Platform · AI with approval
AI in campaign management: the AI drafts. You decide.
The PBM Campaign Platform uses AI for drafts, variants and analyses. Every suggestion is checked and approved by a person before it reaches a customer. And if you don’t want AI, you switch it off.
AI suggestion Checked Approved · Four-eyes
AI suggestion · not yet accepted
Your fixed-term deposit expires on 31 March – secure your terms now before it’s too late. You will find three options in your personal overview.
Compliance & tone check
- no return promises
- Mandatory notice present
The problem
Why do regulated companies hesitate to use AI in marketing?
Because a wrong sentence about a financial or insurance product is a liability risk. And because nobody can explain why a black box wrote something to a customer.
“The AI promises a return that doesn’t exist.”
A compliance check blocks impermissible statements before approval.
“Nobody knows who approved the text.”
Every AI action and every approval is recorded in the audit log.
“Every team uses a different AI tool.”
The AI works inside the platform – with your roles, brands and rules.
Guiding principle
How does AI work in the PBM Campaign Platform?
As a suggestion layer on top of a rule engine. The AI drafts, translates and analyses. Execution takes place in a tested, deterministic engine. No language model makes decisions in the sending path.
-
Suggestion, not automation
The AI delivers drafts. A person accepts, changes or rejects them.
-
Deterministic execution
If you formulate a rule in plain language, the AI translates it into a rule expression – and back into plain language for checking. It is the expression that is executed, not the sentence.
“Show the tax advantage argument if the portfolio value is over €50,000 and relevant for tax purposes”
contact.depotwert > 50000 and contact.steuer_relevant“Show the tax advantage argument if the portfolio value is over €50,000 and relevant for tax purposes”
matches
-
Can be switched off
The AI layer is optional. Without it, the platform continues to work with rules and approved content.
Two layers
AI layer: on
AI layer: off
AI off – journeys, consent check and compliance gate keep running.
Features
What exactly can the AI do?
It helps in five areas – from the first draft to analysis. Every function delivers a suggestion that a person accepts, changes or rejects.
| Area | Features |
|---|---|
| Create | Column mapping during import · template draft from your brand CI · “Build your Campaign”: campaign framework from a goal description · rules in plain language → rule expression with back-translation |
| Personalise | Text variants per recipient · argument recommendation from a pool of approved arguments · image variants · translation |
| Optimise | Variants for subject lines and calls to action with A/B tests · recommendation for send time and channel · lead score forecast as an optional layer on top of the rule-based score |
| Analyse & check | Compliance and tone-of-voice check · questions to your data, e.g. “Why did the campaign perform worse for brand A?” · “Explain your Campaign”: campaign progress in three sentences · access diagnosis: who is allowed to see this record – and why? |
| Agents | Daily overview (Daily Digest) · anomaly monitor · alert to advisers about warm contacts |
Approval
How does an AI text get into a send?
Only via review and approval. No AI suggestion reaches a customer directly.
-
AI step: Suggestion: The AI generates a text variant. It is marked in violet: “not yet accepted”.
-
Compliance check: The check looks for impermissible promises, missing mandatory information, the wrong tone and brand violations. It is blocking: anything that fails does not go any further. This also applies to texts written by people.
-
Human approval: The approval rule is configurable; the default is the four-eyes principle. Each language version is approved individually.
-
Pre-generated: Approved content is produced in batches before sending and stored. Nothing is generated live when a page is accessed.
-
Send: Before every send there is the compliance gate with the consent check. It cannot be switched off.
The compliance gate before every send Roles, permissions & approvals
AI suggestion · not yet accepted
Your fixed-term deposit expires on 31 March – secure your terms now before it’s too late. You will find three options in your personal overview.
Compliance & tone check
- no return promises
- Mandatory notice present
Tone: “too urgent”
Suggestion: “secure your terms now … before it’s too late” → “You can decide at your own pace how to proceed.”
Created by: J. Neumann · Approval by: C. Reuter (four-eyes)
Approved · C. Reuter · Four-eyeslogged · 25 September 2026 10:42
Example · Stadtwerke Musterstadt
How does a sentence become a campaign?
Example: the sales team at Stadtwerke Musterstadt describes its goal in one sentence: “Customers whose electricity tariff ends in 60 days should renew – email if there is consent, otherwise letter; reminder after 10 days.” The AI suggests a campaign framework: one segment, six steps, two template drafts and the goal “Renewal completed”. The team reviews it, adjusts one text and accepts everything as a draft. The campaign only goes live after approval.
Tariff ends in 60 days
From framework to finished journey AI with approval for municipal utilities
“Customers whose electricity tariff ends in 60 days should renew – email if consent, otherwise letter; reminder after 10 days.”
- SegmentElectricity tariff ends in 60 days
- GateConsent per channel
- Emailif consentLetterotherwise
- Reminderafter 10 days
- GoalRenewal completed
- Segment
- 2 templates
- Conversion goal “Renewal completed”
Nothing goes live without approval.
Agents
What are AI agents allowed to do?
Report, yes; act, only with approval. Reporting agents read data and provide information – for example with an overview of your campaigns every morning. Agents that would change something run only behind human approval.
- Agents have their own role with limited permissions.
- Anomaly monitors report irregularities, such as unusually few opens for one brand.
- Advisers receive an alert when a customer becomes “warm”.
Your campaign overview, Friday 25 September
- AnomalyOpens for the Köln brand today well below average
- 14 warm leads for advisers
- 1 content item awaiting approval Waiting
Traceable
Can you trace what the AI has done?
Yes. Every AI action goes into the audit log – with time, function and actor. AI agents appear there as a separate actor type, distinct from people and service accounts. Every approval is logged.
- · AI agent “Compliance check” · tone-of-voice note generated
- · C. Reuter · content approved (four-eyes) · Wiederanlage/DE
Law · As of: September 2026
What do the GDPR and the EU AI Act mean for AI in campaign management?
They require transparency and human oversight – precisely where AI has a say in decisions about people. The PBM Campaign Platform is designed for this. The legal assessment of your specific use remains your responsibility.
AI and customer data in the Trust Centre Guide: hyper-personalisation & GDPR
Frequently asked questions about AI with approval
Can I switch off the AI functions?
Yes. The AI layer of the PBM Campaign Platform is optional. Without it, the platform continues to work with rules, variants and approved content. Journeys, consent check and compliance gate work independently of it.
Does the AI decide which customer receives which message?
No. Which customer receives which message is determined by rules and approved content, executed by a deterministic engine. The AI can suggest variants, arguments, send times or channels. A suggestion is only adopted after a human decision and approval.
Are AI texts checked before sending?
Yes, twice. First, a blocking compliance check looks for impermissible promises, missing mandatory information, tone of voice and brand violations. Then a person approves, by default according to the four-eyes principle and individually for each language version. The compliance check also applies to texts written by people.
Is AI in campaign management a high-risk system under the EU AI Act?
Generally not, but it depends on the purpose of use. In the financial sector, Annex III of the AI Regulation names, for example, creditworthiness assessment and risk assessment in life and health insurance, but not marketing texts or campaign analyses. As of: September 2026. This classification does not replace legal advice.
How does the platform deal with Art. 22 GDPR?
Art. 22 GDPR protects against decisions based solely on automated processing with legal or similarly significant effects. For predictive AI functions such as send time, channel or lead score forecasts, the PBM Campaign Platform provides transparency, human oversight and an option to switch them off. Whether Art. 22 applies must be assessed on a case-by-case basis.
Is every AI action logged?
Yes. Every AI action and every action by an AI agent goes into the audit log, separated by actor type: human, service account or AI agent. Every approval is also recorded with the person and time.