Want to achieve quick results using AI in customer contact? Technology is evolving rapidly and the possibilities are greater than ever, but a successful implementation of an AI solution is rarely just a technical challenge. Consultants Carmen Leegwater and Jasper Joosten from Contact Center Live observe that good preparation, adoption, and further development are crucial for ultimate success. In this article, they share the most important lessons from practice.
“Whether it concerns an internal chatbot, an AI assistant that summarizes conversations, or an application that automatically evaluates customer conversations: implementing AI seems like introducing new software. But there is one important difference. The real work often only begins after going live. AI requires continuous fine-tuning, maintenance, and management to continue delivering the desired quality and results,” says Carmen Leegwater.
Like Jasper Joosten, she is a consultant at Contact Center Live and guides AI implementations at P1, TBAuctions, Dreambeads, Captin, Qbuzz, and Kantar/Candex, among others. “Every organization starts with a different ambition: one wants to reduce the pressure on the contact center, another wants to improve quality or better support employees. Yet, in almost every project, we see the same success factors and the same pitfalls.”
Motivations for getting started with AI?
“Most organizations want to deploy AI to further increase the efficiency of processes in the contact center. Customers want their processes to run structurally faster and more efficiently,” says Joosten. “For example, by reducing manual tasks and thereby saving time or money. Better onboarding of new agents, a higher first-contact resolution, preventing ticket build-up, and improving 24/7 availability also appear as practical objectives. So it is by no means always about lofty strategic goals. Although, of course, there are also companies that aim for higher customer satisfaction.”
Faster results? Get your homework in order first.
Anyone who wants to see results quickly (or faster) during the rollout of an AI application must have done their homework beforehand. Not every company takes care of this. Joosten: “We often start with a strategic workshop in which we clarify the goal: what should AI concretely deliver? What problem do you want to solve? Next, we fill out an inventory sheet together to map out the underlying processes, content sources, look & feel, and desired languages. In doing so, we also determine which tasks AI can take over and which tasks agents will continue to perform themselves.”
Review knowledge base
For clients who already possess a well-maintained knowledge base, there is usually a good foundation for AI implementations, says Joosten. “However, many organizations do not realize beforehand that the quality of their content is decisive for the success of AI. That is why we take on the role of advisor, clearly explain the prerequisites, and guide them step by step through all the preparations. Only then can we get started with the technical setup of AI solutions. We also set up the training of agents and the alignment of escalation procedures.”
How should AI represent your brand?
The client must not only make the content sources available (the knowledge base, the website, and internal documentation) and formulate a vision of how the AI agent represents the brand. For the deployment of Zendesk Copilot, for example, it is also crucial that work processes are clear, so that the assist function can truly support employees. Leegwater: “That all sounds straightforward. In practice, however, we encounter situations where the knowledge base, for instance, is too limited or too superficially structured, rendering the AI solution insufficiently capable of making connections.”
Take your time for a start-up period
“Another thing is that you have to invest time in a start-up period. Of course, you can lay the foundation as well as possible, but fine-tuning requires time and effort from the client. We see this, for example, with the automated handling of tickets, where the results can be meager in the beginning,” warns Leegwater. “An application like the Zendesk QA module makes it possible to have 100% of your tickets assessed by AI, but that requires some preliminary work, which impacts your ROI. It is precisely during this phase that we provide intensive guidance to organizations. By monitoring performance together and optimizing the AI solution step by step, not only does the quality of the outcomes improve, but also the confidence of employees in the technology.”
“The importance of continuous fine-tuning is often underestimated anyway,” adds Joosten. “As mentioned, AI is not a one-off implementation. Implementation times can also vary significantly. Basic implementations easily take two to eight weeks. More advanced projects with multiple integrations or a significant amount of customization can require more time. But it is primarily testing, improving, and training agents that takes the most time.”
Two-thirds are still at the beginning, according to McKinsey – Most organizations are still in the experimental or trial phase when it comes to the deployment of AI, concluded McKinsey end of last year. For roughly two-thirds of the respondents, their organization has not yet started scaling up AI within the enterprise and is at least experimenting with AI agents. Slightly more than a third measure an impact on profit at the company level.
Similar figures come from international research by the Boston Consulting Group (BCG) among more than 1.800 C-level executives in nearly 20 countries: only a quarter of companies say that AI has generated significant financial value.
Redesigning workflows is a key success factor, according to McKinsey: half of the high-performing AI companies plan to use AI to transform their business, and most are in the process of redesigning workflows. Respondents also have varying expectations regarding the impact of AI on their organization's total workforce size in the coming year: 32 percent expect a decrease, 43 percent no change, and 13 percent an increase.
Specifically within customer service, observes BCG that 90% of C-level executives and senior managers in customer service are working on AI initiatives. 87% plan to invest additionally in this over the next 12-24 months.
Be aware of the impact on agents
Joosten: “The impact of AI on customer service representatives is often underestimated. Their role does not change overnight, but evolves step by step. Agents are increasingly taking on a controlling or corrective role, where every change requires a new period of adjustment. That makes an AI implementation fundamentally different from a traditional software implementation. Therefore, we pay a great deal of attention to involving employees in that change.”
“Support doesn’t come about automatically,” states Leegwater. “That is why we involve agents in the implementation from the start and actively ask where they believe AI can provide the most support. As a result, employees feel like co-owners of the change and resistance decreases. Practical examples from organizations that are already successfully working with AI help in this regard. Furthermore, during training sessions, we demonstrate how AI can take over repetitive tasks, giving agents more time for complex customer inquiries and personal customer contact.”
New tasks for supervisors
According to Leegwater, not only agents but also supervisors are taking on a different role due to the deployment of AI and must also be involved in AI implementations. “They will have to adapt to a more evaluative role – for instance, actively monitoring AI performance is crucial. Reports and dashboards will become an important steering tool for supervisors to identify where improvements are possible or necessary. With advanced AI applications, such as integrations with CRM systems, the technical competencies required by this group are sometimes underestimated.”
Review the billing models carefully in advance.
Joosten emphasizes that in AI implementations, most attention is often focused on functionalities and technology. “But it is at least as important to have insight into long-term costs beforehand. Different AI solutions use different pricing models, meaning the total investment can depend on usage.”
“We notice that organizations initially focus primarily on implementation costs, whereas usage can be the determining factor for the final business case. With a solution like Zendesk, many AI functionalities are included, while others are billed as add-ons or based on usage. Therefore, we map out not only the implementation in advance, but also the expected costs and revenues. This way, organizations know where they stand financially and what business case is associated with it.”
AI often leaves you wanting more.
“For many organizations, a first AI implementation is primarily an introduction to what AI can mean in practice,” says Joosten. “As soon as employees experience the first benefits, ideas for next steps often emerge naturally. We regularly see organizations already thinking about further applications during an implementation, such as integrations with CRM or back-office systems, or extending AI to other processes. In practice, AI rarely turns out to be a one-off project, but rather the starting point of a broader development.”
Jasper and Carmen's tips:
- Formulate a vision on AI for customer contact
- Invest sufficient time in the preparation of an AI implementation
- After implementation, set aside time for ongoing optimization efforts.
- The introduction of AI tools changes the work of agents and supervisors; pay attention to this.
(Ziptone/Erik Bouwer)
Featured, Knowledge partners, Technology


