An automated response handling bot is a system that receives incoming messages from customers, identifies the intent or request type, and returns a response or performs an action without manual intervention. It operates across channels like WhatsApp, website forms, or email, and connects to your internal systems - CRM, Sheets, or order management.
We build these bots with n8n, the WhatsApp Cloud API, and integrations to CRM or Google Sheets. The bot receives the message via webhook, passes it through logic that decides what to do, and returns a response or updates a record in the system. All of this happens within seconds, even when the business is closed.
What does an automated response handling bot actually do?
An automated response handling bot receives a message, identifies its content or source, and performs an action according to predefined rules. The action can be sending a text response, updating a CRM field, creating a task for a representative, or routing the inquiry to a specific queue.
For example: a customer sends a WhatsApp message with the word “price”. The bot identifies the word, checks if the customer exists in the CRM, and sends them the updated price list as a PDF file. If the customer doesn’t exist, the bot creates a new record in the CRM with the date, inquiry source, and content, and sends an alert to a representative.
Another bot we built handles responses to website forms. When someone fills out a form, the bot sends an automatic confirmation email, updates the Sheet that centralizes inquiries, and sends a WhatsApp alert to the manager with the inquiry details. All of this happens within 10 seconds of submission.
The bot doesn’t “understand” the message like a human. It works according to exact matches (keywords, fixed expressions) or according to a language model that classifies the intent. If the rules don’t cover a particular case, the bot routes the inquiry to a representative or sends a default response.
How do you build this bot technically?
An automated response handling bot is built around a webhook that receives incoming messages, an n8n workflow that processes them, and connections to systems where information is stored or needs to be updated. The process starts with setting up the webhook in the WhatsApp Cloud API or the platform through which messages arrive.
In n8n, the workflow receives the message, breaks it down into fields (sender, content, date), and passes it through nodes that check conditions. For example: if the content contains “order”, the bot searches for the customer in the CRM by phone number. If the customer exists, the bot sends them a payment link. If not, it sends a message requesting additional details.
The connection to the CRM is made via API or via Sheets if the system doesn’t support direct API. Every update is recorded with a timestamp, so you can track every interaction. If the bot fails to update the CRM (for example, due to a network error), the workflow sends an alert to the team and saves the message in an error queue.
We use OpenAI or Anthropic models when we need to classify intent that doesn’t fit simple rules. The model receives the message content, returns a category (for example: “information request”, “complaint”, “order”), and the workflow continues according to the category. This adds a small cost per message, but allows handling a wider variety of inquiries.
| Component | Role | Example Tool |
|---|---|---|
| Webhook | Receives incoming messages | WhatsApp Cloud API, Typeform |
| Workflow | Processes and decides what to do | n8n, Make |
| Data store | Stores information about customers and inquiries | CRM, Google Sheets |
| Language model (optional) | Classifies intent | OpenAI GPT-4, Claude |
How long does it take to set up a response handling bot?
An automated response handling bot that handles a basic scenario - receiving a message, checking the CRM, and sending a response - goes live within one to two weeks. This includes setting up the webhook, building the workflow in n8n, connecting to the CRM or Sheets, and testing with real messages.
The time depends mainly on the complexity of the logic and the number of systems that need to be connected. A bot that handles three types of inquiries and updates one Sheet is ready faster than a bot that classifies ten types of inquiries, searches multiple data sources, and sends personalized messages.
Part of the time goes into defining the responses themselves. You need to write the texts, decide when to send a file, when to transfer to a representative, and what happens if the bot doesn’t recognize the intent. We work with the client on this document before we start building, so the workflow reflects the process the business wants.
After setup, the bot goes through a monitoring period of two weeks to a month. During this time we check which messages it doesn’t handle properly, and add rules or improve the classification. Most bots reach stability after the first month.
What happens when the bot doesn’t know what to answer?
When an automated response handling bot receives a message it doesn’t know how to handle, it performs one of two actions: sends a default response requesting clarification, or routes the inquiry directly to a representative. The choice depends on the nature of the business and the availability of the team.
The predefined response might be: “Thank you for your inquiry. I couldn’t understand the request, but a representative will get back to you soon”. At the same time, the bot saves the message in a Sheet or CRM with a “not classified” flag, and sends an alert to the team. This way no inquiry falls through the cracks.
We build a simple control panel in Google Sheets that displays all messages the bot didn’t handle. The team sees the content, the sender, and the date, and can decide whether to add a new rule to the workflow or if it’s a one-time inquiry that requires manual handling.
If the bot uses a language model for classification, it can return a confidence level. For example: “I think this is an information request, but I’m not sure”. In such a case, the bot can send a general response and transfer the inquiry to a representative in parallel, instead of risking an incorrect answer.
When is a response handling bot not suitable?
An automated response handling bot is not suitable when most inquiries require human judgment, negotiation, or deep understanding of context. If every customer comes with a unique situation that requires customization, the bot will route most inquiries to a representative, and its value will be limited.
Businesses that sell complex services - legal consulting, financial planning, medical treatment - usually don’t benefit from a full bot. They do benefit from a bot that performs initial filtering: collects basic details, asks screening questions, and transfers the inquiry to the appropriate representative with all the information. This isn’t a full response handling bot, but it saves time.
A bot is also not suitable when the business still doesn’t know what the right process is. If there are no standard responses, there’s nothing to automate. In such a case, it’s better to document inquiries manually for a month, identify patterns, and then build a bot that handles the recurring patterns.
Finally, a bot is not suitable when the business isn’t ready to monitor it. A bot that doesn’t work correctly causes damage - customers receive incorrect responses, inquiries fall through, and trust is damaged. If there’s no one checking the logs at least once a week, it’s better to wait with the setup.
How do you measure if the bot is working?
An automated response handling bot is measured by the percentage of inquiries it handles without human intervention, average response time, and number of errors. We build a simple report in Sheets that updates automatically and displays these numbers.
The central metric is the automated handling percentage: how many inquiries the bot closed on its own, versus how many were transferred to a representative. A good bot handles 60% to 80% of recurring inquiries. If the percentage is below 50%, you need to check if the rules cover the common types, or if customers are asking questions we didn’t anticipate.
Response time is a secondary metric. A bot that works correctly returns a response within 5 to 10 seconds. If the time increases, it means there’s a problem with the API of one of the systems, or the workflow includes a step that takes time (for example, searching a large database). We receive an alert if response time exceeds 30 seconds.
Errors are counted separately: how many times the bot failed to send a message, update the CRM, or read data. Every error is recorded with the date, type, and inquiry that caused it. If there are more than five errors per day, we check what changed - sometimes it’s an update in the external system’s API.
How does the bot connect to existing business systems?
An automated response handling bot connects to existing systems via API, webhook, or direct access to Google Sheets. The choice depends on what the system supports and how frequently data needs to be updated or read.
If the CRM supports API - like most modern systems - we connect n8n directly. The bot sends a request to search for a customer by phone or email, receives the data, and uses it to decide what to answer. If a field needs to be updated, the bot sends an update request with the new value.
When the system doesn’t support API, we use Sheets as an intermediate layer. The bot writes the inquiry to the Sheet, and the team works from the Sheet or imports it to the internal system. It’s not ideal, but it works, and it’s better than manual documentation.
Accounting systems like Morning (Green Invoice) are connected via API when the bot needs to issue an invoice automatically or check payment status. For example: a customer sends “I received the product”, the bot checks if there’s an open invoice in their name, and if so, sends them a payment link.
Every connection is tested separately before the bot goes live. We send dummy requests, check that data returns correctly, and ensure the bot handles errors - for example, what happens if the API doesn’t respond within 10 seconds.
How do you handle messages in different languages?
An automated response handling bot can identify the message language and return a response in the same language, if we’ve defined responses in each language. Identification is done through a simple library that identifies Hebrew, English, Arabic, or Russian by characters, or through a language model that returns both the language and the intent.
If the business serves customers in Hebrew and English, we prepare two versions of each response. The bot checks the language, selects the appropriate version, and sends it. If the language isn’t identified, the bot sends the response in the default language - usually Hebrew for businesses in Israel.
When using a language model for classification, the model receives an instruction to identify the language as well. For example: “Identify the message intent and its language. Return JSON with two fields: intent and language”. The workflow uses the language field to select the correct response.
Businesses that receive messages in many languages can use automatic translation as an additional layer. The bot translates the message to Hebrew, classifies it, and translates the response back to the source language. This adds a small cost and delay, but allows handling any language without preparing responses in advance.
Chatbot systems we build include multilingual support by default when the business requests it. Documentation and testing are done in all languages the bot is supposed to support.
For more on how to evaluate automation platforms for building these systems, see our n8n platform guide.
Frequently asked questions
Can a response handling bot work on both email and WhatsApp?
Yes. The bot receives messages from both channels through separate webhooks, and processes both in the same workflow. You can define different responses for each channel, or use the same logic for both. The bot can also update the CRM with the channel through which the customer reached out, so the team knows where to continue the conversation.
What happens if the CRM API is down when the bot tries to update?
The bot retries twice more, and if that still doesn't work, it saves the inquiry in an error queue and sends an alert to the team. The message to the customer is sent even if the CRM update fails, so the customer doesn't feel that something didn't work. The team handles the error manually and updates the CRM when it's back up.
Can you change the bot's responses without contacting whoever built it?
It depends on how the bot was built. If the responses are stored in a separate Sheet, the team can edit them directly, and the bot will read the updated version on the next inquiry. If the responses are hard-coded in the workflow, you need to contact whoever built it. We usually build with a Sheet when we know the responses will change frequently.
How much does it cost to run such a bot in terms of monthly expenses?
Costs depend on the number of messages and the systems the bot is connected to. WhatsApp Cloud API charges per conversation, n8n costs according to plan (there's also a self-hosted version with no licensing cost), and if you use a language model, there's a small cost per message. For a business receiving hundreds of messages per month, the monthly cost is typically between a few hundred to a thousand shekels, depending on scale.