Building a Deal Intelligence Agent with Outlit

You can build a deal intelligence agent by connecting your call recordings and conversation data to Outlit, then running an AI agent (Claude Code, Codex, Gemini CLI, OpenCode, or a remote MCP client such as Cursor) to query that context on demand. This guide covers four workflows you can set up today: pre-call briefs, closed-lost postmortems, deal handoffs to CS, and weekly rep coaching. Each workflow includes a ready-to-use prompt template and a complete agent instruction file you can drop into a project.

The deal story exists, it just lives in the wrong places

Every deal your team works generates a trail. Call recordings in Fireflies or Granola. Emails back and forth. Slack threads with internal context. The information is there. Getting to it before a conversation is the problem.

A rep preparing for a call with a prospect they haven't spoken to in three weeks has to piece that trail together themselves. Find the last call recording. Skim the email thread. Check if anything was flagged in Slack. By the time they're up to speed, they've spent twenty minutes on prep that should take two.

For closed-lost deals the problem is worse. The recordings exist. The story of what happened is in there. But nobody has time to go back and listen. Patterns that could improve how the whole team sells never get surfaced because the raw material is buried in a folder nobody opens.

An agent connected to Outlit changes this. Outlit indexes your call recordings, email, and Slack into a unified profile per account. It connects to Fireflies, Granola, Gmail, and Slack. The agent queries that profile on demand. Before a call, it produces a brief. After a deal closes lost, it reconstructs what happened. When a deal closes won, it packages the full story for CS before the handoff meeting.

By the end of this guide, you'll have a CLAUDE.md and four prompt templates you can copy into a Claude Code project and run today.

What the agent is actually doing

The agent is not doing anything you couldn't do manually. It is querying Outlit for a unified view of everything recorded about an account, reasoning over that context, and producing a structured output. The value is not magic. It is speed and consistency.

Outlit connects to your call recording tool and indexes recordings per account. This guide uses Fireflies and Granola as the supported call examples. Outlit also ingests email, Slack, and billing data from Stripe where connected, so the account profile the agent queries reflects the full picture of a customer relationship, not just the call history.

When the agent queries Outlit before a call, it is asking: what do we know about this account across every source we have connected? Outlit returns that context in a structured form. The agent reads it, identifies what is relevant for the specific task (a pre-call brief looks different from a postmortem), and produces an output the rep can actually use.

Four workflows, explained

How to set this up

You need Outlit with your call recording source connected, the Outlit CLI authenticated, and Claude Code with the Outlit skill installed. Authenticate the CLI with outlit auth login; if it asks for a key, create one in Settings > API Keys. Remote MCP clients use your workspace MCP URL from Settings > CLI & MCP and complete OAuth in the client.

Step 0: Install the Outlit CLI and authenticate. Run the CLI login flow and let the CLI store the credential.

curl -fsSL https://outlit.ai/install.sh | bash
outlit --version
outlit auth login

Step 1: Connect your call recording source in Outlit. Log into Outlit, go to Integrations, and connect Fireflies or Granola depending on what your team uses. Outlit will start indexing call records per account. Depending on your call volume, the initial index may take a few hours to complete. If you also have Stripe connected, that data will feed into the account profile automatically.

Step 2: Install Claude Code. You'll need Node.js 18 or higher.

npm install -g @anthropic-ai/claude-code
# Verify install
claude --version

Step 3: Install the Outlit skill for Claude Code. This gives Claude Code the Outlit workflow guidance and CLI commands it needs.

outlit setup claude-code
# Verify auth and agent setup
outlit doctor

Step 4: Create your project and add the CLAUDE.md below. Make a new directory for your deal intelligence agent, drop in the CLAUDE.md, and run the pre-call brief for one account. If the output references specific things from your call recordings, the setup is working.

What changes when this is running

Reps stop going into calls cold. Pre-call prep drops from 15-25 minutes to 2-3 minutes. The output is immediately useful and takes less time than any manual prep, so the habit sticks fast. Within a few weeks the question shifts from "should I run this?" to "why would I ever not?"

Closed-lost analysis actually happens. Most teams never do proper postmortems because they take 30-45 minutes that nobody has. With the agent, every closed-lost deal gets a structured analysis within 24 hours. Over a quarter, the pattern becomes visible: deals at a certain stage, with a certain profile, tend to stall at a specific point. That's actionable in a way that anecdotal debrief rarely is.

CS starts with real context. Handoff prep drops from 45+ minutes to under 5. The CSM already knows what was promised, what the customer was hesitant about, and who the key stakeholders are. The customer doesn't have to repeat their own story.

Coaching gets specific. The rep coaching brief gives managers concrete examples for 1:1s rather than generic pipeline conversations. "This objection came up in four of your last six calls and the conversation stalled after it both times" is a different kind of feedback than anything a manager could pull together manually in under an hour.

Get started with the CLAUDE.md

Copy the CLAUDE.md below into your project to get started. It contains the agent's role, instructions for querying Outlit, and all four workflow definitions. Below it are individual prompt templates you can run as-is, with variables to swap out per account.

# Deal Intelligence Agent
 
## Role
You are a deal intelligence agent. Your job is to surface what's known about an account from call recordings, email, and Slack, and turn that context into outputs that help sales reps and managers act faster and more accurately. You do not speculate or fill gaps with assumptions. You work with what exists in Outlit.
 
You do not send anything to a customer or prospect. All outputs are internal. The rep or manager decides what to share externally.
 
## How to use Outlit
Before every response, use the installed Outlit skill and CLI to fetch context for the relevant account. Do not rely on prior conversation context or general knowledge about a company. Fetch current data on every request.
 
Pass the account name or ID. Outlit returns the unified profile: call recordings and highlights, email thread summaries, and Slack context where connected.
 
If a specific piece of data is missing (no call recordings found, email not connected), say so explicitly. Do not invent context.
 
## Output principles
- Be specific. Reference actual call content, actual email threads, actual Slack messages. Generic summaries are not useful.
- Be brief. A pre-call brief should fit in a single Slack message. A postmortem should fit on one page. If it's longer, cut it.
- Distinguish facts from patterns. A fact is something that happened. A pattern is something that happened more than once.
- Never send, forward, or share anything with an external party.
 
## Workflow instructions
 
### Pre-call brief (on demand)
1. Query Outlit for the account: call recordings, email, Slack
2. Identify last interaction date, what was most recently discussed, any open commitments or unanswered questions from prior calls
3. Note any signals that changed since the last call: tone shift, new stakeholder introduced, topic that came up for the first time
4. Produce a brief with three sections: where things stand (2-3 sentences), what to reference or build on from last time, what to watch for in this conversation
5. Produce the brief and surface for review. Keep it under 200 words.
 
### Closed-lost postmortem (triggered)
1. Query Outlit for the full interaction history: all calls, email thread, any Slack context
2. Reconstruct the timeline: first contact, key moments, where momentum slowed, last meaningful interaction
3. Identify recurring objections that were never fully resolved, commitments made but not followed up on, point where tone shifted
4. Produce a postmortem with four sections: deal timeline (key moments only, not every call), where it broke down (specific, not general), objections that went unresolved, one thing that could have changed the outcome
5. Produce the postmortem for team review. Keep it to one page.
 
### Deal handoff to CS (triggered)
1. Query Outlit for all interactions across the sales cycle
2. Extract from the call recordings and email: what was explicitly promised during the sale, concerns or hesitations the customer raised, key stakeholders and their roles, how long the sales cycle took and what triggered the close
3. Produce a handoff brief structured for a CS audience: what the customer thinks they bought (their words if possible), commitments made by the sales team, concerns that are still open going into onboarding, key people to know
4. Produce the handoff brief and surface for review. Do not write this for a sales audience. Write it for someone who has never spoken to this customer before.
 
### Rep coaching brief (weekly)
1. Query Outlit for the rep's call activity in the past 7-14 days
2. Identify across calls: recurring objections, topics handled well, points where conversations consistently stall
3. Note any accounts that have gone quiet or have open questions from the last call that were never followed up on
4. Produce a coaching brief for the manager: patterns across recent calls (specific examples), accounts that need attention, one or two concrete suggestions for the 1:1
5. Produce the coaching brief for the manager's review. Do not share with the rep unless the manager explicitly asks.
 
## What you should never do
- Contact a prospect or customer directly in any form
- Speculate about why a deal was lost without grounding it in actual call or interaction data
- Share a rep coaching brief with the rep without manager approval
- Skip querying Outlit and rely on your own knowledge of a company

Prompt templates

Variables in gold are yours to swap out. Everything else runs as written.

Prepare a pre-call brief for [account name].
 
I have a call with them in [time, e.g. 30 minutes / tomorrow morning]. Rep on the call: [your name or the rep's name]
 
Query Outlit for everything we have on this account: call recordings, email, Slack. Produce a brief with three sections: where things stand, what to reference or build on from last time, what to watch for in this conversation.
 
Surface the brief for [rep name] to review. Keep it under 200 words.

[Account name] just closed lost.
 
Query Outlit for the full interaction history on this account. Reconstruct what happened: key moments, where momentum slowed, recurring objections, last meaningful interaction.
 
Produce a postmortem with four sections: deal timeline, where it broke down, objections that went unresolved, one thing that could have changed the outcome.
 
Produce the postmortem for team review. Keep it to one page.

[Account name] just closed won. Assigned CSM is [CSM name].
 
Query Outlit for all interactions across the sales cycle.
 
Produce a handoff brief for the CSM covering: what the customer thinks they bought (use their words where possible), commitments the sales team made, concerns or hesitations that are still open, key people to know.
 
Write it for someone who has never spoken to this customer. Produce the brief for [CSM name] to review.

Prepare a weekly coaching brief for [rep name].
 
Query Outlit for their call activity over the past [7 or 14] days.
 
Identify across their recent calls: recurring objections or patterns, accounts that have gone quiet or have unresolved open questions, one or two specific things worth addressing in the 1:1.
 
Produce the brief for my review. Do not share with [rep name] unless I ask you to.

Frequently asked questions

How do I build a deal intelligence agent?

Connect your call recording source (Fireflies or Granola) to Outlit, install and authenticate the Outlit CLI, then run outlit setup claude-code for Claude Code. Write an instruction file that defines your workflows: pre-call briefs, closed-lost postmortems, CS handoffs, and rep coaching. The agent queries Outlit for account context before every response and produces structured outputs for review.

How is this different from a call recorder?

Fireflies and Granola capture call recordings or notes. They're the source of the data. Outlit sits on top and unifies those records with email, Slack, and billing into a single account profile that agents can query. The call recorder tells you what was said. Outlit makes it available to the agent that's about to help your team act on it. You use both.

What call recording tools does Outlit support?

Outlit connects to Fireflies and Granola. It indexes call records per account and makes them queryable alongside email and Slack context. Whichever of those two tools your team uses, the workflows in this guide run the same way.

Does the agent listen to calls directly?

No. The agent queries Outlit, which indexes the call recordings and transcripts already captured by your recording tool. It reasons over the structured summaries and highlights, not the raw audio.

Do I have to use Claude Code?

No. For coding agents, Outlit supports outlit setup claude-code, outlit setup codex, outlit setup gemini, outlit setup opencode, and other documented setup aliases. For remote MCP clients such as Cursor or VS Code, use the workspace MCP URL from Outlit and complete OAuth in the client. Custom scripts can use the API with an Outlit API key.

What if my call recording data is patchy?

Output quality reflects input quality. If your team isn't recording calls consistently, the agent will flag gaps in its output rather than invent context. Start by fixing recording hygiene, then roll out the workflows.