The next generation of CCM must be structured, accessible, trusted and AI-ready.
Why AI will change customer communications management
For decades, Customer Communications Management (CCM) has been built around a simple assumption: a company sends a communication, and a human being reads it.
A statement, bill, renewal, claim, reminder and so many more….
The document may be printed, emailed, published to a portal, rendered as PDF, delivered as HTML, archived for audit and/or made available through an app. The channels have multiplied, the templates have improved, and the personalisation has become more sophisticated. But the underlying assumption has remained largely intact.
The recipient is a person.
That assumption is now beginning to break.
Increasingly, the first reader of a customer communication may not be the customer at all. It may be the customer’s AI assistant.
The customer may not even open the statement. Instead, they may ask:
“What changed this month?”
“Is anything important different from last year?”
“How much do I owe, when is it due, and is anything unusual?”
“Give me the important bits.”
and my personal favourite:
“Summarise this for me”
That is not a small change in user experience. It is a structural change in the purpose of customer communications. CCM is no longer only about producing communications that humans can read. It is becoming about producing communications that machines can safely interpret on a human’s behalf.
We have just acquired a brand new type of consumer.
And it changes almost everything.
Accessibility was the warning sign
There is another reader that many organisations have struggled to serve for years: the assistive technology reader.
For a blind or partially sighted customer, a poorly structured PDF is not just inconvenient. It can be unusable. Missing tags, broken reading order, unlabelled tables, image-based text, meaningless links and absent alternative descriptions turn important communications into noise.
This problem is not new. It has been known for decades. But too often, document accessibility has been treated as a compliance obligation, a remediation exercise, or a specialist concern sitting at the edge of the main communications strategy.
AI changes that.
A screen reader and an AI agent are not the same thing. Their capabilities are different, and their users’ needs are different. But they depend on many of the same foundations: structure, sequence, labels, semantics, readable text, meaningful metadata and a clear relationship between content and meaning.
A document that is difficult for assistive technology to parse will often be difficult for AI to interpret reliably. A table that reads as word soup to a screen reader may also become a guessing game for an AI model. A visually impressive PDF that contains little semantic structure may look good to a person and still be a poor source of truth for a machine.
This is where accessibility stops being only an inclusion issue and becomes a customer experience, risk and competitiveness issue.
The European Accessibility Act came into effect on 28 June 2025 and covers important consumer services including banking and e-commerce. The European Commission describes the Act as a way to harmonise accessibility requirements for products and services across the EU, with banking services and e-commerce among the covered areas.
For organisations producing regulated or high-volume customer communications, that matters. The practical disciplines of accessibility, including semantic structure, logical reading order and machine-readable electronic documents, are increasingly aligned with the disciplines needed for AI-assisted interpretation. EN 301 549, the European accessibility standard for ICT products and services, is built to apply across a broad range of ICT, including websites, software and electronic documents.
The point is not that accessibility law was secretly written for AI. It was not.
The point is that the work needed to make communications accessible also prepares them for a world where customers use software to read, summarise and act on their behalf.
Accessibility may turn out to have been the rehearsal.
The document estate is not ready
Most organisations are “uncomfortable” with the state of their document estate.
A large benchmark by Allyant reported that nearly 95% of tested public-facing PDFs were inaccessible against WCAG 2.2 standards. That figure comes from a commercial accessibility vendor, so it should be read with appropriate context, but the scale of the problem is still hard to dismiss.
Academic research tells a similar story. A 2024 study of scholarly PDFs found that fewer than 3.2% satisfied all six tested accessibility criteria, while nearly three quarters failed all of them.
These are not obscure technical defects. They are failures in the basic ability of documents to explain themselves.
And this is before we ask AI to read them.
A human is likely able to compensate for bad structure. They can scan a page, infer context, notice a visual grouping, or use prior knowledge to work out what the document means. AI can do some of that too, and often impressively well. But in regulated customer communications, “often” is not good enough.
A bank statement cannot have an approximate balance.
An insurance renewal cannot have a guessed exclusion.
A payment reminder cannot infer the wrong due date.
A pension statement cannot summarise away a material warning.
A debt communication cannot make the next action sound optional if it is not.
A healthcare letter cannot turn a conditional instruction into a definite one.
This is where CCM becomes central.
The answer is not to let AI stare at a PDF every time and hope it extracts the right meaning. That may be acceptable for low-risk convenience use cases. It is not acceptable for governed communications where accuracy, provenance and auditability matter.
AI should not become the source of truth
Generative AI is probabilistic. That is not a criticism. It is how the technology works.
It is very good at language, explanation, translation, summarisation and pattern recognition. It can make complex information easier to understand. It can help customers navigate dense material. It can reduce friction and improve service.
But in customer communications, especially regulated communications, there is a hard line between explaining the truth and inventing it.
The total due should not be AI generated.
The interest rate should not be AI generated.
The renewal premium should not be AI generated.
The deadline should not be AI generated.
The clause reference should not be AI generated.
The mandatory warning should not be AI generated.
The customer’s rights should not be AI generated.
Those facts should be retrieved from an authoritative record.
Language may be generated. Explanation may be generated. A summary may be generated. A plain-English version may be generated. But the facts and figures themselves need to come from governed data, approved content, controlled business rules and traceable sources.
In other words, the future of CCM is not “AI reads the document.”
The future is “CCM produces the document and the AI-readable truth layer from the same governed source.”
That distinction matters enormously.
A modern customer communication should not only be a visual artefact. It should be a communication package. It should include the human-readable version, the accessible version, the structured facts, the metadata, the provenance, the clause identifiers, the business rules and the boundaries around what may or may not be summarised.
The PDF, the email, the portal view, the mobile notification, the chatbot response and the AI assistant digest should not be separate interpretations of the same event. They should be controlled renditions of the same truth.
That is the architectural shift.
Every AI derived summary is also a communication
There is a legal and operational reason to take this seriously.
When Air Canada’s chatbot gave a customer incorrect information about bereavement fares, the British Columbia Civil Resolution Tribunal found the airline liable for the misinformation. Air Canada argued that the chatbot was a separate legal entity responsible for its own actions, but the Tribunal rejected that argument and treated the chatbot as part of the company’s own website and service.
That case is not the whole legal future of AI, and it should not be overextended. But it is a useful warning.
A customer-facing answer is a customer communication.
If an organisation provides an AI-generated explanation, summary, recommendation or next step, that output cannot be treated as informal just because it was generated conversationally. If it influences customer understanding or customer action, it needs governance.
The same logic will apply in reverse as customers bring their own AI agents.
If a customer’s agent misreads a document, the organisation will want to be able to show that the original communication was clear, accessible, structured and machine-readable. It will want to show that key values, obligations and warnings were not buried inside inaccessible layout, ambiguous language or image-only PDFs.
Ambiguity will become expensive.
The safest position is not merely “we sent the document.”
The safer position is “we sent a document that could explain itself.”
CCM 2.0: from composition to governed interpretation
This is why AI creates such a significant shift for the CCM vertical.
Traditional CCM has focused on composition, personalisation, approval, delivery, archiving and channel management. Those capabilities remain important. In fact, they become more important. But they are no longer sufficient.
The next generation of CCM will need to manage interpretation as well as output.
That means approved content must become more granular and more semantic. Clauses, disclosures, figures, explanations and warnings need identities, not just positions on a page. The system needs to know what each part of the communication is, why it is present, where it came from, whether it is mandatory, whether it can be reworded, and how it relates to the customer’s data.
It also means that accessibility cannot sit downstream as a repair process. If accessible output is created by fixing documents after composition, the organisation has already lost control. Accessibility and machine readability need to be designed into the communication model from the start.
The same applies to AI readiness. It should not be an afterthought where an AI model is asked to inspect a finished document and derive meaning from layout. The structured meaning should already exist.
A CCM platform should be able to produce:
A human-readable communication.
An accessible communication.
A structured data layer.
A plain-language summary.
A machine-readable explanation of key facts.
A provenance trail back to approved content and source data.
Rules defining what may be summarised and what must be reproduced exactly.
Audit evidence showing which version was sent, through which channel, and which facts it contained.
That is not simply document generation. That is governed communication intelligence.
The new competitive surface
Gartner predicts that by 2028, 33% of enterprise software applications will include agentic AI, up from less than 1% in 2024. It also predicts that at least 15% of day-to-day work decisions will be made autonomously through agentic AI by 2028.
Whether those figures prove exact or not, the direction is clear. More software will act on behalf of users. More customers will rely on AI to read, compare, summarise and decide. More enterprise processes will expect content to be structured enough for automation.
That means machine legibility becomes part of the customer experience.
A poorly structured communication will not merely frustrate a human reader. It may frustrate the customer’s AI agent. It may cost more to process. It may produce lower confidence. It may require manual review. It may be excluded from automated comparison. It may be summarised badly. It may make the provider look difficult to deal with.
A well-structured communication, by contrast, becomes easier to understand, easier to verify and easier to act upon.
In the past, organisations competed on the clarity of the visible document. In the future, they may also compete on the clarity of the invisible layer beneath it.
How easy is the document to parse?
How reliably can key facts be extracted?
Can the summary cite the source?
Can the customer’s agent identify what changed?
Can mandatory wording be protected from loose paraphrase?
Can the organisation prove exactly what was communicated?
These questions are about to become CCM questions.
The winners will treat compliance as strategy
Some organisations will treat accessibility and AI readiness as grudging compliance work. They will remediate enough documents to reduce legal exposure, publish enough statements to satisfy policy, and bolt AI onto the side of existing processes.
Others will see the larger opportunity.
They will recognise that the same investment can serve multiple outcomes: accessibility, customer experience, operational efficiency, regulatory confidence, AI readiness and competitive differentiation.
They will not ask, “How do we make this PDF pass?”
Instead, they will ask, “How do we make every communication understandable by design?”
That is the real shift.
The future of customer communications is not just omnichannel. It is not just personalised. It is not just digital. It is not just accessible. It is not just AI-enhanced.
It is governed, structured, explainable and readable by humans, assistive technologies and AI agents from the same source of truth.
For years, the industry talked about getting the right message to the right customer through the right channel at the right time.
That remains true.
But a new requirement is being added:
The message must also be readable by the reader the organisation cannot see.
And in the age of AI, that invisible reader may become the most important one of all.
Sources
EU Directive 2014/55/EU and national e-invoicing mandates (Factur-X, ZUGFeRD) — hybrid human/machine document formats as regulatory direction of travel
Allyant, PDF Accessibility Index: 2025–2026 Benchmark Report — 94.75% of 644,854 public-facing PDFs inaccessible; sector breakdowns
Equidox, PDF accessibility survey — assistive-technology users report 67% of PDFs partially or entirely unreadable
ASSETS ’24 / arXiv, Uncovering the New Accessibility Crisis in Scholarly PDFs — under 3.2% of scholarly PDFs meet six baseline criteria
Hogan Lovells, EAA: what financial services firms should focus on — scope covering contracts and all forms of consumer communication
Dutch AFM, Accessibility of (digital) financial services — durable-medium PDFs must themselves be accessible; multi-sensory channel requirement
Level Access, EAA compliance overview — enforcement from 28 June 2025; first French lawsuits and the 2026 Carrefour ruling
Allyant, EAA and EN 301 549 / WCAG relationship — technical standards for electronic documents
Moffatt v. Air Canada, British Columbia Civil Resolution Tribunal (February 2024) — company liability for chatbot statements
Jeremy Howard / Answer.AI, the llms.txt proposal (2024) — machine-readable site versions for AI readers
Gartner, agentic AI adoption forecasts (2024–2025) — agentic capability in enterprise software by 2028
Research conducted with the use of AI Agents
Leave a Reply