just summarize it.
How to read it
- If you leadEvery chapter opens with the short answer: what to decide, what it costs, what to ask your team. You can stop there.
- If you buildThen it goes deep: the mechanism, the patterns, the failure story, a build checklist and the metrics to watch.
- If you work with an AIEach chapter ends with a structured block you can paste into your assistant, so it reasons with the same concepts.
Contents
Part I
The Case
- 01What Personalization Is, and What It Is WorthWhat is personalization, and why should a leader care now?23 min read
- 02The First EraHow did we personalize before language models, and what did it teach us?27 min read
- 03What Designers KnewWhat did UX research and HCI learn about adapting interfaces to people, and what does it teach AI personalization?25 min read
- 04Shallow to DeepWhy was deep personalization always limited to a few people, and what does "deep" actually mean?18 min read
- 05The New EraWhat exactly changed with AI, and where is it going?27 min read
Part II
Where to Start
Part III
Customer Data
- 07Getting the DataWhat data can we use, where does it come from, and are we allowed to use it?22 min read
- 08Preparing ItHow do we clean, enrich, and use unstructured data?21 min read
- 09One Customer, Many SystemsHow do we unify a customer across CRM, warehouse, and tools without merging two people?25 min read
Part IV
Unified Customer Memory
- 10Memory Built for AgentsWhat changes when the main reader of our customer data is an AI agent, not a person at a dashboard?30 min read
- 11Keeping Memory TrueHow does memory stay accurate, fresh, and traceable?26 min read
- 12The Cost of RememberingHow do we keep memory affordable at scale without losing accuracy?21 min read
- 13Agents That Run for WeeksHow do long-running agents use and maintain memory?22 min read
Part V
Deciding and Acting
- 14The Decision LayerWhat should happen next for this person, including nothing?26 min read
- 15Writing for OneHow do we generate for each person without inventing?24 min read
- 16Beyond the Message: Generative ExperiencesWhat happens when the page, the app, and the notification, not just the message, are composed for one person?27 min read
- 17Now or LaterWhen must personalization be real-time, and when is async better?19 min read
Part VI
Trust
- 18GovernanceHow do we govern what the AI says and does, at the level of the organization and at the level of each customer?27 min read
- 19Privacy and RegulationWhat should we never know, keep, or use, and what does the law require?25 min read
- 20Accuracy and QualityHow do we achieve and keep accuracy at scale?25 min read
Part VII
Proof and Deployment
Part VIII
Playbooks, by team
The foundations applied to one team's work: what to personalize first, what to leave alone, and how to know it worked.
- Marketing · M1Inbound and the WebsiteThey came to us. What should they see and receive next?
- Marketing · M2Accounts, Ads and EventsWe chose them. How do we reach them as individuals?
- Marketing · M3The Physical ChannelWhen each touch costs real money, who deserves one and what should it say?
- Sales · S1New CustomersHow do we open a relationship without sounding like every other AI message?
- Sales · S2Growth and ReturnWhat do we already know, and how do we use it to go deeper or win them back?
- Customer success · CSCustomer SuccessHow does customer success personalize onboarding, adoption and renewal for each account and each user?
- Support · SPSupportHow do we stop asking customers what we already know?
- Product · PRProductHow does the product adapt to each user, and what does the team learn?
- Software · SWSoftware CompaniesHow does a software company personalize when its richest signal is how people use the product?
- Research · RSResearch and StrategyHow does understanding individuals roll up into strategy?
- Across teams · P9One Customer, One ConversationHow do we stop marketing, sales, support, product, and our agents from contradicting each other, or exhausting the same person?
Companion paperGoverned Memory: A Production Architecture for Multi-Agent WorkflowsThe full architecture behind Part IV, with evaluation. arXiv, 2026.
Get chapters by email as they are revised