Generative AI can make contact centre knowledge available faster than ever. Instead of an agent manually searching several articles, an AI assistant can potentially surface an answer, summarise information or suggest the next action while the customer conversation is taking place. But speed creates a new dependency: The AI needs something reliable to work from.
Our first article, Contact Centre Knowledge Management Software: What Customer Service Teams Should Compare, looked at choosing the technology used to capture and retrieve organisational knowledge.
The second, Contact Centre Knowledge Management Best Practices: Keeping Agent Information Accurate and Useful, focused on ownership, review cycles and preventing that information from deteriorating.
AI creates the next stage.
The question is no longer simply:
“Can the agent find the right answer?“
It becomes:
“Can the AI identify the right answer, use it appropriately and show the agent enough context to trust it?“
Treat Knowledge as the Foundation
Generative AI can present information fluently.
Fluency does not guarantee accuracy.
If an AI assistant is grounded in outdated, contradictory or poorly governed knowledge, it can potentially surface that information much faster than an employee could find it manually.
AI Knowledge Principle
Agent assist increases the speed at which good knowledge reaches employees — but it can also increase the speed at which bad knowledge spreads.
This makes knowledge quality more important, not less.
Before deploying AI across a knowledge environment, contact centres should understand which content is considered authoritative.
Define the Authoritative Source
Large organisations often have several versions of apparently similar information.
There may be:
- Knowledge articles
- Policy documents
- Product manuals
- Intranet pages
- PDFs
- Previous scripts
- Training materials
What happens when they disagree?
Contact centres need to establish which sources AI tools are permitted to rely upon and who owns them.
A useful governance question is: What information are we prepared to let the AI treat as authoritative?
If the organisation cannot answer that clearly, the AI may struggle to do so consistently either.
Clean Up Before You Connect
Connecting AI to an untidy knowledge estate does not fix the underlying problem.
Before deployment, identify:
- Duplicated articles
- Outdated information
- Contradictory guidance
- Missing owners
- Unreviewed content
- Obsolete documents
This is also an opportunity to simplify knowledge.
Articles written primarily for human browsing may contain lengthy introductions or duplicated information that makes retrieval harder.
Providers including Verint, Puzzel, Enghouse Interactive, Odigo and Zoho operate across areas of contact-centre technology, customer engagement, knowledge and AI-enabled service where the quality of underlying information is increasingly important.
Decide What AI Is Allowed to Do
“AI agent assist” can describe several capabilities.
An AI system might:
- Retrieve knowledge
- Summarise an article
- Suggest an answer
- Draft a response
- Recommend a next action
- Summarise the conversation
- Populate notes
These activities carry different levels of risk.
Suggesting an article for an agent to read is different from automatically sending an AI-generated response directly to a customer.
Contact centres should therefore establish appropriate levels of autonomy.
Governance Principle
The more directly AI can affect the customer, the stronger the controls around its output need to be.
Keep Humans in the Loop Where Appropriate
Agent assist should help employees make better decisions rather than encouraging them to accept every suggestion automatically.
Agents need to understand:
- What the AI is doing
- Where information came from
- When to verify it
- When not to use the suggestion
- How to flag incorrect output
Where possible, surfacing the underlying knowledge source alongside an AI-generated answer can help employees assess its reliability.
Providers such as Capacity, Kerv, IP Integration, MAINTEL/Mitel and Wavenet operate across combinations of communications, contact-centre, cloud and AI technologies where integration with existing agent workflows becomes particularly important.
Test High-Risk Answers
Not every knowledge topic carries the same consequence if the answer is wrong.
A mistake about store opening hours is inconvenient.
A mistake involving a financial product, vulnerable customer, safety issue or contractual entitlement may be considerably more serious.
Contact centres can therefore identify higher-risk knowledge domains and apply additional testing or approval.
Testing should include realistic customer language.
Customers rarely phrase questions exactly as knowledge articles are titled.
Ask:
Does the AI retrieve the right information when the customer uses unexpected terminology?
Manage Permissions
Not every employee (or AI tool) should necessarily have access to every piece of organisational information.
Knowledge environments may contain internal-only content, customer information or commercially sensitive material.
AI implementations therefore need appropriate access controls.
The NCSC’s secure-AI guidance recommends appropriate controls around APIs, models and data, alongside secure deployment and ongoing monitoring.
The same principle familiar from other enterprise systems applies:
Access should reflect what the user and the application actually need.
Protect Customer Information
Contact centre AI may interact with conversations containing personal or sensitive information.
Organisations should understand:
- What data enters the AI system
- Where it is processed
- Whether it is retained
- Whether suppliers use it for model training
- Who can access it
- How long it remains available
This should form part of supplier due diligence rather than being discovered after deployment.
The NCSC warns that AI systems introduce security concerns alongside conventional cyber risks and highlights confidentiality, integrity and availability as important considerations throughout the AI lifecycle.
Build Agent Feedback into the Workflow
Frontline agents are often the first people to discover that an AI suggestion is wrong.
Make that information useful.
An agent should be able to indicate:
Incorrect
Outdated
Not relevant
Missing information
This feedback can then reach knowledge owners or AI administrators.
Providers such as HGS and Capita Customer Management operate within customer-service environments where the combination of people, processes and technology remains central to delivering service at scale.
Feedback Insight
Every incorrect AI answer is also a signal about something that may need improving — the knowledge, the retrieval, the prompt, the model or the workflow.
Don’t Hide Knowledge Gaps with Fluent Answers
Traditional knowledge search has an obvious failure mode: No results found.
Generative AI creates a more complicated possibility: producing a plausible answer even when the required information is absent.
The NCSC notes that generative AI can produce incorrect statements as facts, commonly described as hallucinations.
Contact centres therefore need systems and processes that allow the AI to effectively say:
“I don’t have enough trusted information to answer this.“
That can be much safer than generating an unsupported response.
Measure Answer Quality
Traditional knowledge-management metrics might include:
- Searches
- Article views
- Zero-result searches
- Article usefulness
- Knowledge adoption
AI adds further measures.
These can include:
- Suggestion acceptance
- Suggestion rejection
- Correct-answer rate
- Escalation
- Agent edits
- Customer outcome
- Knowledge gaps identified
Speed should be included, but not allowed to dominate.
Measurement Principle
A faster wrong answer is not an improvement.
The target should be the fastest reliable answer that resolves the customer’s need.
Compare AI Performance with the Previous Workflow
A contact centre implementing agent assist should establish a baseline.
Before deployment, measure relevant outcomes such as:
- Average handling time
- Hold time
- First-contact resolution
- Transfers
- Agent search time
- Customer satisfaction
Then assess whether AI changes them.
But interpret results carefully.
Reducing average handling time is not necessarily positive if repeat contacts increase because customers received incomplete answers.
Metrics need to be considered together.
Keep Knowledge Owners Accountable
AI does not remove the need for content ownership.
Somebody still needs to decide:
- Whether information is correct
- When it needs updating
- Which version is authoritative
- When it should expire
- Whether AI may use it
Providers such as Sabio Group and other customer-experience specialists can support organisations implementing AI-enabled contact-centre environments, but governance remains an organisational responsibility.
The technology can retrieve knowledge.
It cannot decide the organisation’s policies for it.
Review AI Continuously
AI performance can change as:
- Knowledge changes
- Products change
- Policies change
- Customer questions change
- Models are updated
- Integrations change
The NCSC’s guidance treats secure AI operation and maintenance as an ongoing lifecycle requirement rather than a one-time deployment task.
Contact centres should therefore maintain ongoing review rather than treating successful launch testing as permanent assurance.
A Practical AI Agent Assist Checklist
Contact centre leaders should ask:
- Which knowledge sources are authoritative?
- Have outdated and duplicated articles been removed?
- What is the AI allowed to do?
- Which outputs require human review?
- Can agents see where answers came from?
- Have high-risk topics been tested separately?
- Are access permissions appropriate?
- How is customer information protected?
- Can agents flag poor suggestions easily?
- Can the AI recognise when trusted information is unavailable?
- Are accuracy and resolution measured alongside speed?
- Who remains accountable for knowledge quality?
- How will AI performance be reviewed over time?
Frequently Asked Questions
What is AI agent assist?
AI agent assist uses technologies including generative AI to support customer-service employees during interactions, potentially retrieving knowledge, suggesting responses, summarising conversations or recommending actions.
Can AI replace contact centre knowledge management?
No. AI still requires reliable information from which to retrieve or generate useful responses. Poorly governed knowledge can reduce the reliability of AI output.
Should agents automatically trust AI-generated answers?
No. Organisations should establish appropriate human oversight, particularly for higher-risk interactions, and train agents to understand when information requires verification.
How should contact centres measure AI agent assist?
Measures can include answer accuracy, suggestion acceptance, agent edits, first-contact resolution, handling time, repeat contacts and customer satisfaction. Speed should be considered alongside quality.
Product Guide
Capacity
AI-powered support automation platform providing knowledge, agent-assistance and workflow capabilities.
Website: https://capacity.com/
Capita Customer Management Ltd
Provides customer-management and outsourced customer-service solutions combining people, technology and operational delivery.
Website: https://www.capita.com/
Enghouse Interactive
Provides contact-centre and customer-experience technologies across communications, interaction management and service environments.
Website: https://www.enghouseinteractive.com/
HGS
Customer-experience and business-process specialist providing technology-enabled customer-service solutions.
Website: https://hgs.com/
IP Integration Ltd
Customer-experience technology specialist supporting organisations with contact-centre, communications and related solutions.
Website: https://ipintegration.com/
Kerv
Technology provider delivering cloud, communications, customer-experience and managed technology solutions.
Website: https://kerv.com/
MAINTEL / MITEL
Provide communications and customer-experience technologies supporting enterprise voice, collaboration and contact-centre environments.
Websites: https://maintel.co.uk/ | https://www.mitel.com/
Odigo
Cloud contact-centre platform providing customer-experience, interaction-management and AI-enabled capabilities.
Website: https://odigo.com/
Opus Technology
Business technology provider supporting organisations across communications, contact-centre and managed technology services.
Website: https://www.opustech.co.uk/
Puzzel
Cloud contact-centre platform providing customer-service, workforce, interaction-management and AI capabilities.
Website: https://www.puzzel.com/
Sabio Group
Customer-experience technology specialist supporting organisations with contact-centre transformation, AI and digital customer-service solutions.
Website: https://sabiogroup.com/
Verint
Customer-experience automation provider offering contact-centre, knowledge, workforce and AI technologies.
Website: https://www.verint.com/
Wavenet
Technology and managed-services provider operating across communications, cloud, cybersecurity and customer-experience solutions.
Website: https://www.wavenet.co.uk/
Zoho Corporation Limited
Business software provider offering customer-service, CRM, communications and wider enterprise applications.
Website: https://www.zoho.com/
From Knowledge Management to AI-Ready Knowledge
Across the three articles, the contact-centre knowledge journey becomes:
choose → maintain → activate
First, organisations need technology that allows knowledge to be created, governed and retrieved effectively.
Second, they need operational processes that keep that information accurate.
Only then does the third opportunity become truly valuable: allowing AI to use trusted knowledge to support agents in real time.
AI can dramatically reduce the distance between question and answer.
The challenge is making sure it does not reduce the distance between question and wrong answer at the same time.
The Contact Centre & Customer Services Summit connects senior customer-service professionals with carefully selected suppliers through pre-arranged one-to-one meetings, providing an opportunity to explore knowledge management, AI, agent-assist technology and wider contact-centre solutions.
Related Reading
This article follows Contact Centre Knowledge Management Software: What Customer Service Teams Should Compare and Contact Centre Knowledge Management Best Practices: Keeping Agent Information Accurate and Useful.
Together, the three articles cover technology selection, knowledge governance and AI-enabled activation, providing a practical roadmap from building a knowledge environment to using it safely at the point of customer interaction.
Sources
- National Cyber Security Centre – Guidelines for Secure AI System Development – https://www.ncsc.gov.uk/collection/guidelines-secure-ai-system-development
- National Cyber Security Centre – AI and Cyber Security: What You Need to Know – https://www.ncsc.gov.uk/guidance/ai-and-cyber-security-what-you-need-to-know
Image credit: https://unsplash.com/photos/a-black-keyboard-with-a-blue-button-on-it-kECRXz0m42A
