Customer service is often summarized in one sentence: "that's all handled by chatbots anyway." In practice, the department consists of dozens of tasks that have little to do with each other. An incoming call about a deviating delivery requires something different than entering an order, and drafting a quote requires something different again than handling a complaint. Our taxonomy counts 82 tasks in this domain. Some of these AI can already largely take over today, others require an employee who approves or rejects, and part remains human work. The pattern behind that division is more useful than a list of 82 separate lines.
The tasks that are most fully automatable share two characteristics: they are high-frequency and the answer or next step can largely be derived from existing data. Checking and communicating order status is a good example of this: the system already knows where an order stands, and the task essentially consists of looking up that information and translating it for a customer. Conducting live chat and chatbot conversations also largely falls into this category, especially for standard questions about opening hours, return policy or common product questions. Entering orders into the system is a third example: if the entry follows from a clear customer request with fixed fields, little interpretation is needed and there is little risk in an incorrect choice.
What these tasks have in common is that a mistake is usually quickly visible and correctable, and that the decision space is small. There is a clearly correct answer or a clear next step, and that can be automated without a human having to constantly monitor it.
A large part of the work in customer service sits in an intermediate position: AI can do most of the work, but an employee must approve or reject the result before it reaches the customer or enters the system. Drafting quotes is a telling example here. AI can compile a price proposal based on the customer request, including products, services and terms, but the final margin, exceptions and customer-specific agreements usually require a look from someone who knows the customer relationship. Processing order changes falls into the same category: implementing an address change is trivial, but a change that affects delivery time, price or stock across multiple systems requires a check.
Registering complaints also belongs here, though on the cautious side: recording the facts is something AI can do well, but assessing whether something is an isolated incident or a signal of a structural problem is something an employee must judge before the registration takes on further life within the organization. In all these cases, the gain is not that the task disappears, but that most of the work shifts from executing to assessing.
Handling the substance of complaints is the clearest example of work that remains human work. Here it is not about following a procedure, but about weighing an individual situation, gauging what a customer needs to feel heard, and making a decision that does not follow from a script. Following up on outstanding quotes also partly belongs here: sending a reminder can be automated, but the conversation that follows when a customer hesitates or negotiates requires someone who can adapt. Complex escalations on social media, where a dissatisfied customer responds publicly and the organization's tone is at stake, also cannot be left to a model without creating reputational risk.
The common denominator in this category is that the task revolves around judgment, relationship or unpredictability, and that a wrong assessment cannot simply be corrected afterwards.
If you overlay these three categories onto the 82 tasks, no sharp division emerges between "front office stays human" and "back office becomes AI." The division runs across both: some front-office tasks such as live chat are readily automatable, while some back-office tasks such as assessing an unusual order change require oversight. The question is per task: how predictable is the outcome, and how great is the damage in case of a mistake?
This pattern is not unique to customer service. Also with what can AI take over in wholesale and with what work can AI take over in purchasing, the division between the three categories lies less in the job title and more in the nature of the individual task. Those looking for structure would therefore do well to look at each task individually rather than generalize by department or role.
One final note: if these findings are used in a conversation about staff numbers or positions, separate legal requirements apply to that, which this page neither adds to nor detracts from. This overview describes tasks, not personnel decisions.
To see how this division plays out for your own customer service, it is of little use to guess based on job titles. The ratio between the three categories depends on the mix of tasks your employees actually perform, the systems involved, and the complexity of your customer questions. Ftetoai offers a free quickscan for this: twelve questions, no account required, resulting in an indication of what portion of the hours in your profile can be taken over by AI today. The full work scan, which goes deeper into individual tasks, is still under construction; we deliberately do not offer that here before it is ready.
Vraag maar. Ik ken de kennisbank van deze site; wat ik niet weet, zeg ik erbij.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.