A chatbot reply is not enough when the customer is still waiting for a quote, a callback, a decision or an answer the bot cannot give. In those cases, record the request, assign it to someone and set a deadline for the next action. A question about opening hours can end with a factual answer. A request for a site visit is different: ‘Thanks, we’ve received your message’ confirms receipt, but it does not arrange the visit. Judge whether the conversation is finished by what the customer needs, not by whether the messages have stopped.
Decide which conversations need a handover
Sort conversations by what the customer needs next. Someone checking whether you serve their area may just need a factual answer. Someone asking for a quote for two locations by Wednesday needs a follow-up. An existing customer asking about a missed appointment needs help with an active job, not a new sales opportunity. The requests may arrive through the same chat window, but they belong with different people.
Review a handful of recent chats with the colleagues who handle requests. Look for cases where customers had to ask twice, the bot offered human help, or it was unclear what would happen next. Use those examples to set routing rules. Send tailored quote requests to sales, for instance, and questions about active jobs to the person coordinating the work. Give ambiguous requests a named reviewer. A shared inbox can collect messages, but someone still needs to check it and assign the work.
Tell the customer what to expect after the handover. Should they stay in the chat, watch for an email or call the team directly? If nobody monitors the queue in the evening, don’t imply that a colleague is about to join. And if a customer asks to speak to someone, give them a way to do that without making them answer unrelated questions first.
Record enough for someone to pick up the conversation
Design the record for the colleague who opens it tomorrow morning. They need to know who asked, what they want, which details are missing and what the bot has already told them. Capture the contact method the customer provided, the source of the request, a brief summary, the person responsible and the next action with a deadline. ‘Quote for two offices; second address missing; Maya to email by 11 am on Tuesday’ is more useful than a record labelled ‘New lead’ with an unedited transcript attached.
Keep customer statements separate from assumptions. ‘Ideally next month’ is a preference, not a confirmed start date. Likewise, the date the customer wants a quote is not necessarily the deadline for your first reply. Record both if they matter. If the bot says someone will call by Friday, put that commitment where the person expected to call will see it, not just in the chat history.
When evaluating a sales CRM for managing requests and follow-up, decide who needs access to the original conversation and when a short summary is enough. Avoid copying full transcripts by default, especially if they include personal details unrelated to the request. Agree on how to handle repeat messages, too. A matching email address can help you find the right contact, but it cannot tell you whether the person is updating an open quote request or asking about an unrelated job.
Worked example: a quote for two offices
Imagine a four-person commercial cleaning business. At 4:40 pm on Monday, an office manager asks in the website chat whether the team can clean two offices in the evening. She supplies a work email address and the first site’s address. A colleague still needs to confirm the second address, but she would like a preliminary quote by Wednesday. The bot acknowledges the message without confirming availability, a price or a start date.
Maya, the sales coordinator, takes responsibility for the request. She has until 11 am on Tuesday to send her first email. The team records that deadline separately from Wednesday, the date the customer would like the quote. Maya asks for the second address and the preferred cleaning days. After sending the email, she marks the request as waiting for information and sets a reminder for Wednesday morning. If the details are still missing, she can explain what is holding up the quote and when she expects to send it once the information arrives.
Suppose the customer returns to the chat on Tuesday afternoon with the address. The team adds it to the existing request instead of assigning someone else a second quote. If Maya is away, the colleague covering for her can see both addresses, the email Maya sent and the date the customer requested. The record moves to ‘Quote sent’ only after the quote goes out. The owner can then set a follow-up date based on when the customer expects to review it.
Test the handover before relying on automation
Start with one kind of request, such as tailored quotes. Transfer the first few conversations manually and ask the colleague handling them what was missing. You may find that the team needs the service location but rarely refers to the transcript. If you’re considering AI workflow automation, decide which step is worth automating and which decision still needs a person. A correct reply in the chat does not prove that a usable CRM record was created. Connect the tools you plan to use and test that particular transfer before counting on it.
Use test contacts to check an ordinary request and several exceptions: the customer leaves out their email address, returns to add a detail, or submits a message while the CRM connection is unavailable. Check what the customer sees, whether the right person can find the request and where the team can recover a message that did not transfer. If the workflow cannot confirm that the sales team received the request, the bot should not say it has reached them.
During the first week, review a few requests rather than just counting new records. Compare the acknowledgement in the chat with the action the customer actually received. Did they have to give the same information twice? Did the promised callback happen? Did a message adding a detail create a duplicate opportunity? If customers are still chasing replies, check who owns the requests and whether the deadlines are realistic before adding more automated steps.
Practical questions to settle before launch
Should every handover create a sales opportunity?
No. A customer may need help with an invoice, a rescheduled visit or an active job. Record the request where the person responsible can find and resolve it. If the bot cannot tell whether a message is about new business or existing work, have someone review it rather than creating a sales opportunity by default.
What if the customer doesn’t leave contact details?
Ask for one suitable way to get in touch and explain why you need it. If the customer chooses not to provide it and you cannot reply in the same conversation, show them another way to reach the team. Don’t promise a callback when you have no number to call.
Can the bot confirm a price or a start date?
It can share approved general information and explain what it covers. For instance, if a published starting price applies to a single office, the bot should not present it as the total for cleaning two sites. A tailored quote, a departure from standard terms or a firm start date should be confirmed by the person responsible after they check the details.
