AI in customer contact: Five takeaways from our June round table

On Wednesday 3 June 2026, we brought together customer contact leaders from companies like ABN AMRO, PostNL, Univé, CZ, Vattenfall, ICS, Centraal Beheer and Landal for a round table discussion. At Medux’s headquarters in Utrecht we discussed AI and the future of customer contact.

All participants shared their challenges and experiences. From the discussion several themes emerged:

• How AI is changing customer contact
• What is and isn’t working
• What the leaders are doing concretely to get ahead

This article covers five takeaways from our session.

1. 01

   AI-native contact centers will look very different

   AI agents will not just take over part of the tasks of current contact center agents. They will become true collaboration partners.

   There are the obvious cases of AI directly handling customer requests or AI helping human agents with things like finding information, creating summaries or drafting responses.

   Less obvious, but maybe even more impactful are AI agents that can ask humans for input, which they can use to continue. Imagine AI agents requesting approval on a compliance-related flow, for discount strategies during a negotiation, or to get datapoints that they don’t have access to yet because of missing IT integrations. Suddenly the scope of use cases that can be handled with your AI agents expands to approach 100% and your service levels will improve with a much lower cost to service.

   This means that there will be smaller, yet highly capable teams in the contact center. They will receive more challenging questions and have more of a supervising and orchestrating role. A different rhythm is needed, with different rest periods around complex cases. Expertise will be extremely valuable (think for example about financial regulation certifications).

   This shift won’t happen overnight, but it might be faster than you think. Planning accordingly will help.

2. 02

   Start small, with “IT-light” use cases. Expand from there

   When it comes to AI, perfect is the enemy of good. What separates those that succeed from those that don’t is the fact that they got started within the existing limitations and iterated from there.

   Missing IT integrations, limited channels, undocumented knowledge. Everyone is facing the same challenges, but they should not stop you. Identify what you can do now, while working in parallel to address the long-term opportunity. There is a good business case in that first part.

   Examples of how organizations got started are:

   • Replacing their IVR system with a voice agent to improve customer experience and routing accuracy from 60 to 95%.
   • Answering a selection of FAQs with fallback to a human agent for any other topics, resolving 40% autonomously.
   • Sending data gathered as an e-mail to an RPA robot or human agent for processing, saving 90% of handling time.
   • Auto-drafting e-mail responses with a human in the loop before sending.

   Early results can be used to build business cases for larger investments. To build the required APIs for customer authentication, direct data processing and documenting knowledge on other FAQs.

   As these unlocks become available they expand their AI agents on top of all the lessons they have already learned by being in production for weeks or months.

3. 03

   Create a clear and rapid governance process

   The most successful organizations have set up a clear path for governance of their AI activities. They often have a single group of decision makers ("AI board") that meets regularly with the full mandate to approve experiments and use cases. Organizations that lag behind often have multiple groups, with overlapping responsibilities that each need to approve every experiment and use case.

   Critical for success is that this board contains all required decision makers with the mandate for approval. Typical participants are leads on IT, security and the various involved business units.

4. 04

   Build an AI innovation team that embeds itself in the business

   AI capabilities are hard to build, and you cannot expect business teams to develop them alongside their core activities. Having a central team with the skills and vendor relationships to drive the identification, design and implementation of AI use cases has worked very well for several organizations.

   The crucial difference with other typical center of excellence setups is that the team is brought to the work instead of the work brought to the team. Basically, this team goes to the various business units and works with them to identify cases, run experiments and bring use cases into production. Always focused on real value and business results.

   The power of this mechanism is that you build expertise that grows over time and is spread throughout the business. It also solves many of the typical capacity constraints that stifle innovation at the start.

   Best practice is to have the team train the business unit staff (for example contact center staff) to manage and iterate on the AI use case over time, so skills are transferred and the new way of working becomes embedded.

5. 05

   Experiment to experience the difference between vendors

   While almost every vendor seems to offer roughly the same at the surface, the performance differences can be significant. The best organizations have found a way to quickly test different options before committing to a solution for the long-term.

   Make sure you evaluate the options on the market, including building yourself, to get a good feel for the differences. Run benchmarks on realistic flows, preferably in production, to really compare.

   Contrary to conventional wisdom, start with the hardest problems first. For example assess performance on complex flows, start with voice instead of chat, try difficult languages and accents. This stretches the possibilities of each platform and shows you where the limits are. It prevents you from finding out about them only when expanding to these more challenging cases.