Getting AI Right Starts With Getting Knowledge Right: Our Takeaways From eGain Solve 26

For all the excitement around AI in customer experience, one message cut through clearly at eGain Solve 26 event in London: if your knowledge is a mess, your AI will be too.

Across two days of customer stories, product discussions, partner conversations and practical examples, the event kept returning to the same idea: AI success in contact centres doesn’t begin with another chatbot. It begins with trusted, structured, governed knowledge.

In this recap, we’ll look at the biggest themes and insights CCH Director Xander Freeman gleaned from the companies at eGain Solve 26: why knowledge management has become the foundation for successful AI, how leading organizations are applying it in real contact centre environments, and what they need to get right before AI can truly deliver on its promises.

The Problem Isn’t AI, It’s Companies Foundations Underneath It

In Xander’s conversation with eGain CEO Ashu Roy, he was very intentional not to frame the industry as getting AI “wrong”. In fact, he started from a more optimistic place, saying organizations are right to see AI as a way to improve customer satisfaction and drive efficiency.

The issue, he argued, is what happens when businesses rush too quickly towards value.

That point was reinforced by one of the event’s opening presentations: “Knowledge is not unstructured data. It is instruction for your AI systems. Treat it so.”

In this case, language matters. If knowledge is instruction, then the quality of that instruction determines what AI can do.

Bad instructions create bad answers. Outdated policies create risk. Fragmented information creates inconsistency. And in a contact centre, inconsistency impacts everything. It becomes longer handle times, agent frustration, poor customer experiences, compliance issues and a loss of trust.

A slide set from the event set out the practical advice plainly:

• Fix your knowledge foundation before building your production AI house.
• Assess your knowledge maturity and partner with a proven, best-in-class provider.
• Don’t deliver yet another chatbot.
• Deploy an end-to-end, trusted AI knowledge pipeline in production for a representative use case in 30 days.

It was a refreshingly grounded message. Less “AI will transform everything overnight”. More: “please sort out the plumbing before you turn the tap on”.

The Knowledge Maturity Gap Is Bigger Than Many Want to Admit

One of the most useful frameworks from the event was eGain’s AI Knowledge Maturity Scale, which showed five levels of readiness: Ad Hoc, Managed, Designed, Optimized and Transformative.

At the “Ad Hoc” or lowest level, organizations are dealing with knowledge silos, no formal capture or curation process, and no consistent taxonomy.

By the “Managed” stage, there may be a single source of truth, but no formal writing guides or structure, with duplicate and inconsistent information still floating around.

The higher levels move towards consistent structure, metadata, trusted knowledge, contextual personalization, orchestration, automation and, ultimately, knowledge-driven generative AI automation.

The reality check came from another presentation showing where enterprises currently sit:

• 80% are at low maturity, with siloed, unstructured, stale knowledge that isn’t AI-ready.
• 16% are at medium maturity, with some structure but inconsistent governance.
• Only 4% are at high maturity, with governed, AI-ready knowledge that is continuously improving.

Roy suggested that the biggest movement over the next year will likely come from that middle 16%, the organizations that recognize the problem but need a clearer way to solve it.

The real commercial opportunity there is not convincing people that AI matters (because that argument is largely won), but in helping them understand why AI won’t work properly until knowledge becomes a board-level operational priority.

We were lucky enough to sit down with some of the speakers and other CX innovators during the event, all with their own individual use cases and success stories.

You can catch the highlight reel below, but we’ve chosen to put the spotlight on three specific stories that we think have key knowledge management lessons for our audience to take away.

Cathay Pacific Case Study: When Knowledge Becomes Mission-Critical

The Cathay Pacific use case brought the whole conversation into sharp focus.

Tannie Kwong, Head of Customer Support at Cathay Pacific, described airline customer care as one of the most complex service environments imaginable. His team handles suggestions, enquiries, complaints, baggage claims, disruption, customer social media and more, with hundreds of people spread around the world.

That complexity becomes even more serious when you consider the emotional context of travel. Kwong pointed out that customers may be flying for funerals, to care for end-of-life family members, or dealing with aircraft diversions involving sick passengers. In those moments, advisors need accurate information quickly.

Cathay Pacific began with a proof of concept around 12 months ago. The goal was to move beyond old-style query-based systems where agents type in search terms and still struggle to find what they need.

Two months before our conversation, the airline launched its platform.

One month later, it opened it up with AI capabilities. Now, hundreds of agents around the world are using it, supported by engagement initiatives and prizes for those using the platform most effectively.
The most interesting part, though, was Kwong’s reflection on what he would do differently.

Rather than focusing mainly on converting existing manuals, policies and procedures into a new platform, he said organizations should think harder about what AI makes possible next.

His example was brilliantly simple: why should an airline website require customers to click through “20 buttons” to find the information they need? Why couldn’t the website become one large knowledge management solution, where customers type what they need and receive the right answer?

In other words, knowledge management isn’t just there to help agents any more. Done well, it has the potential to become the foundation for a better customer experience everywhere.

Key Takeaways

• Complex CX environments need fast, accurate and governed knowledge.
• AI should not simply digitize old manuals and rigid navigation.
• Agent adoption matters, and Cathay Pacific’s gamified approach helped build engagement.
• The future of customer access may look much more like conversational search than traditional website menus.

BT Group Case Study: Bring Knowledge Back to Basics

BT Group’s Hamish Corry offered another practical, grounded example through the launch of Instant Answers, an AI search capability for call centre advisors.

His core message was beautifully direct: “It’s really let us bring knowledge back to basics.”

Previously, BT had been “very workflow heavy”. To make knowledge useful for both advisors and AI, the team moved towards smaller, clearer articles with a more direct narrative. That meant reworking the topic structure around what people actually ask rather than how internal teams prefer to store information.

This is such a common trap in knowledge management. Internal structures often make perfect sense to the people who built them, because they’ve been staring at them for months. But they don’t always reflect how agents search under pressure, or how customers phrase their problems.

Corry said BT went back to basics by asking: what are people looking for?

The other smart part of the project was how closely it involved frontline users. Subject-matter experts were brought in from day one. They joined training sessions, wrote content, worked with the design team and helped create a feedback loop as the content evolved.

“It’s kind of built by advisors for advisors,” Corry said.

That is exactly the kind of practical detail that separates a decent AI project from a usable one.

Key Takeaways

• AI-ready knowledge needs clear, searchable, well-structured content.
• Taxonomy should be based on user questions, not internal filing systems.
• Frontline advisors should help shape the knowledge base from the start.
• Simpler content can be more powerful than complex workflows when AI search is involved.

PMI Case Study: Turning Customer Care Into a Profit Centre

Imran Awan from Philip Morris International brought a different lens: commercial transformation.

He described PMI’s shift from a traditional FMCG organization into a more CX-led business, with customer care moving from “a cost centre to a profit centre”. To support that, the company needed to become “AI ready and future-proof”.

PMI had built an in-house agentic AI capability, but needed trusted knowledge sitting behind it. As Awan explained, “we needed trusted knowledge infrastructure” that could be exposed “in a safe and scalable way”.

The result was a pilot combining the eGain knowledge layer, PMI’s in-house capability and MuleSoft, with more markets planned.

His advice to anyone starting the same journey was one of the most useful lines from the event: ask ten people what knowledge management means, and you may get ten different answers. So start with the business outcome, then work backwards.

That’s the grown-up version of AI strategy. Don’t start with the tool. Start with the commercial goal, the customer goal, the employee goal, and the governance requirement. Then decide what knowledge needs to exist to make that possible.

Key Takeaways

• AI projects stop being interesting very quickly if they can’t prove commercial value.
• Internal AI capability still falls apart without trusted knowledge underneath it.
• “Knowledge management” is too vague to be a strategy on its own. The business goal has to come first.
• The future probably isn’t one giant AI platform. It’s interconnected ecosystems stitched together properly.

Final Thoughts

The real story from eGain Solve 26 wasn’t that contact centres need more AI. That talking point has been run into the ground for the past two years.

The sharper message was that AI is only as useful as the knowledge system underneath it, and the AI race will be won by the best-instructed organizations.

The organizations making progress are not treating knowledge management as admin, housekeeping or a back-office content project. They’re treating it as essential infrastructure.

If AI is going to speak to your customers, assist your agents and represent your brand, it needs to be taught properly. And AI’s teaching starts with your knowledge.

Author: Xander Freeman
Reviewed by: Robyn Coppell