Picture this: you’re meeting a valued client to resolve a problem with your product. To protect the relationship – and your brand – you need to fix it for them quickly and seamlessly.
But, halfway through the meeting, the team member tasked with fixing the problem starts randomly interrupting the client, misunderstanding basic questions and inventing answers that don’t make sense and are completely untrue. When challenged about their lies, the team member denies any wrongdoing.
The truth is, you’d probably never hire someone so unreliable. Not if you wanted to have customers.
Yet this is the risk too many organisations are taking when they rush to deploy fully agentic AI in contact centres before these products are ready – and before their organisations are ready.
Contact centres are messy, human environments. Customers don’t speak in scripts or follow neat, deterministic patterns. When booking appointments, for example, customers answer questions with phrases like “I dunno… um first thing or end of day” or “ASAP”. These phrases require context, judgement and negotiation. Humans can handle that nuance. Most deterministic AI can’t, and while agentic AI is great at inferring meaning, it sometimes struggles to apply the right rules.
Despite that, organisations are running expensive proof of concepts of agentic AI that has all the latest bells and whistles, including one case that even went as far as including background noise of a contact centre into the automated agent – that noise annoys us when calling a real call centre, why would we want it in a bot? The problem is that testing for all the possible ways it can fail becomes exponentially larger the bigger we make the bot. That means if we start with a large vision for the bot on day one, we have a huge task to confirm it will stay on script.
This is because some organisations are being sold an unrealistic dream by vendors who want to redefine the problem to fit their product and don’t really grasp the reality of contact centres. This is built around two false premises.
The first is that targeting full automation is the best way to get ROI and transform performance – switch on a bot in October and cut call volumes by 50 per cent in November. However, the truth is that going fully agentic requires a significantly longer development cycle – often more than six months – which is leaving any savings you could achieve in the short term on the table. This focus on automation above all else is wasting your customers’ time and torpedoing your customer experience.
The second false premise is that the only way to benefit from Agentic AI is to fully automate tasks from start to finish – resulting in no need for the call to hit the call centre. This is failing to see that Agentic AI can be used to improve efficiency, create better routing, and ask deeper questions to help the contact centre agents reduce their AHT.
I’ve worked in this industry for more than 30 years, including a decade building AI-driven contact centre systems. While the technology has changed significantly during this time, the basic structure of human conversation has not.
All contact centre agents must be able to answer three questions: Who are you? What do you need? Can I do that for you?
The biggest opportunity in contact centres isn’t aiming for full automation overnight. It’s getting the basics right across “who are you and what do you want” first, before improving your CX incrementally and using AI where it genuinely helps, not just where it looks impressive.
If you’re considering making your entire contact centre fully agentic, remember that the old adage still applies – if it sounds too good to be true, it probably is.
If you wouldn’t hire a staff member who lies and hallucinates, don’t unleash one on the poor unsuspecting customers who call your contact centre with AI that is not fit-for-purpose.