“Where’s my order?” Twenty times a day. Your team can do better, if the machine does the boring part.
Customer service is the department where AI delivers something fastest, and breaks something fastest. The questions are often the same, the answers are somewhere, and yet a customer waits for hours. At the same time, almost nobody wants to talk to a bot. The solution isn’t a chatbot at the front door, but an assistant next to your employee: one that does the groundwork, prepares the answer and lets the person decide.
// A report from a fictional online shop with a physical shop, eighteen employees, two people on service, a shared mailbox, WhatsApp on the owner’s phone and a telephone nobody answers during lunch. The company is fictional, the situations are real: we run into them again and again.
Sound familiar?
- “Where’s my order?” makes up a quarter to half of all questions.
- Customer emails wait for hours, sometimes a day; the weekend doesn’t count.
- The phone rings during lunch or a customer visit and nobody picks up.
- Opening hours, returns policy, invoice question: typed out afresh every day.
- A complaint sits in a personal mailbox and surfaces after the holiday.
- WhatsApp with customers runs through the owner’s personal phone.
- Twenty minutes on the phone and nothing recorded; the next colleague starts over.
- An “invoice attached” email to the wrong address: the most common data breach.
What usually runs here
The software we come across here most often. It does not have to go; we build around it and in between.
A day in customer service & communication: now and next
On the left, the day as it usually runs now. On the right, the same day once the dull parts are automated and AI does the groundwork. Watch what changes: not the people, but where their time goes.
- 08:30ServiceShared mailbox: 47 new emails. Twelve times “where’s my order”. Each reply: look up the order, copy the track-and-trace, type.
- 10:00ServicePhone: customer wants to return something. Colleague looks up the returns policy in an old Word document. Calls back at 11:15.
- 12:30ShopLunch. The phone rings four times. Voicemail full.
- 14:00OwnerGets a complaint about a damaged delivery on his personal WhatsApp. Forwards it to service. Service sees it tomorrow.
- 15:30ServiceAngry customer on the line, twenty minutes. Resolved, not recorded. The customer calls again next week and starts over.
- 16:30ServiceOne-star review on Google: “never got a reply”. Nobody responds.
- 17:45Service19 emails still open. Same again tomorrow.
- 08:30ServiceThe twelve status questions were answered overnight with the real delivery status from the system, with a notice that an AI replied and a button for “I’d rather have a person”. Of the remaining 35 emails, 28 have a draft reply ready, sorted by urgency.
- 10:00ServiceReturns question: the draft reply pulls the returns policy from the knowledge base, with a source reference. Colleague reads, clicks approve. Done at 10:02.
- 12:30ShopLunch. The phone is answered by a voice assistant that identifies itself as such, notes the question and offers a call-back slot. Two customers choose “call me at 13:30”.
- 14:00OwnerWhatsApp runs through the business number in the shared inbox. The complaint is recognised as a complaint, sits at the top, and service calls within the hour.
- 15:30ServiceThe twenty-minute call was summarised with consent and linked to the customer. Next week, every colleague immediately sees what was agreed.
- 16:30ServiceThe one-star review got a draft reply within an hour; the colleague adjusted the tone and posted it. And called the customer.
- 17:45ServiceZero emails older than four hours. The day ended at 17:00.
Where AI makes the difference here
Per use case: what the automation does, what a person keeps doing, and whether it is proven or still promising. We say which one honestly.
Status questions answered automatically
Questions about an order, delivery or appointment are recognised and answered with real data from your systems, with an AI notice and a button to reach a person.
Still human: picks up everything that isn’t standard.
Draft replies from your own knowledge base
For every email or chat, a reply is ready, built from your policy and documents, with a source reference. No made-up promises.
Still human: reads, adjusts and sends. That’s a line we don’t move.
One inbox, all channels
Email, WhatsApp, chat and phone notes in one shared environment, automatically classified and prioritised; complaints and angry customers at the top.
Still human: decides the order and the tone.
A phone that always gets answered
A voice assistant that identifies itself as AI, notes the question, answers simple questions and schedules a call-back. In Dutch the quality still varies; we test it at your business first.
Still human: calls back and has the real conversation.
Conversations recorded, knowledge kept
With consent, conversations are summarised and linked to the customer; recurring questions become proposals for the knowledge base.
Still human: manages the knowledge base; what’s in there is the truth.
Preventing data breaches through wrong recipients
A check before sending: wrong address, attachment with personal data, unusual payment requests.
Still human: gets a warning and decides.
Honestly: what AI does not do here
- A chatbot at the front door is the fastest way to lose customers: in the Netherlands, only 12% of customer questions are answered well by chatbots. We put AI next to the person, not in front of them.
- AI answers can be wrong, even with a knowledge base. Air Canada was held liable for a promise made by its chatbot. So: source references, a defined set of topics and a person who sends.
- Since August 2026, customers must know they’re talking to AI. We arrange that properly, without disrupting the conversation unnecessarily.
- Klarna replaced two thirds of its chats with AI and later hired people again because quality dropped. Fewer people isn’t the goal; better service with the same people is.
The objections we hear — and whether we can close them off
We looked them up and heard them from clients. For each objection we say whether it is really solvable, partly solvable, or a risk that stays and that you accept knowingly.
Can it be closed off: yes. Rightly so: almost two thirds of customers would rather not have AI in service and 87% demand a human option. That’s why, with us, AI does the groundwork and only answers factual status questions itself, always with a button to a person. The person remains the face.
- 01Week 1: core session on the service desk. We go through a week of emails and calls and count which questions keep coming back.
- 02Weeks 1 and 2: one shared inbox for email and WhatsApp, with automatic classification and priority.
- 03Weeks 2 and 3: build the knowledge base and draft replies with source references; connect status questions to your systems.
- 04Weeks 3 and 4: pre-send check against wrong recipients; review monitoring with draft replies.
- 05Week 4: measure. Response time, questions resolved per hour, customer satisfaction. Only then the voice assistant.
First the free AI scan, then a core session on your floor (a half-day of 4 hours, €596 excl. VAT and travel costs) with a core report and an honest go/no-go. We build on a project basis with fixed hours: €110 per hour for straightforward work, €165 for complex development.
More hours only with your written approval. If we cannot deliver, you pay nothing for what was not delivered.
Want to know what this means for your customer service & communication?
Take the free AI scan or book an intro call with someone who builds it themselves. A reply within one working day, no slides, no obligations.
Most businesses start with the tool and then get stuck. It is better to start with one piece of work that comes back every week and nobody enjoys.
The build costs are usually the easiest part of the bill. The surprises are in the clean-up beforehand, the monthly usage and the maintenance afterwards.
- NBER, Generative AI at Work (working paper 31161) (2023)
- Gartner, 87% of customers say companies using GenAI must provide access to a human agent (2026)
- Security.nl, Dutch DPA data breach report 2024 (2025)
- Frankwatching, Nationale Voice Monitor 2026 (2026)
- Customer Experience Dive, Klarna says its AI agent is doing the work of 853 employees (2025)
- CBC, Air Canada found liable for chatbot’s bad advice (2024)
- Legalz (summary of the ACM/AP recommendations on AI chatbots) (2025)
- ACM ConsuWijzer, Rules for recording phone calls (2025)
- Gartner, 64% of customers would prefer that companies didn’t use AI for customer service (2024)