Using custom actions to update CRM records in chat

Every CRM platform fields the same few account-maintenance requests during support: “change my billing address,” “update the contact owner,” “mark this lead as disqualified.” In a busy SaaS team those small edits still cost an agent several minutes of context-switching - looking up the record, navigating to the right screen, and confirming the change back to the customer.

When you power your support chat with Chatref custom actions, the AI agent handles those updates inside the conversation. The agent can ask for exactly what it needs (such as the record ID and the new value), validate the input with the customer, and then call your own webhook to apply the change. That keeps the interaction flowing, saves your support team the manual steps, and works especially well for CRM platforms where operators spend a large share of their day performing small record edits on behalf of customers.

A custom action consists of two parts:

1. A schema of inputs the agent collects - the data it asks the customer for (record identifier, field name, new value, optional notes).
2. A webhook URL in your backend - Chatref sends a POST request with the collected payload whenever the action is used.

When a customer says something like “please update our company name to Northwind Inc,” the agent can recognize the intent from your knowledge-base examples and the action’s description. It requests the missing pieces (for instance, “I’ll need your Account ID to make that change - do you have it?”), confirms with the customer (“I’ll change your company name to Northwind Inc, correct?”), then calls the endpoint. Your backend authenticates the request, updates the CRM record, and returns a success or error message. Chatref takes the HTTP response and shows a plain-language confirmation to the customer. If the update fails, the agent can explain why and offer to escalate the conversation to a human through the shared inbox.

The process turns a back-and-forth chain of emails or tickets into a single chat turn, with no copy-pasting between tabs.

Identify the two or three highest-frequency CRM updates your support team performs manually. Common candidates: updating a billing or shipping address, changing a lead status, reassigning a contact owner, or modifying a custom field.

Then, in Chatref, create a custom action for each one. You will provide:

• A descriptive name so the agent knows when to use it (e.g., “Update Lead Status”).
• The input fields the agent should collect - typically a record ID (account, lead, or deal ID) and a value field. You can mark fields as required.
• The webhook endpoint URL. Chatref will send a POST request with a JSON body containing the collected field values when the action fires.

On your backend, set up the endpoint to:

• Verify the request (an HMAC signature, a secret token, or an API key you control).
• Look up the record via the ID.
• Apply the update.
• Return a 2xx status with a message body the agent can use.

Test the action in the Chatref playground before enabling it for live visitors. Simulate a customer asking for the update, let the agent collect the data, and watch that your endpoint receives and processes the call correctly. If your CRM already has an internal API, you can wrap it with a small integration service that validates and formats the payload before updating.

Once the action is working, consider adding a short explanation and example conversation to your agent’s training content so it knows precisely when to trigger the action instead of just offering a help doc.

Start with the most repetitive record edits, then layer on additional patterns over time - for example, creating a new lead from a chat (combining custom actions with lead capture) or applying a bulk tag when the customer mentions a campaign code. The same pattern works for any record type: contacts, deals, tickets, or custom objects.

Protect data integrity with a confirmation step. Always build your custom actions so the agent echoes the change back and asks for explicit confirmation before the webhook fires. That small prompt prevents bad data from a misunderstood field.

Log every update attempt, including record ID, requested change, and response status, in your own application. When something goes wrong - an invalid ID, a CRM lock conflict, a network timeout - the agent should give the customer a clear next step, such as “Sorry, I wasn’t able to update that. Let me connect you with a support specialist.” That handoff to the shared inbox keeps the interaction human when it needs to be.

Use the action’s input schema to implement basic validation. For example, if a field only accepts a few allowed status values, list them in the description. The agent will then present them as options, reducing back-and-forth.

Finally, revisit the action monthly. As your product changes, the fields and values customers ask about change too. Small adjustments keep the action matching what your team actually needs.

Can chat edit CRM records?

Yes. A Chatref custom action collects the necessary details from the customer and sends them to a webhook you control. Your backend updates the CRM record using whatever logic and permissions you choose.

Is it safe to let chat write data?

You decide the safety. The webhook endpoint lives on your infrastructure so you can enforce authentication, validate every input, require confirmation from the customer, and log every attempt. The chat cannot reach your CRM directly; it only sends a predetermined payload to an endpoint you own.