A customer wants to know where their order is. An employee looks up the status in the ERP or logistics system and reports it back, by phone, email or chat. This is one of the most recurring tasks within customer service, and also one of the most predictable. That is precisely what makes it suitable for AI to largely carry out.
Three axes are decisive here: structuredness, volume and customer contact.
Structuredness scores high. An order status is not an open-ended question: there is an order number, a system with a current status, and a fixed set of possible answers (shipped, in transit, delayed, delivered). Nothing needs to be invented, only looked up and reported back.
The volume is generally high. At companies with many orders, this type of question makes up a large share of incoming contact volume. High volume combined with a fixed pattern is exactly where automation pays off: the same action, hundreds or thousands of times, without the content changing.
Customer contact scores low toward automation (2 out of 5), but that is less of an obstacle here than it seems. Customer contact only becomes a problem for AI when the conversation is sensitive or requires customization. With a status question, the contact is functional: the customer wants an answer, not a conversation. As long as the message is correct and arrives on time, the "who" is secondary.
Three axes keep the score mixed, and it is relevant not to gloss over that.
Discretion (2) and creativity (5, noted unfavorably because no creativity is needed here and that is exactly the point) show that little interpretation is needed — but with deviations, such as an order that has been showing "in transit" for three days without a scan event, an assessment is indeed needed: is this a system error, a carrier delay, or a lost order? That requires a human at that point, or at least an escalation rule.
Compliance (5) means there are few legal complications involved in reporting back a status. Cost of errors (3) sits in the middle: reporting an incorrect status to a customer is annoying, but rarely harmful. This does, however, vary strongly by sector. With medical devices or time-sensitive B2B deliveries, an incorrect status message can indeed have consequences, and then this axis shifts to a higher risk than assumed here.
What AI can handle here today is an agent: a system that independently looks up the status in the logistics system, translates this into an understandable text, and communicates this to the customer, without a human sitting in between every time. This is not a full takeover of the role, but it is of a large part of the repetitive actions within that role.
Three conditions determine whether this works in practice:
At companies without a real-time connection between the order system and logistics system — for example, where the status is tracked manually in an Excel file — structuredness is actually lower than assumed here, and this task is not yet ready for an agent. That directly relates to a similar task: providing stock and delivery time information leans on the same data sources and runs into the same problem when that connection is missing.
Companies with many custom orders or B2B customers with contractual delivery agreements see the discretion axis weigh more heavily: a delay can mean a breach of contract there, and human judgment is needed sooner than with consumer packages.
This task is also connected to what happens further along in the process. Incorrect customer data leads to incorrect status messages, so keeping customer data up to date in the CRM is in fact a precondition, not a separate task. On the other side of the process, generating shipping and return documents and the origin of the order itself, via entering orders into the system, also play a role: the cleaner those steps are set up, the more reliable the status information that comes out later.
This is not personnel advice and not an argument for shrinking a customer service team. Whether and how an organization makes personnel decisions falls under its own legal requirements and own judgment; this page only describes which part of the work lends itself to automation, not what the personnel consequence of that should be.
Whether this task is automatable to a large or small degree at your company depends on your systems, your order volume and the extent to which deviations occur. The free quickscan from FTE TO AI consists of twelve questions, without an account, and gives an indication of which part of the hours in this type of work profile can be taken over by AI today. The full work scan, which calculates this down to task level for your own organization, is still under construction.
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.