AI in customer winback enables operationally usable copy optimisation within a workflow. The Paperfly AI agent transforms existing content into a version that works for each channel: concise, active, easy to understand and focused on one clear primary action. At the same time, the workflow ensures that delivery, deadlines, fallbacks and evidence are handled reliably as part of the process.

The goal is a winback process that measurably leads customers to the next step more often and can be scaled reliably when willingness to switch is high and time windows are tight.

In this article, we explain the rules behind channel-specific copy optimisation, how clarity can be measured, why DIWA works as a reader-guidance model in winback, and how these requirements can be translated into a genuine end-to-end workflow as quality gates.

Series: Winback as a Workflow for Utilities

What to expect in this article:

Why Copy Quality in Winback Is Suddenly a Process Issue

In customer winback, copy quality is not a detail but a process-critical lever. When willingness to switch is high, speed matters, but so does whether customers understand the message immediately, can respond without friction and do not end up in follow-up questions or drop off.

What Copy Quality Means in Practice:

  • Reduce Friction: Information overload and complicated emails or letters lead to refusal or follow-up questions. Optimisation aims to improve clarity and make it easier to act.
  • Test Instead of Guessing: Optimisation is measured, variants are tested against each other and iterated, meaning the test is repeated multiple times.
  • Operationalisation: Rules, such as sentence length, structure and reader guidance, are embedded as standards within the workflow.

Copy Optimisation with the Paperfly AI Agent

Copy optimisation by an AI agent does not mean that AI invents arbitrary new marketing copy. Instead, it transforms existing content according to a fixed set of rules into a version suited to the channel and intended action: shorter, more active, less padded with stock phrases and more precise.

At its core, this is 'agentic AI': the objective and AI guardrails, meaning clearly defined boundaries, are predetermined, and the agent optimises execution across all touchpoints without reinventing the content. The AI optimises execution, not the underlying message. Approved copy remains unchanged; the guardrails define what may be said, while the agent controls only when, where and how it is delivered.

What an AI Copy Agent Does in Winback, and What It Does Not Do

AI copy agent, rule-based and operationally usable:

  • Apply Channel Profiles (email ≠ SMS ≠ letter): automatically adapt length, structure and tone.
  • Improve Readability & Concision: break up long sentences, rewrite passive phrasing into active voice, and reduce filler words and stock phrases.
  • Adapt Style and Tone to the Target Audience: for example, factual, clear and respectful, while implementing or suggesting better and more natural phrasing.
  • Enforce the 'One-Action' Principle: one primary action, with a clear next step.
  • Automate Execution, for example by generating variants and deploying them after approval.

Control Mechanisms in Implementation:

Approved Copy Modules: Der Agent arbeitet mit versionierten, abgenommenen Modulen (z. B. Legal-approved) und darf diese nur im festgesetzten Rahmen semantisch verändern, formatieren oder anders platzieren.

Rule-Based Risk Checks: Statements and details, such as price, term, withdrawal rights and bonus conditions, are safeguarded using allowlists and blocklists; risky wording is blocked where necessary.

What an AI Copy Agent Does Not Do:

  • No 'Creative Reinvention' thanks to rules and compliance guardrails*. No hallucination of facts: the agent does not invent new discounts, deadlines, product benefits or justifications. It only uses approved content.
  • No Changes to Core Statements or Liability Wording: The agent does not alter the substantive content, such as offer terms, legal notices or commitments. If an adjustment is required, it switches to suggestion mode and waits for approval.
  • No Replacement for Segment Strategy, offer design or timing. The agent optimises copy, not product strategy.
In short: the agent automates execution and improves readability, but it operates within approved guardrails. Content changes go through an approval process rather than running on autopilot.

* Guardrails (boundaries): predefined limits and rules for AI outputs, including tone, length, prohibited wording and mandatory elements such as opt-out notices, so that results remain consistent, safe and compliant.

Making Readability Measurable: Using HIX as a Reference Model for Clarity

The Hohenheim Comprehensibility Index (HIX) is used here as a German-language reference model for measuring clarity. It makes clarity objectively comparable by combining several readability formulas validated for the German language with additional text parameters on a scale from 0 to 20.

With an AI agent, this becomes a practical workflow step: input the copy, receive an analysis, including indications of clarity barriers and concrete optimisation suggestions.

Internal terminology and language rules can also be embedded. This is valuable for winback because cognitive load directly translates into drop-offs, follow-up questions and delayed decisions, which in turn means missed time windows.

Translating HIX Thinking into Winback Channel Profiles

Channel Typical Reading Situation Common Clarity Killer HIX Lever
SMS 3-5 seconds of attention Sentences that are too long, excessive context Break up sentences, communicate one benefit and one action. Remove or explain unfamiliar terms.
E-Mail Text is scanned, decision is made quickly Noun-heavy style, nested sentences Reduce passive voice, use bullet structures and explain terms briefly.
Letter with QR Code Trust and orientation Bureaucratic language, high information density Use a clear sequence of steps, reduce abstract wording and keep terminology consistent.

Reader Guidance Instead of 'Nice Copy': DIWA Structure in Winback

DIWA is not a framework for writing 'nice copy', but a guidance model that leads readers respectfully and directly towards the desired action. Together with clarity measurement using HIX, it forms a practical, action-oriented pattern that can be used as a quality gate in AI winback workflows.

DIWA Applied to Winback:

  • D - Direct (Reason/Why?)
    'You have cancelled. Before the switch is completed, check our best tariff option here...'
  • I - Interest
    'Many customers are currently reviewing their energy costs. That is why we make it easy for you to compare the right offer quickly.'
  • W - Value
    'Check your tariff in 20 seconds / receive a callback within 60 minutes / no waiting time.'
  • A - Action (One Clear Action)
    'Show best offer' / 'Request a callback'.

Channel Profiles: How Winback Copy Differs Across Email, SMS and Letters

In winback for utilities, a channel is not merely a delivery medium, but a cognitive context. An SMS is typically processed within seconds and therefore requires maximum brevity and an immediately understandable call to action (CTA).

Emails can carry more context and structure, while SMS works more as a time-sensitive prompt. That is why combining both channels often works better than an either-or approach.

For letters with a QR code: the CTA must not only prompt an action, but also make clear what happens after the scan. Short, unambiguous wording increases the likelihood that the QR code will actually be used.

Across all channels, the goal is a consistent omnichannel journey: customers are offered different channels without having to start again from the beginning, otherwise friction and drop-off increase.

Why 'One Text for Everyone' Makes Winback Worse

Winback scheitert nicht daran, was kommuniziert, sondern wie uniform kommuniziert wird: wenn Wertklasse, Kündigungsgrund und Abwanderungsrisiko unterschiedlich sind, muss sich auch Ton, Angebotstiefe und Kanalfolge unterscheiden.

Too long for SMS, too thin for email and too promotional for a letter is not the goal.

That is exactly why an AI winback workflow explicitly applies personalisation, segmentation and channel and timing fit.

Why 'One Text Fits All' Creates Concrete Problems in Winback

  • Customer Value Is Ignored: High-value segments may need more explanation and service options, while lower-value segments are better served with shorter, more cost-efficient offers.
  • The Cancellation Reason Is Not Reflected: 'Price' and 'service' require different arguments and CTAs, otherwise the message feels generic.
  • Channel Preference and Timing Are Missing: Winback matters, but so does using the right medium at the right time. A uniform response ignores working hours, channel preferences and the appropriate copy length for each channel.
  • Channel Switching Without Losing Context Does Not Work: Omnichannel thinking means the conversation can continue on another channel without starting over.
  • Budget: Without segment routing, you do not optimise for cost to save, but distribute effort evenly, including where winback is unlikely.

Compliance & Governance: Consent, Opt-Out and Evidence

Governance is a mandatory part of automation. The risk is not 'good copy', but missing consent logic, unclear opt-out paths and insufficient evidence.

At the same time, consent must be managed by channel where separate permissions have been given for each channel. And opt-out is not merely a link in the footer, but a workflow event that stops or reroutes journeys.

Governance Components

Governance Component What Must Be Stored? Why Is It Important in Winback?
Consent per Touchpoint Channel and specific touchpoint 'Email is permitted, SMS is not' must be technically enforceable
Opt-Out as an Event Opt-out timestamp and channel Stops the journey or triggers rerouting and prevents violations
Evidence Exportable logs and durable evidence Without evidence, compliance and legal risks increase
Copy Versioning Templates, AI-optimised version, approval status and delivery channel AI may operate only within defined guardrails

Conclusion: AI Winback Becomes Scalable When Copy Optimisation Is Part of the Process

Customer winback for utilities becomes truly effective when copy is treated not as 'communication', but as a process-ready component with rules, measurement points and clear execution.

An AI agent is not a magic solution that automatically solves every problem. But it does consistently optimise copy for each channel within defined guardrails: quickly, repeatably and without every variant having to be created manually.

This is where the real leverage comes from: you combine measurability, reader guidance and workflow orchestration. You improve reachability and make automated winback journeys robust.

3 Takeaways for Decision-Makers

  • Standardise Instead of Improvising: Use copy quality as a quality gate, based on HIX or heuristics, rather than relying on gut feeling.
  • Guide Customers Instead of Persuading Them: DIWA and the one-action principle reduce friction and increase the likelihood of completion.
  • Scale Through Processes: Channel and segment rules, consent and fallbacks belong in the workflow. The AI agent is the accelerator, not the governance layer.

Quick Answers: The Most Important Questions About AI Winback

What Exactly Is an AI Agent in Customer Winback?

An AI agent in winback is a rule-based assistant that optimises copy for each channel, such as email, SMS or letters with QR codes, for example by making it shorter, more active and focused on one clear primary action. In parallel, the workflow manages delivery, deadlines, fallbacks and evidence, turning winback into an executable process.

What Is the Hohenheim Comprehensibility Index (HIX)?

The Hohenheim Comprehensibility Index (HIX) is a metric that assesses clarity objectively by combining several readability formulas and text characteristics into a comparable score. This makes it possible to standardise copy, compare variants and define channel-specific clarity targets as quality criteria.

What Does DIWA Structure Mean in Winback Messages?

DIWA is a reader journey that guides recipients towards action: Direct (reason), Interest (relevant context), Value (specific benefit) and Action (one clear primary action). In customer winback, DIWA reduces friction, prevents information overload and increases the likelihood that customers immediately understand and complete the next step.

How Do I Measure Whether Copy Optimisation Actually Works?

Effectiveness can be measured easily through A/B testing by segment and channel: variant A, the baseline, versus variant B, the optimised version, using leading KPIs such as delivery rate, CTR or QR scan rate, response time and completed flow rate, as well as lagging KPIs such as return rate, time to reactivation and cost to save for economic evaluation.

Mini Glossary

AI Agent (Copy Agent)
A rule-based assistant that optimises existing copy within defined guardrails, for example making it shorter, more active and focused on a clear primary action, and generates variants. It does not replace segment strategy or offer design, but improves execution.

Agentic AI
An approach in which a system pursues an objective and selects and executes steps within clear guardrails, for example adapting copy to a channel profile, creating a variant and submitting it for approval.

Channel Profile
Specifications for each communication channel, such as email, SMS or letter with QR code: objective, maximum length, structure, CTA format, tone and mandatory information.

One-Action Principle (Primary Action)
A rule that each message clearly prioritises exactly one primary action, for example 'Check option', and offers at most one secondary alternative, such as 'Contact / FAQ'.

CTA (Call to Action)
A prompt that makes the next step unambiguous, for example a button, link text or QR callout. In customer winback, CTA clarity is a key quality factor.

DIWA Structure
Leserführung in vier Schritten: Direkt (Anlass), Interesse (kurzer Kontext), Wert (konkreter Nutzen), Aktion (eine klare Handlung). Ziel ist Reibung zu reduzieren und die Entscheidung zu erleichtern.

Touchpoint
A specific contact point within the journey, such as email, SMS or a letter with a QR code, where a customer receives information and can respond.

Segmentation Logic
Rules used to divide customers into groups, for example by customer value, cancellation reason or churn risk. Segments determine tone, depth of offer, channel sequence and timing.

Channel Sequence (Contact Journey)
A defined sequence of channels and time windows, for example email → SMS → letter with QR code, including reminder cadence.

Fallback
A rule defining what happens when a step fails, for example due to non-delivery or no response: channel switch, reminder, manual clarification step or termination of the journey.

Opt-Out
Unsubscription from communications. Within the workflow, an opt-out is treated as an event that stops or reroutes the journey and is logged.

Leading KPIs
Early indicators of performance at each touchpoint, for example delivery rate, CTR, scan rate, response time and completed flow rate.

Lagging KPIs
Later-stage outcome metrics, for example return rate, time to reactivation, cost to save and net retention effects.