AI in customer service: availability, speed and hours that come back

More and more organisations look at AI for customer service: a chatbot on the website, an agent that handles mail or a help for your service desk. The question remains: what does it deliver? In this article you see three things you can count: availability, speed and hours that come back. After that you work through an example in three steps and we compare a chatbot with an AI agent. The numbers in the example are a calculation example. Your own figures decide what it delivers in your customer service.
What does AI deliver in customer service?
Three things you can count:
- Availability: customers get an answer outside office hours too. You measure the share of questions that get an answer directly.
- Speed: an answer is done sooner. You measure the lead time from receipt to done.
- Hours that come back: the hours an agent takes over go back to your team. You use those hours for customers with a question that needs an employee, and for work that is left undone now.
Quality can be measured too: the share of answers that an employee approves without changes. For every agent we set one main goal with a number up front, and what an agent really does we count from the AgentWorks log and not from an estimate.
Step 1: Record your current volume and time
Collect the figures of the past three months:
- Number of tickets or contacts per month
- Average handling time per ticket in minutes
- The share that keeps coming back and is predictable
- The share that arrives outside office hours
Calculation example: 1200 tickets × 8 minutes / 60 = 160 hours per month.
Step 2: Decide which part an agent takes over
Not every ticket suits an agent. Divide your tickets by kind:
| Kind of ticket | Suits an agent? |
|---|---|
| Frequently asked questions and product information | Well |
| Status and order updates (with a connection to your systems) | Well |
| Returns and complaints | The agent does the groundwork, an employee decides |
| Questions about a purchase or quote | Partly |
| Technical or legal questions | An employee, with the agent as preparer |
How many of your tickets an agent really takes over you only know after go-live. You measure that in the log. In this calculation example we assume the agent takes over 40%. That is an assumption for the example and not a promise.
Calculation example: 1200 × 40% × 8 / 60 = 64 hours per month. That is 768 hours per year. At € 46 an hour (source: CBS (Statistics Netherlands), labour costs per hour worked, 2024) that is € 35,328 of capacity per year, with which you help more customers without hiring more people for the recurring work. Expressed in digital FTE that is 0.46 (140 hours per month is one digital FTE).
Step 3: Put it next to your price
An AI agent costs a fixed amount per month. Build, hosting, maintenance and use of the AI model up to a fixed credit are included and you pay from go-live and nothing up front. What your agent costs is set out up front in your proposal. Put the capacity from step 2 next to the monthly price × 12. That is the value of hours and not a saving on wages: nobody leaves your payroll, your organisation gets room.
Count the other two as well. What is it worth to you that customers get an answer at ten in the evening? And that an answer is done the same day instead of the next? You cannot always put that value in euros, but you can measure it, and you do that with the goal you agree up front.
Chatbot or AI agent: what suits your questions?
The term AI for customer service covers two different things.
Chatbot
- Answers fixed questions from a knowledge base or decision tree
- No connection to your back office
- Biggest risk: it does answer, but does not solve the problem, so the customer calls anyway
AI agent
- Consults live data, such as order status, stock and customer history
- Carries out actions: creating a return, booking an appointment or preparing an invoice
- Biggest risk: a wrong answer if the agent is not set up well
Rule of thumb: does your customer need an action as well as an answer? Then choose an AI agent. Is it about information? Then a chatbot can already catch a lot. With an agent you decide up front what it may do by itself and what an employee decides.
What stays with you?
Even with a fixed monthly price some work stays with your team:
- Keeping the knowledge up to date: new products, changed terms, questions that return per season. If your knowledge is scattered, a Company Brain (from € 795 per month) puts it in one place.
- Following along in the first weeks: an employee looks at the answers and gives back what can be better.
- A good handover: if the agent cannot work it out, the conversation goes to an employee with all the context.
If the agent processes personal data, a data processing agreement is part of it. Read more under security and management.
Measuring from day one
Without measuring you do not find out whether it works. In the AgentWorks log you see every week how much the agent does, how fast and how much goes to an employee. That way you adjust, and after a few months you know what has come back.
Get started
Want to know what this delivers for your customer service? Run an AI scan for a first opportunity map or book an advice session. Within two working days you have a proposal with the price. More about what an agent does for customer service is under AI agents.
Frequently asked questions
Short, clear answers so you can decide faster.
When do the hours come back?
From the day the agent goes live. An agent goes live in two to eight weeks, depending on the agent, and you pay from that day. How many hours come back, you see in the AgentWorks log.
What share of tickets does an agent take over?
That differs per kind of question and per organisation. In the calculation example above it is an assumption. You measure your own share in the log after go-live. Plan with a low and a high scenario.
Do I have to replace my helpdesk?
In most cases not. In the advice session we discuss which systems the agent has to work with and how the connection runs.
What if the agent gives a wrong answer?
Decide up front what the agent may do by itself and pass the rest on to an employee with the context. Check the answers in the first weeks and use the log to adjust.



