Chatbot workflow automation: what happens after the chat

The conversation is over — then what?

The chatbot answered the customer, took a phone number, accepted a return request. The conversation ended. And then? In many companies «then» is manual work: an agent reads the conversation, copies the details into the CRM, messages a salesperson, sets a reminder. The bot works fast, but a person carries its result into the system — and that step is always late, and sometimes forgotten.

A chatbot workflow is the sequence of things that happen automatically after a conversation. This article covers the four-part workflow model — trigger, condition, action, approval — six real business examples, where human approval is needed, error handling and practical steps to start. The technical integration architecture is a separate topic; here the subject is designing the business process.

The four parts of a workflow

1. 1TriggerThe event that starts the workflow: a phone number was recognised, the customer wrote «I want to return this», the conversation was handed to an agent, a set time passed.
2. 2ConditionThe check that decides whether to act: the lead's category, the channel, working hours, the amount, the customer's previous status.
3. 3ActionThe work done in a system: create a lead in the CRM, notify a salesperson, add a tag, set a reminder, query a company system.
4. 4ApprovalDoes a person need to look before or after the action? Wherever money, contracts or exceptions are involved — yes.

Steps that work out of the box in Vexvon

Part of a workflow is ready in the platform, and part depends on integration with the company's own systems. What works in Vexvon out of the box:

• When a phone number is recognised in a conversation, a lead is created and the customer card is tagged «left a number»
• A trigger on a new lead: a notification to the sales group in Telegram, a lead to Bitrix24, or the internal CRM
• Assigning the lead to a salesperson in four modes: manual, rotation, by channel, by workload
• Reminders for the salesperson — the system checks them every minute
• Queries to the company's own APIs during the conversation — for example, reading an order status

Actions outside this list — changing an order in a warehouse system, say — are built with the company's own API and the corresponding connection. Which systems can be connected is shown on the integrations page.

Six business examples

1. 11. An ad leadTrigger: a number recognised in an Instagram DM. Condition: category «sales». Action: a lead is created, the sales group notified, the lead assigned in rotation. Approval: none.
2. 22. A lead at nightTrigger: a lead outside working hours. Condition: high priority. Action: the customer is told «we'll contact you at 09:00», and the salesperson gets a morning reminder.
3. 33. A return requestTrigger: the customer wants to return an item. Condition: the order was found. Action: the reason and a photo are collected and the request passed to an agent ready to go. Approval: the agent.
4. 44. An unhappy customerTrigger: a complaint or the word «agent». Action: a notification, the bot stops, a note on the customer card. Approval: the manager on duty.
5. 55. Order statusTrigger: a status question. Condition: identity checked. Action: the status is read from the order system and told to the customer in plain language.
6. 66. Follow-upTrigger: the customer asked a price, gave no number and left. Action: if there is consent, one gentle reminder after a set time; if no reply, it stops.

Where human approval is needed

The appeal of automation is doing everything without people. But some actions must not happen without approval:

• Money: refunds, discounts, compensation, credit
• Contracts: changing terms, cancellation, renewal
• Exceptions: decisions outside the rules
• Irreversible actions: cancelling an order, deleting data
• Sensitive data: health, finance, personal documents

Here the workflow does not perform the action itself — it prepares the request, collects the details and sends it to someone with authority. After approval the action can continue automatically.

Error handling

A workflow's most dangerous error is a silent one: the lead never reached the CRM and nobody knows. Four questions for every workflow:

• What happens if the action fails — a retry, or a notification to a person?
• Who hears about the error, and where?
• What is the customer told — especially if a promise was made?
• Where are errors logged, and when are they reviewed?

How failed requests are handled on the technical side is covered in chatbot API and webhook integration.

The workflow document

Each workflow is described in a one-page document:

• Name and purpose — one sentence
• Trigger, condition, action, approval
• What the customer is told
• What happens on error
• Owner — who is responsible for the workflow working
• Success criterion — what is measured

A filled-in example: an ad lead

1. 1Name and purpose«Instagram lead» — a customer who leaves a number in DM reaches a salesperson within 10 minutes.
2. 2Trigger and conditionA number was recognised; the conversation is in the sales category; the customer is not already an active lead.
3. 3ActionA lead is created, a Telegram notification goes to the sales group, the lead is assigned in rotation, a 10-minute reminder is set.
4. 4What the customer is told«A salesperson will write to you within 10 minutes.»
5. 5On errorIf the notification fails or nobody takes the lead in 10 minutes — the head of sales is notified.
6. 6Owner and criterionHead of sales; the share of leads taken within 10 minutes.

Common mistakes

• Building a workflow «in the settings» with no document — after a few months nobody knows what it does
• Not thinking through the error case — a silent failure
• Automating money and contract actions without approval
• Tying several workflows to the same trigger — the customer gets two messages
• Not naming an owner for the workflow

Practical steps to start

1. 1List the manual workFor one week, write down every manual task agents and salespeople do after a conversation.
2. 2Pick the most repeatedUsually writing the lead into the CRM and notifying a salesperson.
3. 3One workflow, one channelLaunch the first workflow on one channel and watch it for two weeks.
4. 4Read the errorsRead every error and every half-finished action, and tighten the conditions.
5. 5ExpandMove to a second workflow only once the first runs reliably.

What to measure

• How often the workflow fires, and its success rate
• Silent errors — an action should have happened and did not
• Time from the end of the conversation to the action
• The fall in time spent on manual work
• How long requests wait for approval

Limits

A workflow can act only in connected systems. If a company system offers no API, that step stays manual — the bot only prepares the request and sends it to a person. A very complex workflow — dozens of conditions, interlinked steps — is hard to test and maintain; simple, single-purpose workflows are more reliable.

Conclusion: the conversation is not the end of the work

A chatbot workflow carries the result of a conversation into your systems without manual work: the lead to the CRM, the notification to the salesperson, the request to the agent, the reminder to the calendar. When trigger, condition, action and approval are written clearly, error cases are thought through, and every workflow has an owner, automation becomes dependable.

How leads are kept in the CRM is on the CRM page; the details of the status workflow are in the order status chatbot. The rest of the section is in this category. To write your workflows together, ask for a demo.

Frequently asked questions

1. 1Does a chatbot workflow need a developer?The built-in steps — lead, notification, assignment — are configured in settings. Acting in the company's own system needs an API and a technical connection.
2. 2Which workflow should we start with?The most repeated manual task — usually writing the lead into the CRM and notifying a salesperson.
3. 3Can a workflow make decisions about money?No. Where money, contracts or exceptions are involved, the workflow prepares the request and a person decides.