Avoiding customer service ‘black holes’ when embedding AI

Nearly all of our regular clients have told us they’re aiming to use more AI-driven solutions in 2024. Every single transformation plan or roadmap we’ve reviewed this year includes it to some degree.

But which use case is most likely to deliver efficiencies and cost-savings, without unnecessary disruption – or even the unintentional creation of a customer service ‘black hole?’

Typically, we see AI embedded in customer service for three reasons:

• Virtual assistance; automatic customer service handling, deflecting demand to contact centres
• Service augmentation; real-time analytics to give existing advisors more powerful tools and guidance. System-generated notes and suggested next best actions
• Insight; understanding trends, demand drivers, compliance, and customer call propensity.

While each option represents a pursuit of the same broad goals, your choice of application can prove detrimental to the customer experience, should you repeat the mistakes made by some other major companies.

When service fails – Virtual assistance gone wrong

Indulge me for a second, but I recently had a dreadful (but only mildly inconvenient) customer service experience with a food delivery company. Unfortunately, the courier made a mistake and had delivered my order to another home, before arriving at my door with very much the wrong order.

The courier had marked the order as delivered which, unknown to me at the time, made it impossible for me to get support from the company’s customer service team that evening.

Their approach to virtual assistance had ultimately failed, because not all customer outcomes and process-driven eventualities had been considered.

• No contact number was available via the app – only an automated chat that required a photo of your order to proceed
• A number was listed on the website, but it worked by matching the caller’s number to their account, and prioritising ‘in progress’ orders
• Calls were disconnected if your order wasn’t ‘in progress’, instead being directed back to the automated chat
• Social media interactions were also bot-driven – with no signs of a known route to resolution

A refund message did arrive via the app on the next day, but this had overlooked my needs as a customer. If all I required was a refund, then a sub-24 hour resolution would have felt highly efficient. However, the virtual assistance made it impossible to reach the outcome I actually needed – posing some literal ‘food for thought’ on AI’s effectiveness in customer service.

Creating safety nets for virtual assistance

I got stuck in some fatally flawed automated processes while trying to get food, which was inconvenient, but could be far more impactful for an energy or financial services customer, who requires immediate assistance.

As companies work on increasing the level of virtual assistance across customer journeys, it’s crucial to think about avoiding ‘black holes.’ Can your customers become trapped in virtual servicing under any scenario? If so, what are you doing to prevent this?

Previously, “fast-lane” phone numbers were used to give customers access to immediate human support, but were commonly shared online as friction-free routes to resolution prompting most to be removed.

Today, we see virtual assistance used as a major way to drive demand deflection. This means implementing live chat or automated service handling, to prevent the need for an agent to serve the customer. Or, when they do, equipping them to be sniper-focussed in addressing the customer’s specific need.

The food delivery company implemented several demand deflection systems, including:

• Decision tree-based digital contact
• Limited visibility of phone numbers
• Automated social media responses
• Blocking certain contact channels to specific customers (a bit nuclear)
• Scenario-based customer prioritisation

However, in this scenario, the company failed to see and respond to a customer that was clearly stuck in the system.

We believe that when using automated virtual assistance to this extent, there must be a safety net in place. This is akin to customers who previously button-bashed to bypass telephony IVR, and in the modern day, should be driven through real-time analytics – identifying any customers clearly going above and beyond to make contact.

After recognising these customers, they should be elevated into the right process channel. This can be supported by an intervention team, who make contact with the customer via phone, live chat, or email – lifting the customer out of any virtual assistance pathways.

Alternative: Prioritise service augmentation over virtual assistance

Of course, prioritising service augmentation is a different option, offering friction-free contact, followed by efficient and effective handling. Doing this means switching the focus away from complex automated processes for call categorisation and routing – instead leveraging AI to deliver insight that empowers your agents, and reduces cost to serve through quicker, sharper contact.

We’ve seen this work for high quality, online customer journeys, providing the majority of the customer base is digitally-engaged. If a customer drops out of the digital journey, and into the contact centre, then this is made visible to the agent.

Within these companies, we often see a very thin/light implementation of an IVR, as the customer’s need is already understood from their attempted journey.

When the customer does reach an agent, the agent is supported with system-driven prompts, and a summary to support the customer as efficiently as possible.

However, while this provides a better experience, some customers do recognise it as the most efficient path to resolution, and their propensity to contact remains high. The job of the agent then becomes different. Can they educate the customer to self-serve next time?

We’re continuing to support clients at both ends of the AI-spectrum, helping to set a broader strategic direction for AI-powered customer service, as well as tackling demand at a tactical level, through actions like call volume deflection.

For more on how we can help, contact Matt Turner or Jonathan Paton.