Why AI pilots fail in contact centres (and what to check first)

Why do so many AI pilots stall?

Most AI pilots start with the technology. A vendor shows a great demo, a business case promises fewer calls, and a pilot is launched to prove it.

The problem is that AI sits on top of everything else in your contact centre: your knowledge, your processes, your data and your people. If those foundations aren't right, AI doesn't fix them. It exposes them, faster and in front of your customers.

That's not a reason to avoid AI. It's a reason to check your foundations first. Here are the eight issues I see most often, and where to look before you start.

1. Nobody agreed the problem it was meant to solve

"Use AI to reduce costs" isn't a problem statement. Without a clear target, such as a specific contact reason, channel or customer group, a pilot can't succeed or fail. It just drifts.

Check first: whether you can name the contact reasons the pilot targets, and the current volume, handle time, resolution and cost for each. That's your baseline. Without it, you can't prove any value.

2. Your knowledge base isn't ready

AI answers questions using whatever your knowledge base says. If articles are out of date, contradict each other or live in agents' heads, AI will give customers confident, wrong answers.

Check first: ask three experienced agents where they really go for answers. If it isn't the knowledge base, fix that before you automate it.

3. You're measuring deflection, not resolution

Containment and deflection rates can look impressive while customers simply give up and call back, or complain. A contact that's deflected but not resolved isn't a saving. It's a cost you've moved.

Check first: whether your pilot measures repeat contact and complaints alongside containment. If customers come back within a few days, the AI didn't resolve anything.

4. You automated a broken process

Many contacts exist only because something went wrong earlier: a confusing letter, a missed callback, an unclear bill. Automating the answer to a contact that shouldn't exist makes a bad process run faster.

Check first: your top five contact reasons, and how many of them are caused by your own processes. Removing that demand is often a bigger saving than automating it.

5. The frontline wasn't involved

Your agents know exactly where customers get stuck and which answers cause confusion. Pilots designed without them miss that insight, and agents who expect AI to replace them have little reason to help it succeed.

Check first: whether agents and team leaders helped design the pilot, and how you've explained what it means for their roles.

6. The workforce plan didn't change

When AI handles the simple contacts, the ones left for your people are harder and take longer. Average handle time rises, occupancy changes and old forecasting assumptions stop working. Without an updated plan, service levels drop and burnout climbs.

Check first: whether your forecasts, staffing models and handle time targets have been updated for the contacts AI will leave behind.

7. Nobody owns the quality of AI conversations

Traditional quality assurance samples a handful of contacts per agent each month. AI conversations need checking too, but in many pilots nobody owns their quality, so errors go unnoticed until customers complain.

Check first: who reviews AI conversations, how often, and against what standard. It should be the same standard you hold your people to.

8. It isn't linked to your customer strategy

"AI-first" is often a cost strategy in disguise. When AI isn't linked to what the organisation is trying to achieve for its customers, it gets judged only on savings, and the customer experience quietly suffers.

In my experience, this is where many contact centre leaders struggle. Around three in four started on the phones, and many have only ever worked in one contact centre. Nobody taught them how to link the contact centre, or its technology, to the organisation's strategy.

Check first: whether you can explain how the pilot supports your customer strategy, not just your cost targets.

What should you have in place before an AI pilot?

You don't need a perfect contact centre to use AI well. But you do need the basics:

• A clear problem, with a baseline for the contact reasons you're targeting
• An accurate, well-owned knowledge base
• Measures of resolution and repeat contact, not just containment
• Processes that are worth automating
• A frontline that understands and supports the change
• An updated workforce plan
• Quality assurance that covers AI conversations
• A clear link to your customer strategy

Contact centres are a global industry with defined best practices, such as the global Contact Centre CX Standards, and they apply to AI just as much as to your people.

Not sure your contact centre is ready?

A call centre health check gives you an independent view of your foundations before you invest, so your AI pilot starts on solid ground. I'm vendor-neutral, with no technology to sell. Book a free 30-minute call to talk it through.