How to get feature requests from sales calls into your product backlog automatically

The short answer. To get feature requests from sales calls into your backlog automatically, have an agent read every call, email and Slack thread, find each request, and add it to the issue it belongs to with the customer attached, so every issue lists the customers who asked and links to what they said. Day AI does this for teams in place of their CRM, or on top of Salesforce or HubSpot: a skill runs when each meeting recording is ready and adds the requests to Linear through Linear's MCP server.

Say you run customer success at a data-infrastructure company, and it's 2:15 on a Tuesday. A renewal call has just ended, and halfway through it the customer's admin asked, almost in passing, for audit log exports they can hand to their security team. Before your next call, your Day AI agent has read the recording, found that request and searched Linear. There's already an issue for audit log exports, so it added a comment there: the customer's name, the company, what the admin asked for and why, and the call it came from. It found a second request that nothing in Linear matched, so it drafted a new issue and asked you in Day AI before creating it. You read it, approved it, and the new issue went into the backlog with the customer attached. Nobody pasted anything into a ticket, and the product manager who opens the audit log issue on Thursday sees one more customer under it, with the call to go read.

Where the usual answer goes wrong. Most advice tells you to give reps a feature-request field, a form or a Slack channel, or to wire a workflow that opens a new ticket after every call. The first depends on reps remembering to type, and the second fills the backlog with duplicates that have lost the customer. Day AI's view is that the conversations are the record: the request, who made it and why are already in the call, so an agent should do the filing and a person should approve what matters. You may already have a recorder on your calls and a channel where requests get posted. That was the right start.

Christopher O'Donnell, Day AI's CEO, made this argument in his October 20, 2026 newsletter: product gets closer to revenue by getting closer to customers, to what they do in the product and what they say on calls. "If you're customer-oriented, the revenue tends to take care of itself." In the same issue he describes how Day AI's own product team works: when a customer asks for something on a call or in an email, it lands in Linear on the issue it belongs to, with the customer attached, so six requests on an issue are six customers you can name and go read.

The request trail, with Day AI

A request is only useful to product if it arrives with three things: who asked, what they said, and where it belongs. The request trail is five steps that keep all three, from the moment a customer speaks to the moment someone plans the work. Any setup that says it files requests for you should cover all five.

1. Capture every conversation. Requests come up on calls, in email threads and in shared Slack channels, so the trail starts with all three. In Day AI, the notetaker records meetings, Gmail and Google Calendar connect per person, and Slack channels flow in once the Day AI app joins them.
2. Find the request. Something reads each conversation and pulls out what the customer asked for and why, in their terms. In Day AI, a skill does this when a meeting recording is ready, and on a schedule for email and Slack.
3. Match it to the issue it belongs to. Search the backlog for the existing issue before creating anything, because most requests are repeats. In Day AI, the agent searches your issue tracker through its MCP server and proposes a new issue only when nothing fits.
4. Attach the customer. Add the customer's name, company and the conversation it came from to that issue, so a count becomes a list of names. In Day AI, the agent writes that comment on the issue, and every answer it gives later cites the call clip or the message it came from.
5. Read it twice. The call that carries the request also tells you whether the deal will close and who else has to sign off. In Day AI, product reads it for the roadmap and RevOps reads the same call for the forecast, from one customer record.

How to get feature requests into Linear with Day AI

These steps use Linear. For another issue tracker, connect its MCP server in Day AI the same way: agents read from any MCP server you connect, and write to it where the server has write tools and you allow them. If you run Day AI in place of your CRM, the customer record in step 6 is Day AI's own; on top of Salesforce or HubSpot, the agent also updates them through their MCP servers.

Step 1: Connect calls, email and Slack to Day AI

Turn on Day AI's notetaker so it records meetings and writes each one into Day AI when it ends. If your team records in Gong, connect it and your call history comes in with transcripts; Granola meetings come in with their transcripts too. Each person connects Gmail and Google Calendar and sets their own sharing rules, so the agent sees only what they chose to share. Add the Day AI app to the Slack channels where customers and your team talk about the product.

Step 2: Connect Linear to Day AI and choose its tool settings

Linear's MCP server has "tools available for finding, creating, and updating objects in Linear like issues, projects, and comments" (Linear MCP docs, read November 2, 2026). In Day AI, open your settings, go to Connectors and connect Linear with your own Linear login. Then set each Linear tool to Always allow, Needs approval or Off. A good start: searching issues and adding comments on Always allow, creating issues on Needs approval, and anything that deletes or closes work Off.

Step 3: Create the feature-request skill in Day AI

In Day AI, create a skill for your agent and set its trigger to run when a meeting recording is ready. Write its instructions in plain language:

• Find every feature request in the meeting: what the customer asked for, the problem behind it, and how strongly they said it.
• Search Linear for the issue each request belongs to. If one matches, add a comment with the customer's name, company, what they said and the meeting it came from.
• If nothing matches, propose a new issue with the same details and wait for approval.
• Skip requests already on that issue from the same customer.

Attach the Linear connector to the skill. Add a second skill on a schedule, for example every Friday, that does the same for the week's email and Slack, since many requests arrive in writing.

Step 4: Start the Day AI skill in learning mode

In learning mode, Day AI reports what the skill would have added to Linear and asks for feedback before it's allowed to act. Read a week of those reports. When the skill matches a request to the wrong issue, reply and tell it why; the agent edits the skill, and later runs use the change. Every run is kept in the skill's run history, so you can see what it did and where it delivered.

Step 5: Ask Day AI which customers asked

Once requests are attached to issues, anyone in product can ask the agent in Day AI: "Which customers asked for audit log exports this quarter?" or "What did they say about it on calls?" The answer names the customers and cites the call clip or the message each request came from. From Claude, the same question works through the Day AI MCP.

Step 6: Let RevOps read the same calls in Day AI

The renewal call that carried the request also carries the timeline, who else has to sign off, and the risk. Day AI keeps that conversation on the customer record for RevOps, and a post-call skill can propose the deal updates for the rep to approve (see how to update your CRM from call recordings).

Picture yourself running customer success at a developer-platform company on Salesforce. Your CSMs get a Monday renewal forecast from Day AI. After a call, the follow-up email arrives about 30 minutes after it ends, written in the CSM's own voice, and Day AI asks in Slack or email to attach the people from the call to the opportunity in Salesforce; the CSM approves it there. Add the feature-request skill to your team's agents and the same calls feed the backlog as well as the forecast. Product reads the call for the roadmap; RevOps reads it for the forecast. To go further on the product side, Day AI's free course The Revenue-Oriented Product Team covers how a product org gets closer to customers and revenue.

Where the usual way breaks

• A field reps have to fill in. Reps remember the request that came up loudly and skip the one mentioned in passing, and what they type is a paraphrase. In Day AI, the skill reads the call itself, so a request doesn't depend on anyone remembering it.
• One new ticket per call. A workflow that opens a ticket after every call turns ten customers asking for the same thing into ten tickets nobody merges. In Day AI, the skill adds each request to the issue it belongs to and asks before creating a new one.
• Votes without names. A count of "+1"s tells you demand exists but not who to call or what they meant. In Day AI, each request on an issue carries the customer and the conversation, and the agent answers "who asked?" with citations.
• Only the calls a recorder saw. Requests in email threads and shared Slack channels never reach a call-only workflow. In Day AI, the same skills read email and Slack alongside meetings.
• Private conversations leaking into a shared backlog. A homemade script with one admin login can post from mail its owner never meant to share. In Day AI, an agent works with the permissions of the person it works for, and a skill deployed to the whole workspace acts only on what's shared with the workspace.
• Product and RevOps reading different copies. When the request is retyped into a ticket and the deal notes are retyped into the CRM, the two drift apart. In Day AI, both teams read one customer record built from the same conversations.

What Day AI adds

Day AI provides the context graph and agent control plane to quickly and securely upgrade your company brain. For a product team, that means the backlog stops being a list of features with vote counts and becomes a list of customers you can name, with what each one said, while the same calls keep the forecast current. See pricing for how agents are priced.