What problem does it solve?
Marketing teams know that relevant messages work better than broadcast ones, but personalization has been limited by two bottlenecks. The first is decisioning: choosing, for millions of customers, which of hundreds of offers and messages is most relevant right now, while respecting frequency caps, consent and eligibility. The second is content: even a good decision engine was only as personal as the handful of creative variants the studio could produce and legal could approve.
Machine learning has handled the first bottleneck for years, in recommendation engines and next best action systems. Generative AI now attacks the second: copy, images and video variants per segment, language and context, without a separate studio brief for each one. Together they make segment of one campaigns practical. The risks grow with it: invented claims in generated copy, offers that exploit vulnerable customers, profiling without a lawful basis, and brand damage from content nobody reviewed. The workable approach is a library of approved building blocks within which AI personalizes.
How does it work?
- Unify the signals. Customer profile, consent, product holdings, behaviour on web and app, and context (location, time, channel) are brought into one decisioning layer.
- Decide the next best action. Models predict propensity and value for each eligible action, including service messages and doing nothing, and arbitration rules pick one within caps, eligibility and business priorities.
- Assemble the content. The chosen action is rendered from approved building blocks; generative AI produces copy and creative variants within brand, claims and disclosure rules.
- Check before sending. Automated checks cover required disclosures, prohibited claims, consent and vulnerability flags; new templates and campaigns get human approval.
- Deliver and learn. The message goes out in the right channel; responses feed back into the models, and experiments with control groups measure the real uplift.
- Audience
- Back office
- Autonomy
- Supervised agent
- Adoption
- Mainstream
- Channels
- Email, Mobile app, Web chat, API and system to system
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Conversion uplift | Too few to pool | 30% | 1 | 1 vendor |
| Cycle time reduction | Too few to pool | 10x | 1 | 1 vendor |
| Cycle time reduction | Too few to pool | 40% | 1 | 1 vendor |
| Productivity gain | Too few to pool | about 50% | 1 | 1 vendor |
| Revenue uplift | Too few to pool | at least 20% | 1 | 1 vendor |
Value drivers: Revenue growth, Customer experience, Employee productivity, Speed and cycle time.
Indicative value
A consumer business with USD 200 million a year in revenue from marketing driven campaigns
USD 800,000 to USD 4.8 million
Incremental contribution from personalization per year
How this is calculated
Formula: campaignRevenue * uplift * margin. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Annual revenue attributed to targeted campaigns campaignRevenue, USD per year | 200,000,000 | 200,000,000 | The reference organization. Use revenue measured against a holdout, not last click attribution. |
| Incremental revenue from personalization versus current targeting uplift, fraction of campaign revenue | 0.02 | 0.06 | Editorial assumption, kept well below the benchmarks on this page (Google Cloud reports revenue from Radisson Hotel Group's AI powered campaigns up by more than 20% and Catchtable's reservation conversion up 30%), because those figures are vendor reported and do not state their baseline or whether a control group was used. |
| Contribution margin on incremental revenue margin, fraction of revenue | 0.2 | 0.4 | Editorial assumption, replace with your own. |
What it leaves out: An uplift estimate that only holds if measured with control groups. It leaves out content production savings, the cost of data, platform and people, and any revenue lost to customers who opt out after poorly judged personalization.
Who already uses it?
7 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Amazon
United States · Retail and ecommerce · 2025
Since the end of 2023 Amazon lets independent sellers create a product listing from a few words or a single image, and since March 2024 also from the URL of their own web page: generative AI on Amazon Bedrock drafts the title, bullet points, description and attributes, and the seller submits the draft, with Amazon encouraging a review first. Bulk creation from a spreadsheet followed, and Enhance My Listing, which Amazon said in May 2025 had begun rolling out in the US, suggests updates to existing listings based on shopping behaviour. In May 2025 Amazon reported that sellers accept the AI generated content with little to no edits about 90% of the time. The same post describes Amazon using generative AI to personalize product recommendation categories and product descriptions shown to customers on the website and in the shopping app, based on a customer's shopping activity; no outcome number is given for that side of the work.
- Users served: at least 900,000, selling partners who have used the listing tools, by May 2025
"Now, more than 900,000 Amazon selling partners have embraced these tools, with sellers accepting AI-generated content with little to no edits approximately 90% of the time."
Claimed by: organization - Quality score uplift: 40%, overall listing quality of listings created with the tools, reported by May 2025
"When sellers use our Gen AI tools to create listings, they see a 40% increase in overall listing quality, helping them create content that enhances customer engagement and boosts sales potential."
Claimed by: organization
Commonwealth Bank of Australia
Australia · Banking · 2022
Commonwealth Bank's Customer Engagement Engine (CEE), built on Pega Customer Decision Hub, suggests in real time the next best conversation to have with each customer, whether in the branch, on the phone, online or on a mobile device. Beyond suggesting conversations, the bank uses it to match customers to government benefits and rebates they may be missing (Benefits finder) and to reach customers hit by natural disasters with same day support such as a loan deferral. The same decisions feed digital channels and prompts for branch and contact centre staff.
No outcome disclosed.
Catchtable
South Korea · Travel and hospitality · 2025
Catchtable, a restaurant discovery and reservation app from South Korea, built personalized recommendation models on Vertex AI that combine language models with custom embeddings to read search intent and recommend restaurants in real time. Google Cloud reports a 30% increase in reservation conversion rates and a 150% increase in impressions per restaurant search.
- Conversion uplift: 30%, reservation conversion rate
"Catchtable uses Vertex AI and Kubeflow with GPU optimization to build personalized restaurant recommendation models, achieving a 30% increase in reservation conversion rates and a 150% increase in impressions per restaurant search."
Claimed by: vendor
Swarovski
Austria · Retail and ecommerce · 2025
Swarovski, which sells in more than 140 markets, launched Génie in 2023, a generative AI portal on Vertex AI and Gemini, on top of a BigQuery data foundation consolidated with the partner CloudSufi. More than 1,000 employees use it for tasks including content translation into over 20 languages, creative asset generation and testing visuals and descriptions for different regions. Google Cloud reports that campaign localization became 10 times faster through AI assisted translation and asset adaptation, and that Génie's AI personalized email campaigns see 17% higher open rates and 7% higher click through rates. Every AI application is evaluated against an internal ethics and risk model.
- Cycle time reduction: 10x, speed of campaign localization
"Campaign localization is 10x faster, thanks to AI-assisted translation and asset adaptation"
Claimed by: vendor - Users served: at least 1000, employees using the Génie portal
"Over 1,000 employees now utilize Génie for tasks such as contract review, content translation into over 20 languages, creative digital asset generation, and campaign inspirations, and product design cost estimation."
Claimed by: vendor
Virgin Voyages
United States · Travel and hospitality · 2025
Virgin Voyages uses Google's generative video and image models to create thousands of personalized advertisements and emails in its brand voice, and has deployed specialized agents that turn behavioural signals into personalized campaign actions. Google Cloud reports that campaign creation time fell by 40%.
- Cycle time reduction: 40%, campaign creation time
"Virgin Voyages have already deployed more than 1,000 specialized agents to reduce campaign creation times by 40% by turning behavioral signals into personalized action."
Claimed by: vendor
Radisson Hotel Group
Belgium · Travel and hospitality · 2024
Radisson Hotel Group worked with Accenture to personalize its advertising at scale with Vertex AI and Gemini models, trained on extensive datasets stored in BigQuery. Google Cloud reports that ad team productivity rose by around 50% and that revenue from the AI powered campaigns rose by more than 20%. The comparison group for the revenue figure is not stated.
- Revenue uplift: at least 20%, revenue from AI powered campaigns
"By training them on extensive datasets stored in BigQuery, its ad teams saw productivity rise around 50% while revenue increased from AI-powered campaigns by more than 20%."
Claimed by: vendor - Productivity gain: about 50%, advertising team productivity
"By training them on extensive datasets stored in BigQuery, its ad teams saw productivity rise around 50% while revenue increased from AI-powered campaigns by more than 20%."
Claimed by: vendor
Square Enix
Japan · Media and entertainment · 2024
Square Enix uses customer data to develop AI optimized marketing assets and send players personalized emails matched to their preferences. Google Cloud reports a 20% increase in email opens and a 10% increase in retention.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Customer profile and behavioural data with recorded marketing consent per channel
- Product and offer catalogue with eligibility rules
- Approved brand guidelines, claims library and required disclosures
- Response history and holdout groups to measure uplift
Systems to integrate
- Customer data platform or data warehouse
- Decisioning or next best action engine
- Marketing automation, email and push platforms
- Content management and digital asset management
- Consent and preference management
Complexity: Medium
Generating variants is easy; everything around it is not. Unified customer data with consent, a decisioning layer with arbitration, a library of approved content blocks and experiment discipline decide whether personalization pays off.
- 1
Fix consent and data first
Know for each customer and channel whether marketing and profiling are allowed, and make the decisioning layer enforce it. Personalization on data you may not use is a liability.
- 2
Build an approved content library
Break campaigns into building blocks (claims, offers, disclosures, images) that legal and brand have approved, and let generative AI personalize wording and combinations within them.
- 3
Start with one journey and a holdout
Pick one high volume journey, such as onboarding, cart abandonment or renewal, and measure uplift against a randomized control group before scaling.
- 4
Add arbitration across campaigns
Move from campaign by campaign targeting to one decision per customer that weighs all eligible actions, including service messages and no message, with frequency caps.
- 5
Automate checks, keep approvals for the new
Run automatic checks for disclosures, prohibited claims and vulnerability flags on every variant, and require human approval for new templates, offers and audiences.
- 6
Review fairness and customer outcomes
Check who receives which offers and prices, and whether customers in vulnerable circumstances are targeted with products that may harm them.
Guardrails
- Personalization only on data with a lawful basis and recorded consent for the channel
- Generated copy restricted to approved claims and must include required disclosures
- No targeting that exploits age, disability or financial difficulty; vulnerable customers excluded from high risk offers
- Frequency caps and an easy opt out in every message
- Synthetic images and video labelled or marked where the law requires it
KPIs to instrument
- Incremental conversion and revenue versus a randomized holdout
- Opt out, unsubscribe and complaint rates per campaign
- Share of generated variants passing automated compliance checks first time
- Campaign production time from brief to launch
- Distribution of offers across customer groups, including vulnerable customers
Human in the loop
Marketing owns the strategy, audiences and offers; brand and legal approve templates, claims and new campaigns; analysts own the experiments. The AI selects and assembles within those approvals, and people review samples of what customers actually received.
Common failure modes
- Uplift that is really attribution
- Personalized campaigns take credit for sales that would have happened anyway. Measure against randomized holdouts.
- Invented claims at scale
- Generated copy promises a benefit or price that does not exist. Restrict generation to an approved claims library and check every variant.
- Creepy or harmful targeting
- Messages reveal inferences customers did not expect, or push credit or gambling to people in difficulty. Limit sensitive inferences and exclude vulnerable customers.
- Too many messages
- Each campaign optimizes for itself and customers are flooded. Arbitrate across campaigns with frequency caps.
What are the risks and rules?
EU AI Act
Depends on design
Most personalization and content generation is minimal risk. Providers of systems that generate synthetic audio, image, video or text content must mark the output as artificially generated, and deployers must disclose deep fakes (Article 50(2) and 50(4)). Personalization that deploys manipulative or deceptive techniques, or exploits vulnerabilities due to age, disability or a specific social or economic situation, in a way that causes or is reasonably likely to cause significant harm, is prohibited under Article 5(1)(a) and (b). Using AI to assess creditworthiness or to price life and health insurance is high risk under Annex III point 5(b) and 5(c) and belongs on its own page. Outside the AI Act, the FCA Consumer Duty applies only to FCA regulated firms (the financial services slice of this use case), and the Telephone Consumer Protection Act applies only to campaigns delivered by call or text message in the US.
Guidance
- Article 5, prohibited AI practices (European Union, Europe). Points 1(a) and 1(b) prohibit manipulative techniques and the exploitation of vulnerabilities that cause significant harm.
- Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). Marking of synthetic content and disclosure of deep fakes used in campaigns.
- Article 21 GDPR, right to object (European Union, Europe). People may object at any time to processing for direct marketing, including profiling related to it.
- Direct marketing and privacy and electronic communications (UK Information Commissioner's Office, Europe). The UK regulator's guidance hub on direct marketing under PECR and data protection law, covering consent for email and text marketing and choosing a lawful basis.
Controls to put in place
- Consent and preference checks enforced in the decisioning layer
- Approved claims and disclosure library with owners
- Automated pre send checks plus human approval of new templates and audiences
- Holdout based measurement for every personalized journey
- Periodic review of offer distribution for vulnerable customers and fairness
Frequently asked questions
- What results do companies report from AI personalization?
- Vendor reported results are large. Google Cloud reports that revenue from Radisson Hotel Group's AI powered campaigns rose by more than 20% and ad team productivity by around 50%, that Catchtable's personalized recommendations raised reservation conversion by 30%, and that Virgin Voyages cut campaign creation time by 40%. None of these states its baseline, so treat them as upper bounds and measure against your own holdout.
- How is this different from an offers and rewards agent in banking?
- The banking offers page covers choosing and explaining offers and rewards from transaction data inside the bank's app and conversations. This page covers the wider marketing job in any industry: deciding and generating personalized campaign content at scale across email, app, web and paid media.
- Can generative AI write campaign copy without legal review?
- Not freely. The workable model is an approved library of claims, offers and disclosures that legal and brand sign off once, with AI personalizing wording and combinations inside it, automated checks on every variant and human approval for anything new.
How to cite this page
Blits.ai AI Use Case Library, "AI marketing personalization at scale", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/personalized-marketing-at-scale. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published