Close More Deals With Codex Sales Followups

Start by downloading the starter folder from this guide's resources. Then create or open a clean blank folder for the workflow in Codex. In the demo, the folder is named codex-sales-followup-demo.

The starter folder has this shape:

approval-rules.md
AGENTS.md
data/
  accounts.csv
  crm-fields.md
  email-style-guide.md
  product-notes.md
  slack-signals.md
  meeting-notes/
    acme-robotics-call.md
prompts/
outputs/

The data/ folder is where the raw context lives. The outputs/ folder is where Codex saves review artifacts. The approval-rules.md file tells Codex what it can and cannot do.

For the demo, all accounts, companies, deal values, and notes are fake. For a real setup, replace the demo files with exports from your CRM, notes app, and Slack workspace.

The attached template files are the starting point. Copy them into the blank project, then replace the sample content with your real exports or let Codex generate fresh versions from scratch if you do not have them yet. Keep the file names stable so Codex can find the expected inputs every time.

Pro tip: Keep approval-rules.md open while recording or running this for the first time. It gives Codex a source to cite when it refuses risky actions like sending email, changing CRM data, or inventing pricing.

Your approval rules should say:

Codex may inspect files, rank accounts, draft messages, and propose CRM updates.

Codex must not:
- send customer emails
- post Slack messages
- update a live CRM
- invent pricing, discounts, implementation dates, security claims, or product commitments
- use real customer data in a demo
- overwrite source data in data/

Every customer-facing message needs human review.
Every CRM update must be returned as a proposed update table.

If you are starting from scratch, open your empty project folder in Codex and have it create the workspace.

If you already downloaded the attached template files, drop them into the blank folder first. Then open the folder in Codex, or use this prompt to have Codex scaffold the same structure for you.

Use this prompt:

Create a sales follow-up workspace in this folder.

Set up:
- AGENTS.md
- approval-rules.md
- data/
- data/meeting-notes/
- prompts/
- outputs/

Create starter files for:
- account data
- CRM field rules
- product notes
- email style rules
- Slack or team-channel signals
- meeting notes

The workflow should help me:
- extract account details
- rank accounts by follow-up priority
- create customer follow-up packets
- propose CRM updates for review
- create seven-day close plans

Important rules:
- do not connect to live accounts yet
- do not send emails
- do not post Slack messages
- do not update CRM records automatically
- all outputs should be saved as Markdown files in outputs/

Codex should create a folder structure similar to the demo and leave you with prompt files you can reuse.

Pro tip: Start with exported files, then connect your note-taking app, Slack, and CRM after the local review flow works. That keeps setup simple and prevents Codex from touching live sales data too early.

Under Plugins, connect your note-taking app, CRM if you have one, and anywhere you talk to sales prospects only after the local review loop is working. That way Codex can pull account data and propose updates before it touches live records.

Once the local version works, connect your real tools carefully:

• Connect your note-taking app so Codex can read meeting summaries.
• Connect Slack or export account-team signals so Codex can see what changed.
• Connect your CRM only after the review workflow is working locally.

For the first real run, use exports or read-only access. Do not start by letting Codex write back to the CRM.

Before Codex can prioritize accounts, it needs a clean picture of the pipeline. Start with a one-time extract.

Use this prompt:

Perform a one-time extract of account details from the available source files or connected sales tools.

Create a normalized account summary with:
- account name
- owner
- deal stage
- deal value
- forecast category
- close date
- champion
- known blockers
- latest interaction
- next step
- risk level
- source evidence

Save the result in data/accounts.csv or a Markdown table if CSV is not practical.

Do not update the CRM. This is an extraction and cleanup pass only.

Then ask Codex to inspect the workspace before it drafts anything:

Inspect this sales follow-up workspace.

Do not draft messages or propose CRM updates yet.

Return:
- which files you can read
- what each file contributes
- what context is missing
- the approval rules you will follow
- where you will save outputs

This step matters because it forces Codex to show its working area before it starts making recommendations.

Now have Codex decide where follow-up matters most.

In the recording, the prompt was intentionally plain:

Look at Slack signals and meeting notes and use them to prioritize our accounts and make a follow-up priority queue.

You can make that more reusable:

Use the files in data/ to build a follow-up priority queue.

Rank each account by:
- urgency
- clarity of next action
- risk if no one follows up this week
- evidence quality
- closeness to revenue

For each account, include:
- rank
- account
- deal stage
- reason for priority
- source evidence
- next-best action
- confidence
- human review note

Save the result as outputs/followup_queue.md.

In the demo, Acme Robotics ranked first because it had a recent meeting, a security blocker, and a clear procurement deadline. Nimbus Finance moved up because it had the highest deal value, a commit forecast, and a fresh pricing signal.

That is what makes this useful. Codex is not just writing a faster email. It is deciding which follow-up deserves attention and showing the evidence behind that recommendation.

Once you know which account needs attention, run the post-call workflow.

In the recording, the live example used Nimbus Finance. The first prompt was:

Create a follow-up packet for Nimbus Finance.

Codex created outputs/followup_packet_nimbus_finance.md with:

• an account summary
• a customer email draft
• an internal Slack update draft
• proposed follow-up assets
• claims and assumptions to verify
• an approval checklist

The next prompt was:

Okay, update the CRM entry for this account and then create a close plan.

Codex did not edit the source CRM data. It created review files:

• outputs/crm_update_review_nimbus_finance.md
• outputs/seven_day_close_plan_nimbus_finance.md

That is the safer pattern. CRM changes should show current value, proposed value, source evidence, reason for update, risk, and an approve/reject field. The close plan should separate internal actions from customer-facing actions.

After approving the CRM review and close plan, the recorded prompt asked Codex to draft the email locally:

CRM update approved. 7-day close plan approved. Draft any follow-up emails in my Gmail. Don't really do that in Gmail; this is a demo.

Codex created a local file instead:

• outputs/gmail_draft_nimbus_finance.md

That file explicitly notes that no Gmail action was performed and no customer-facing send happened. In a live workflow, this is where a rep should review the email, check the assumptions, and decide whether to send.

For a reusable post-call prompt, use:

I just finished a sales call for [ACCOUNT_NAME].

Read the latest meeting notes, account record, Slack or team-channel signals, CRM field rules, and product notes.

Create:
- a customer follow-up email draft
- an internal account-team update
- a list of claims or assumptions to verify
- proposed CRM updates in a review table
- a seven-day close plan

Rules:
- do not send email
- do not post Slack messages
- do not update the CRM
- do not invent pricing, dates, discounts, security claims, or product commitments
- save all outputs in outputs/