AI Agents in customer contact: currently limited deployment, but boundless ambitions

In the customer contact sector, there is a lot of talk about the use of AI. However, the number of organizations that actually apply AI in customer contact is limited, it is said. Is that image correct or does it speak of impatience or disbelief?

The most well-known use cases in customer contact are focused on agent assist – automated summarizing, providing dialogue suggestions and automated quality monitoring. This can be concluded from conversations, polls and conference visits. Contact centers that use GenAI mainly work with the customer conversation – or text. The output of the AI agent in the mentioned use cases also comes down to text: a summary, a conversation suggestion or an analysis. A completely different category is that the AI agent goes to work for you and carries out transactions or changes.

Assisting with conversations or carrying out processes

In practice, both types of agents will complement or even overlap, says Jan Verbrugghe, Senior Director Solution Engineering at Salesforce. Last month, that company launched Agentforce – an entire army of bots that can perform all kinds of tasks. Instead of distinguishing between dialogue agents and action agents, Verbrugghe prefers to talk about autonomous agents and assist agents. “The agents from Agentforce can work in the background – for example, picking up a process in the backend – but can also help in the front office or have a conversation. The assist agent is there for the employee, the autonomous AI agent is there for the end customer. Both use the same reasoning engine from Salesforce. Agentic AI is what we call 'acting on behalf'. So that is more than just looking up information and presenting it to an employee; it is also about executing.”

Examples

One example of Salesforce’s agentic AI being used is Open Table, a reservation site. Verbrugghe explains: “Among the frequently asked questions that the contact center receives is the adjustment of reservations that have been made. That is intervening in a process: as a third party on behalf of the customer, leaving the reservation at a restaurant by half an hour. Is there a slot available? Does the restaurant allow that?”

That sounds like further automating a self-service page, but Verbrugghe points out that an AI agent can work across channels, while a self-service page forces customers to go to a screen. “I think we will also see a shift here: from self-service as a page you have to go to, to arranging matters via a dialogue, where you can even switch from one channel to another. An AI agent in the background can then take into account that a customer has just registered online and that an app message may be about a first order. Self-service is often still a knowledge base or an FAQ in which you as a customer can literally search yourself; some organizations then want to put an AI layer on top of that, while you can also think of a completely different set of capabilities that we add to that self-service to make it multimodal.”

Adoption

So there is indeed a distinction between dialogue-oriented and transaction-oriented AI agents. What is the state of adoption of both flavors? Is it true that customer contact professionals predominantly see agent-assist applications that are focused on dialogue?

Verbrugghe has seen organizations start working with the 'autonomous agents' since November. "That is a process where you have to gain experience step by step: there are very few companies that say 'let's go wild' and turn everything on right away. Organizations start small - think of a home care organization in Sweden that schedules appointments for domestic help, and that develops an AI agent that can reschedule appointments. Secret Escapes, a British 'membership only travel agency' who offers trips with substantial discounts, will start working with an agent who can arrange matters with, for example, a hotel on behalf of customer contact employees - something that normally has to be arranged by an employee, for example by approaching the hotel, asking the question, and reporting back the answer. You help the customer faster and you also remove some of the workload from the contact center that adds little value.”

Who does what now

We started this conversation to find out what the adoption of AI in customer contact is: who is doing what with AI? Verbrugghe can only name a few customers for now, such as Secret Escapes, Open Table and Remarkable. Many organizations are now taking their first steps, he says. “Remarkable wants to capture the most frequently asked questions with a autonomous agent, and then tackle other subprocesses such as returns, where the AI agent assesses the request and sends a return label to the customer, but also sends a message to logistics to indicate that a package is coming. Adecco uses AI agents in the field of recruiting to analyze resumes for inconsistencies, among other things.

“In Northern Europe, we have an airline that is going to deploy an AI agent for multilingual support in the next few days. After that, we expect to offer it via AI agents and accept payments for upgrades and meals. There are many possibilities, but companies are choosing to implement it in small steps. They start small and if successful, other concrete processes are taken up. And that is very sensible. We have now done sessions with more than 100 customers who are all thinking about the use cases they can tackle.”

Deployment of AI Agent synonymous with 'fewer FTE'

At the last Spicy Contact Center Café During the panel discussion it was noted with some indignation that when new technology is introduced, the first result that is announced is usually about 'fewer FTE'. It was also suggested that new generations of customers would prefer to see faster and better service, otherwise they will leave you.

Verbrugghe does not recognise that first image of only cost savings. “Organisations want to grow, scale up, expand. Think of new markets for which the languages are not easily available, or expanding the service window because the end customer also works between 9 and 5 and therefore prefers to have contact in the evening. End customers increasingly want to be helped via multiple channels. An additional channel rarely leads to less contact, on the contrary. Here too, there are possibilities to 'turn on' digital channels, without it having an impact on the company.”

End of January: 1000 AI agents at work

What does Verbrugghe expect in terms of adoption over the next one to three years? Are we on the eve of a breakthrough or a turnaround? Verbrugghe thinks that will come fairly quickly.

“The biggest difficulty for most customers to take steps is formed by the underlying business applications and the data management. Adoption will not so much have to do with the technology, but more with the maturity of customer organizations. We can already do a lot with limited data, but productivity will increase if the processes and data flows are better aligned. Furthermore, Salesforce wants to introduce Agentforce 2.0 before Christmas, according to Marc Benioff you won't believe your eyes when you see that,” says Verbrugghe, “So it won't be a quiet Christmas.”