Will 2025 be the year in which AI in customer contact will prove itself? There is an overkill of supply. Which applications have the best credentials to break through and what obstacles will we encounter in the coming year?

Ziptone spoke with Leon Driessen (solutions consultant Frontline Solutions), Stuart Dorman (chief innovation officer Sabio) and Stef van den Oever (Group Practice Manager at Nextview Consulting). 

 

Eighty-five percent of customer service leaders will explore or pilot a conversational generative AI (GenAI) solution by 2025, according to research from Gartner. 44% of 187 customer service and support leaders surveyed indicate that they are exploring a customer-facing GenAI voicebot. 11% are testing this technology and 5% have already implemented a customer-facing GenAI voicebot.

Those predictions look a bit paltry compared to the many billions that have been pumped into GenAI solutions. Now we are getting to the point where it actually has to yield something, says Leon Driessen, solutions consultant Frontline Solutions. “In addition to the low-hanging fruit such as agent-assist applications, I expect more advanced applications in the coming year. My idea is that within a few years you will be able to say 'these are good agent conversations' and that AI can train itself to conduct conversations independently based on real-time listening. That is a pipe dream, but I don't think too far into the future: all the technology for this is already available. Controlling it is a concern, but your own knowledge base and RAG are important tools here.”

'GenAI is becoming a commodity'

According to Peter van der Putten, Pega’s Lead Scientist and Director of AI Lab, GenAI core models will become “a commodity” in the coming year. The sheer volume of models and solutions will lead to fierce competition and pricing pressure; smaller independent developers will drop out due to lack of funding or high operational costs. But Pega is also championing what the company describes as “the comeback of non-generative AI,” including applications such as process mining, predictive modeling, and real-time decisioning. “But the biggest innovation in 2025 will come from technology that makes AI itself more autonomous, namely agentic AI,” Van der Putten said in a recent Pega post.

Jarno Dursma recently posted on LinkedIn some critical comments on the reasoning capabilities of advanced AI solutions. He referred to research from Apple that exposed weaknesses. 'Reasoning language models' – AI systems that could solve complex problems step by step – were said to score well according to standard benchmarks. But as soon as the questions from those benchmarks were adjusted even slightly, the quality of the answers dropped dramatically.

Although solutions for AI in customer contact are developing rapidly, the pace at organizations is much slower, Driessen notes. “The security officer alone wants to know what is involved. For the time being, contact center managers are still busy with agent-assist applications, for example. Nevertheless, I expect that AI in customer contact will remain hot next year. And of course, the fact that various contact center platforms will stop supporting their on-premise packages in the coming months also plays a role. Contracts are expiring. So there is plenty to do.”

AI like water from the tap

AI is on its way to becoming a 'general purpose technology', like electricity and the internet, says Stuart Dorman, Chief Innovation Officer Sabio. “The question of whether something is powered by the internet or electricity has long since ceased to be relevant, and that is also the direction AI is heading. The big players – big tech and chip manufacturers – are reinvesting half of their profits in R&D. There is a huge arms race going on. Whether the limits of AI are in sight, as Google’s Sundar Pichai suggested, or whether we are just at the beginning, is completely unclear at the moment. The underlying models are becoming a commodity, but the real development and implementation work is in the applications that are based on them. Think of Agentforce or other applications. We still have a long way to go in that area, not only to make the applications a success, but also to train everyone. The speed at which everything is happening now increases the chance of a crash if the expectations that are now being created are not met quickly enough.”

“I expect that AI will get the place it deserves in the coming year,” said Steve van den Oever, Group Practice Manager at Nextview Consulting. “For the customer contact industry, that is one of the enablers to improve customer contact. The customer can be helped better or faster, the employee can be relieved of tasks that do not contribute to someone's career or job satisfaction."

Impact of AI on employee and customer

In the average contact center, despite all predictions, 70% of contacts are still telephone, 20% email and 10% other channels. Automating a significant portion of that 70% has quite a few consequences for the customer, says Dorman. Implementation should not be underestimated either. “AI in customer contact is not something you can just turn on.”

“Automatic summarization? An agent needs that after-work time to catch their breath.”

With AI, you can increase the productivity of customer contact employees. For example, by automatically summarizing conversations. Because this eliminates most of the after-work time, this can have a negative impact on the well-being of employees, according to Dorman. An agent needs that after-work time to catch his breath. “Before you make a decision: also check the track record of a service provider. Keep in mind that this can be difficult, because there is not that much experience. We are only at the beginning. Incidentally, there are also solutions that we have been working with for years and that, even if we do not call them that, come down to AI.”

The implementation path depends largely on what you want an AI agent to do, says Van den Oever, but you can indeed develop a concept in a month. “For example, an AI agent that helps with contact deflection. Think of a live agent that has to combine information from different sources to provide a good answer to a customer question. That is exactly the sweetspot of GenAI. Anyone who has a good knowledge base can set this up relatively quickly.”

What increases the pace of development is the fact that you can develop AI agents using natural language instead of coding. The downside of this is that coding can be tested and then generates standard output, while AI agents do not. Therefore, developing AI agents is a matter of continuous development, also because changes in the various supporting systems can lead to different behavior of the agent, warns Van den Oever.

The Costs of AI: A Headache in the Making?

Gartner analysts also warned at the Australian ICT market research conference that AI use could be alarmingly more expensive. One of the reasons lies in the frequent adjustment of the rates: generally by leaps and bounds. But they also point to insufficient overview of the AI ​​use within an organization. Something similar happened with the transition to cloud computing, partly because SaaS solutions could easily be purchased outside the IT department. Cloud bill shock is a possible outcome of this.

Van den Oever points to the shift that is visible in the tariff models: from user-based billing to consumption-based billing. “The former leads to well-predictable costs, the latter does not. There is a visible shift from user-based billing to consumption-based billing. I notice that companies are reluctant to make decisions because of this. Next year it will become clear whether companies can create more clarity for themselves in this area. In addition, large providers often have multiple billing models for different services, which are all related in AI applications.” This does not make calculating the TCO and ROI of AI in customer contact any easier.

Driessen also points out the financial uncertainties surrounding the use of AI in customer contact. “The tooling will become broader in the coming year: not only for agents, but also for supervisors and administrators. My tip for contact center managers is therefore that they do not get seduced too quickly. First look at what you really need, then look at what you already have in-house and certainly do not forget to look at the cost structure. We notice that many parties find it difficult to estimate the costs in advance. This also applies to an application such as automated summarization - it is an additional reason to start with small steps and to look carefully at your realized savings. Some contact centers hardly have any after-work time and then automated summarization quickly becomes an additional cost item.

Turning AI On and Off? That's Not How It Works

What is particularly difficult in this context is that you cannot always simply switch off AI agents if your usage budget is exceeded. Do you then go back to your old situation? At the moment, for example, you can purchase bundles of AI conversations – and the more you purchase, the lower the price per conversation.

Van den Oever says that many companies still find it difficult to accurately predict the consumption volumes for AI in customer contact. “Forecasting is becoming complicated. It would help if providers came up with better and simpler billing models.”

Read also: Stef van den Oever on AI in customer contact. “Look beyond FTE reduction"

As for the pricing models, Dorman expects that they will be adjusted in the foreseeable future to provide insight into ROI. On average, handling a call costs roughly six euros, for a chat that will be three to four euros. The AI ​​applications that will replace these will have to be considerably lower in price. The purchase variants are now quite high and that is excluding the additional costs for implementation and management.

AI oversight: 'an accident waiting to happen'

As organizations increasingly use AI in customer contact, oversight is becoming increasingly important. Various specific European and national laws apply to the use of AI in business processes. The European AI Act, which came into effect in August 2024, is leading. Its implementation is a step-by-step process, but there are at least clear frameworks within the AI ​​Act. The supervision The use of AI is a different story.

This is spread over several supervisors. This requires a sectoral and integrated approach, according to a joint final advice of the National Inspectorate for Digital Infrastructure (RDI) and the Dutch Data Protection Authority (AP) of November this year. The choice not to add a new AI counter to the supervision should prevent the situation from becoming too complicated for companies.

That seems like a sensible choice, but it can also lead to conflicts. DNB is the designated supervisor for the financial sector. But it is also DNB that demands that banks and insurers they sift through their data to comply with all kinds of other laws. The various supervisors must therefore work together and coordinate everything among themselves, and that is a risk. Supervisors have lost a lot of executive power in recent years. Cooperation in complex ecosystems is not the strongest point of the Dutch government. For example, it could mean that supervision will be substandard in the coming years, which means that companies can consciously or unconsciously make mistakes for a longer period of time. Moreover, there is a rush: the AI ​​Act has already come into effect, as mentioned. And from February 2025, the first set of rules (the ban on the riskiest AI systems) will come into effect. By then, supervision must be in order. (Ziptone/Erik Bouwer)

Also read:

Fear of unpredictable costs with AI is unnecessary

 

Featured, Technology