AI Personalisation in Marketing: Examples and How to Get Started
Amazon generates 35% of its revenue from AI recommendations. Netflix saves $1 billion per year in retention through personalised content. Spotify's Discover Weekly drives 40% of all streaming activity. These are not coincidences — they are the result of the same underlying approach, applied at scale. This guide breaks down what that approach is, what results it produces for teams without those budgets, and how to get started with the tools available today.
Quick Summary
- 1Companies that excel at personalisation generate 40% more revenue and reduce customer acquisition costs by up to 50%
- 271% of consumers expect personalised experiences — 76% get frustrated when they don't receive them
- 3AI personalisation goes beyond merge tags: it predicts behaviour and acts on signals before a customer explicitly states their intent
- 4First-party data is the foundation — AI tools cannot personalise without behavioural signals from your own channels
- 5You can implement meaningful AI personalisation at five touchpoints: email, website, ads, search, and product recommendations
35%
of Amazon revenue from AI recommendations
+40%
more revenue for personalisation leaders
+26%
average conversion rate increase from AI recs
−57%
lower customer acquisition cost with AI targeting
What AI personalisation is — and what it is not
Basic personalisation is rules-based: if a contact is tagged as a customer, show them a different banner. If they are in Germany, send a localised email. These are useful, but they do not require AI — they require data and a conditional logic builder.
AI personalisation is predictive. It analyses patterns across thousands or millions of customer interactions to predict what an individual is likely to want, do, or respond to — before they signal it explicitly. The difference is the scale and the direction: rules-based personalisation reacts to what a customer has done; AI personalisation anticipates what they will do next.
| Rules-based personalisation | AI personalisation | |
|---|---|---|
| How it works | If/then logic set by marketers | Models trained on behavioural data |
| What it uses | Segment membership, tags, attributes | Purchase history, browse patterns, engagement signals |
| What it predicts | Nothing — it reacts | Churn risk, purchase intent, LTV, next best action |
| Scale | Dozens of rules, manually maintained | Millions of individual-level decisions, automated |
| Setup time | Hours to days | Weeks to months (data collection required) |
How the big platforms do it — and what marketers can copy
The core technique behind Netflix, Amazon, and Spotify is the same: collect first-party behavioural signals, train a model to find patterns across similar users, and surface content or products predicted to be relevant to each individual. The scale is different, but the logic is replicable.
Amazon
35% of total revenue from recommendationsHow: Collaborative filtering: analyses what customers with similar purchase histories bought next. The "Customers also bought" section uses this model. The key input is purchase data — not demographics.
Marketer takeaway: If you have ecommerce data, a Klaviyo or Shopify AI recommendations block replicates this at a fraction of the cost. The more purchase history you have, the better the predictions.
Netflix
80% of viewing from recommendations; $1B/year saved in retentionHow: Surfaces content based on viewing history, time of day, device, and what similar users watch. Personalises even the thumbnail image shown per subscriber — a different still frame per person for the same title.
Marketer takeaway: Dynamic content blocks in email (showing different product or article images per segment) are the accessible equivalent. Klaviyo, Iterable, and Braze all support this natively.
Spotify
Discover Weekly drives ~40% of streaming activityHow: Combines collaborative filtering (what people with similar taste have liked) with audio analysis (tempo, key, danceability) to build weekly playlists that feel personally curated.
Marketer takeaway: For content marketers, this is the logic behind AI-curated email newsletters: surface articles the individual has not read based on what similar subscribers engaged with. Beehiiv and some ESP AI tools do this.
5 types of AI personalisation — with realistic examples
You do not need to implement all five at once. Start with the one that maps to your highest-traffic channel and where you already have the most data.
Email personalisation
Rather than sending one newsletter to all subscribers, AI segments by engagement pattern and purchase history — showing different product blocks, articles, or offers per recipient within the same send. A subscriber who clicked three articles on AI tools sees AI-focused content; one who bought a course sees course recommendations.
Website personalisation
A first-time visitor from a LinkedIn ad sees a B2B-focused homepage headline. A returning customer sees a continuation of where they left off. A high-intent visitor (visited pricing three times) sees a demo CTA. None of these require manual rules — AI learns which experience drives conversion for which visitor type.
Product and content recommendations
On a product page, instead of showing bestsellers, AI surfaces items most likely to convert for that specific visitor based on what similar buyers purchased next. For content publishers, this is article recommendation widgets that surface pieces the individual has not read but similar readers engaged with.
Paid advertising personalisation
Dynamic creative optimisation (DCO) automatically assembles ad variants — headline, image, CTA — matched to each audience segment, device, and placement. Rather than creating 20 ad variants manually, a DCO tool generates and tests hundreds of combinations and routes spend to the best-performing combinations per audience.
Predictive segmentation
Instead of segmenting by who bought in the last 90 days, AI builds segments based on predicted future behaviour: who is about to churn (suppress from ads, trigger a win-back email), who is about to buy (increase bid, send a nudge), who has high lifetime value potential (invest more in retention). These are signals that do not exist in historical data — the model infers them from patterns.
The data foundation: first-party data is non-negotiable
Every AI personalisation system is only as good as the data feeding it. Third-party cookie data is largely gone — browser restrictions and privacy regulations have eliminated the easy route to behavioural tracking across the web. In 2026, AI personalisation runs on first-party data: signals collected from your own properties with explicit consent.
The four first-party data types that power meaningful AI personalisation are:
Behavioural data
Pages visited, time on page, scroll depth, content engaged with, search queries on your site
Source: Website analytics, heatmaps
Transactional data
Purchase history, order value, category affinity, return rate, repurchase frequency
Source: CRM, ecommerce platform
Engagement data
Email opens, clicks, subject line preferences, send-time patterns, content category interests
Source: ESP / marketing automation
Declared data
Preferences, interests, and goals a customer explicitly shares via surveys, onboarding flows, or preference centres
Source: Forms, quizzes, onboarding
McKinsey's research consistently shows that companies excelling at personalisation run 40% more revenue from these activities than competitors — but the differentiator is not the AI tool, it is the data quality. Two companies using the same personalisation platform get very different results depending on how rich and accurate their first-party data is.
Tools by personalisation type
Start with the tools already in your stack before adding new platforms. Most modern ESPs, CDPs, and ecommerce platforms have built-in AI personalisation features that are underused.
| Use case | Tool options | Entry cost |
|---|---|---|
| Email personalisation | Klaviyo, ActiveCampaign, Brevo, Iterable | Free / $20/mo |
| Website personalisation | RightMessage, Personyze, VWO Personalise | $97–$199/mo |
| Product recommendations | Shopify native AI, Klaviyo flows, Nosto | Free (Shopify) / $99/mo |
| Predictive segmentation | Klaviyo AI segments, Bloomreach, Insider | $20/mo (Klaviyo) / custom |
| Dynamic ad creative (DCO) | Meta Advantage+, Google Performance Max, Smartly | Free (Meta/Google native) |
| Content recommendations | Beehiiv AI, Recombee, Coveo | $39/mo (Beehiiv) |
How to get started: 5 steps
Audit your first-party data
Before selecting a tool, map what data you already collect and where it lives. Typical gaps: website behaviour not connected to email contacts, purchase history sitting in an ecommerce platform not synced to your ESP, and engagement data locked in one tool and not shared with others. Data you cannot connect cannot be personalised against.
Pick one channel and one use case
Do not try to personalise everything at once. Start with your highest-traffic channel and the personalisation type most likely to move a business metric you already track. For most teams, that is email (highest data density) or product recommendations (most direct revenue connection).
Enable native AI features in your existing tools
Check whether your current ESP, ecommerce platform, or CMS already has AI personalisation features you are not using. Klaviyo predictive segments, Shopify product recommendations, and Meta Advantage+ creative are all available on existing plans and require no additional tool budget.
Set a baseline and define success
Before turning on AI personalisation, record your current conversion rate, AOV, or engagement metric for that channel. Define the improvement that would justify the setup time. AI personalisation typically takes 4–8 weeks to accumulate enough data to deliver reliable predictions — judge results at 90 days, not 2 weeks.
Expand once one use case is working
Once your first personalisation use case is producing measurable results, add the next adjacent layer — not an entirely new channel. If email personalisation is working, add send-time optimisation. If product recommendations are lifting AOV, add predictive segments to your ad targeting. Each layer compounds the previous one.
Privacy and compliance
AI personalisation using first-party data is generally compliant under GDPR and CCPA where explicit consent was collected at opt-in. However, automated decisioning systems that affect pricing, offers, or access — and predictive profiling of EU individuals — may trigger additional transparency and fairness obligations under the EU AI Act (Article 50, high-risk system provisions). If you use AI to make or influence commercial decisions about EU contacts, document your system and be prepared to provide a plain-language explanation of how decisions are made. See our EU AI Act guide for marketers for full details.
Related guides
AI marketing intel,
delivered weekly.
Join marketers getting the digest free. No spam, always.