You open your phone in the morning: eighteen unread messages on WhatsApp, six on Instagram, a few more by email. Half of them ask the same three questions, two are angry, one really matters and will not get its turn until evening. This is not a talent problem; it is a volume and timing problem. This guide explains how a digital worker takes over your inbox, where it stops, and how it is set up in the first week.
Why can't you keep up?
Four reasons, almost the same in every business:
- The channels are scattered. The same customer writes on WhatsApp and continues on Instagram; you open each app separately.
- Most questions are the same. "Where is my order", "what does it cost", "are you open today", "is there an appointment". The answers are known, but you type each one again.
- Messages do not keep office hours. What arrives in the evening or at the weekend waits until the next day, and the customer writes somewhere else in the meantime.
- The few that need judgment get lost in the many. The three messages you genuinely need to look at wait in line behind twenty routine ones.
Framing the problem correctly matters, because the solution is not "type faster". The solution is for the routine majority to leave your hands, and for the minority that needs judgment to arrive in front of you, flagged.
What is the difference between a canned auto-reply and a digital worker?
Messaging apps have their own auto-replies, and there are ready-made bots; most businesses try those first. Knowing the difference lets you set up the right thing.
| What it does | Where it stops | |
|---|---|---|
| Auto-reply | Sends the same canned text to every message ("We will get back to you shortly"). | Answers no question; it only extends the wait politely. |
| Ready-made bot | Makes the customer pick from a menu, or gives fixed answers to fixed questions. | Stalls at the first question outside the menu; cannot look at your order record. |
| Digital worker | Reads the message, looks in your own systems, writes a reply specific to that customer, leaves anything sensitive as a draft until you approve. | Stops on matters that need judgment and flags them to you; that is a design, not a limit. |
A concrete example: "Where is last week's order?". The auto-reply says "we will get back to you shortly". The ready-made bot says "type your order number", and once the number is typed it usually lands on you anyway. The digital worker opens the order record, pulls the carrier status, reads the reason if there is a delay and answers "Your order was handed to the carrier yesterday and shows delivery tomorrow"; if the parcel is lost, it does not send a reply but flags the customer to you.
What does a digital worker actually do in the inbox?
An inbox worker's job sorts into three buckets, and every message lands in one of them:
- Answers itself. Anything that can be answered from recorded information: order status, opening hours, price list, address, return conditions, appointment availability. It takes the information from your system, not from memory.
- Leaves a draft. It prepares the reply but does not send it; you approve with one tap. In the first weeks most messages sit in this bucket; as trust grows they move to the first.
- Flags to you. Complaints, negotiation, personal-data requests, health or legal topics, anything that needs judgment. The worker writes no reply; it summarises the message and puts it in front of you.
Beyond that, scheduled work can belong to the same worker: listing the conversations left unanswered for you in the evening, sending appointment reminders at a set hour, prioritising the weekend's messages on Monday morning.
You decide what falls into which bucket, and you can change it later. A good start is to keep everything in the second bucket for the first week.
How does it work on your own number?
This is the most misunderstood point: a digital worker does not give you a new number or a new inbox. It connects to your own WhatsApp Business number, your own Instagram account and your own email domain. Your customer sees no change; they write to the same number and get their reply in the same chat.
The same goes for records. Order information stays in your own e-commerce or accounting system, appointments in your own calendar, customer records in your own CRM. The worker connects to those systems through skills (integrations) and builds the context between them; it does not move the data, it reads it where it lives. On Botonom a worker is hired exactly this way: the channels are yours, the systems are yours, the worker stands in the middle of all of them.
The practical consequence of this design: on the day you switch the worker off, nothing is lost. The number is yours, the records are yours; you simply go back to reading the inbox yourself.
How is it set up in the first week?
Five steps, all of which fit in one afternoon:
- Pull out the twenty most frequent questions. Look at the last month of messages; list the questions that repeat and the answers you gave. That list is the worker's knowledge base.
- Write the red lines. Topics the worker will never answer: price negotiation, complaints, return decisions, personal data, anything else you decide. The "third bucket" rule applies to these.
- Connect the channels and the records. Your own WhatsApp number, Instagram account and email; your order and appointment systems.
- Start in draft mode. In the first week, no message leaves without your approval. Read every draft; correct the wrong ones and write the correction back as a red line or a knowledge-base note.
- Sort the buckets. At the end of the week you can see which questions went through without a single correction; move those to the first bucket (answers itself). The rest stays in draft.
This sequence is the inbox version of the general way a digital worker is hired; for choosing the first job and the seven-day plan, see the where to start guide.
Where should the worker stop?
The inbox is where the most expensive mistakes happen: a wrong reply has reached a customer and cannot be taken back. That is why the limits are set from the start.
Where the worker should stop: price and discount negotiation, complaint and return decisions, health or legal content, personal-data requests, and every message it does not understand. When it does not understand, it must not guess; it must flag. We described which actions should run freely and which should wait for approval as a matrix in approval gates.
Personal data is a separate heading. Customer messages contain personal data, and that flow is a decision that comes before the digital worker decision. Check it against the data protection regime you operate under (KVKK in Türkiye, the GDPR in the EU) and ask your own legal side which jobs fall in scope. This line is not legal advice.
What should you measure?
Three numbers, weekly:
- First-response time. How long a message waits for its first reply. It should fall from the first week; if it does not, the worker is answering too few messages itself.
- Share left to a person. Out of a hundred messages, how many landed in front of you. High at first, falling as corrections are written into the knowledge base. Zero is not the goal; messages that need judgment should always reach you.
- Recalled replies. The number of replies you had to correct after they were sent. It should stay near zero; if it rises, narrow the first bucket.
These three numbers tie how much the worker has really taken over to figures rather than feel. Decisions (widening permissions, changing the buckets, opening a second channel) are made on these numbers.
Frequently asked questions
Will my customer know they are talking to a bot?
The worker replies on your behalf, in your tone; but when asked whether it is an AI, saying so is both honest and the right design. On routine questions the customer notices a fast, correct answer; on matters that need judgment a person steps in anyway.
Does it work at night and at weekends?
Yes; it answers first-bucket questions regardless of the hour and lists the rest for you in the morning. An "are you open today" that arrives after hours no longer waiting until the next day is, in most businesses, the first difference people feel.
Does the same worker handle WhatsApp and Instagram?
Yes. Channels are the worker's skills; even when the same customer writes on two channels, they are handled with one inbox logic.
What if it gives a wrong answer?
In the first weeks no reply leaves without your approval, so a wrong answer is caught in draft. After permissions are opened you keep the weekly count of recalled replies, and if it rises you narrow the set of questions the worker answers itself.
How many messages before it makes sense?
There is no threshold; even a few messages a day are worth it if they are eating your evenings. The measure is not volume but how much of your time the routine messages steal.