There’s a pattern playing out across contact centres right now.

An organisation invests in a “best-in-class” AI platform. The business case stacks up. The demos are compelling. The promise is clear: faster resolution, lower cost-to-serve, better customer experience.

Six months later, the story has changed.

The bot is underperforming. Escalations are clunky. Agents are frustrated. Customers are repeating themselves. And internally, the same conclusion keeps coming up:

“The AI just isn’t good enough.”

It’s an easy conclusion to reach.

It’s also usually the wrong one.

Most organisations don’t have an AI problem. They have a design problem.

AI Doesn’t Fix Broken Systems, It Exposes Them

AI is often positioned as a solution layer, as something you can simply deploy to improve performance.

In reality, it behaves more like a multiplier.

If your environment is well-structured, AI can accelerate outcomes. But if it’s fragmented or inconsistent, AI doesn’t fix those issues, it actually amplifies them.

You see it quickly:

• Inconsistent knowledge leads to inconsistent answers
• Unclear processes create confusing journeys
• Weak escalation paths turn into high-friction handoffs

What gets labelled as “AI failure” is usually something else entirely operational with the gaps being exposed at scale.

The difference is visibility. AI surfaces these issues faster, more frequently, and more publicly than human interactions ever did.

The “AI-First” Trap

Most AI initiatives go off track before they even begin.

Organisations often start with the technology. They evaluate vendors, comparing their features, and selecting platforms before they’ve clearly defined the experience they’re trying to deliver.

That creates an “AI-first” mindset, where the tool starts shaping the solution.

Instead of asking:
“What is the right experience for our customers and our employees?”

The question becomes:
“What can this platform do?”

It sounds subtle, but it’s a fundamental shift.

You move from designing with intent to designing around constraints. The result is often a solution that looks impressive in a demo, but struggles in the real world because it was never grounded in how the operation actually works.

A design-first approach flips this. Define and design the experience first. Then use technology to enable it.

Contact Centres Are Ecosystems, Not Tools

Another reason AI struggles is because contact centres are treated like systems.

They’re not. They’re ecosystems.

Every interaction sits across a network of interdependencies through customer journeys, knowledge, processes, agents, channels, and backend systems. AI doesn’t sit outside this, it cuts across all of it.

When deployments are driven purely by technology, they tend to ignore the operational reality of the contact centre floor.

They don’t see how agents actually navigate systems. Where knowledge breaks down. What customers are really trying to achieve.

Without that context, AI is being designed in isolation. And isolated solutions rarely perform in connected environments.

The Foundations That Actually Drive Performance

When AI works well, it’s rarely because of the platform or the model. It’s because the foundations underneath it are solid.

Four areas consistently separate high-performing implementations from the rest.

Knowledge (KM) is one of the biggest. If your content is duplicated, outdated, or inconsistent, your AI will reflect that instantly. Well-structured, governed knowledge is non-negotiable.

Process design is equally critical. AI needs clarity. Where processes rely on human interpretation or workarounds, automation becomes unpredictable and fragile.

Contact understanding is often overestimated. Many organisations believe they know why customers contact them, but without a clear, data-driven view, it’s difficult to prioritise the right use cases or design effective journeys.

Then there’s architecture and integration. AI’s ability to retrieve data, trigger actions, and pass context depends entirely on how well it connects into your ecosystem. Poor integration is one of the fastest ways to limit value.

These may not be the most exciting parts of an AI program, but they are almost always the deciding factors to a successful AI program.

Design for Handoffs, Not Just Containment

Containment is one of the most common success metrics in AI.

On its own, it’s not enough.

Even well-designed AI won’t resolve everything. When it reaches its limit, the experience is defined by how effectively it transitions to a human.

This is where many implementations break down.

Customers are forced to repeat themselves. Context is lost. The interaction resets. What should feel like a continuation feels like starting over.

At the same time, agents inherit frustrated customers, with little visibility into what’s already happened.

A better approach is to design intentionally for these moments.

That means ensuring context is passed, customer intent is preserved, and ownership between AI and agent is clear.

A well-designed handoff doesn’t feel like failure. It feels like progress.

What a Design-First Approach Actually Looks Like

A design-first approach isn’t about slowing things down, it’s about getting the sequence right.

Start by defining the experience you want to deliver. Then align your environment to support it.

In practice, that means:

• Mapping key customer and service journeys before selecting technology
• Fixing knowledge and process gaps early
• Defining clear roles for AI and human agents
• Designing escalation and augmentation paths as part of the experience

Only then does the technology come into focus.

The result is not just better performance and experience, it’s a solution that can evolve, scale, and improve over time.

The Practitioner Truth

When AI underperforms, the instinct is to look at the model. Tune it, retrain it, or replace it.

But that’s often treating the symptom, not the cause.

The better questions are:
Where is the experience breaking down?
What are customers actually trying to achieve?
What are our agents needing to compensate for today?

Because in most cases, the problem won’t be solved with technology alone.

You don’t have an AI problem. You have a design and integration problem.

And once that’s addressed, the technology tends to perform exactly as it should.