A Practical AI CRM Follow-Up Workflow Template That Closes Faster

Most leads go cold because nobody replies fast enough. An AI CRM follow-up workflow fixes that by drafting the first reply within minutes of a form submission or missed call — before the lead moves on to the next tab.

The catch: most teams either automate too early (and send sloppy generic replies) or never start (and lose the lead to silence). The fix is a narrow, draft-first workflow with an approval gate. The AI prepares a contextual follow-up. A human reviews it. Then it ships.

Below is a copy-paste template, a decision rule for when to automate, and the boundaries that keep this safe.

The short answer

• Trigger: form submission or missed call.
• Source: CRM record plus the original inbound message.
• Output: a draft follow-up with context — not a sent message.
• Approval gate: owner (or assigned rep) reviews before send.
• Expand later: only after the first workflow runs cleanly for two weeks.

The AI prepares. The human decides. That is the whole model. If you skip the approval gate, you are not automating follow-up — you are outsourcing your first impression to a model that has not earned it yet.

If you cannot fill in the trigger, source, and approval gate, the process is not ready for automation. Fix the manual version first. A broken follow-up process automated is still a broken follow-up process, just faster.

The concrete example

A consulting firm gets 30 leads a week from a contact form. The owner checks email between client calls, so the fastest reply often comes 4-6 hours after the lead submitted. Half those leads have already emailed a competitor by then.

The new workflow: when a form submission lands, the AI reads the CRM record and the submitted message, drafts a reply that references what the lead actually asked about, and posts the draft into the owner's review queue. The owner taps approve or edits one line. The reply sends within 10 minutes of the original submission.

Same owner. Same judgment. Sixty percent faster first touch. No generic autoresponders, no risk of a bad reply going out unchecked.

Copy-paste runbook

Paste this into your agent's runbook file or workflow tool. Edit the bracketed fields.

# Runbook: AI CRM Follow-Up Draft

## Purpose
Draft a contextual first follow-up for new inbound leads, then route
to the owner for approval before sending. The assistant never sends
directly and never changes lead status without approval.

## Trigger
- Form submission received, OR
- Missed call logged in CRM with a caller match, OR
- New lead created with source = "inbound" and status = "new"

## Sources (read-only)
- CRM lead record: name, company, source channel, form answers
- Original inbound message (form body, email text, or call note)
- Owner's saved reply preferences (tone, length, signature)
- Offer or pricing runbook if the lead asked about a specific product

Do not read unrelated records. Do not pull in data the owner has not
approved for this workflow.

## Output
A single draft follow-up message containing:
- A greeting using the lead's first name
- One specific reference to what they asked about (not a generic line)
- One clear next step (book a call, reply with details, or confirm interest)
- Owner's standard signature
- Suggested send time (default: now, if approved)

Format: posted into the owner's review channel (Slack, email draft, or
CRM activity). Never sent directly to the lead.

## Approval gate
- Owner reviews the draft within 30 minutes of trigger.
- If approved: assistant sends the message using the approved channel.
- If edited: assistant sends the edited version.
- If rejected: assistant logs the rejection reason and stops.
- If no response in 2 hours: assistant sends one reminder to the owner
  and then holds. It does not auto-send.

## Stop conditions
- Lead has already been contacted today — do not draft a duplicate.
- Lead source is missing or untrusted — escalate, do not draft.
- The inbound message is empty, abusive, or clearly spam — flag, do not draft.
- The lead is marked "do not contact" — stop immediately and log.

## Receipt
After each run, log:
- Trigger type and timestamp
- Lead ID and source
- Draft produced (yes/no)
- Approval status (approved / edited / rejected / no response)
- Time from trigger to approved send
- Any missing context the owner had to add

## Never
- Send a follow-up without approval.
- Change lead status, value, or owner assignment without approval.
- Draft for leads flagged "do not contact" or "existing client."
- Use a generic template that ignores what the lead actually wrote.

When to automate vs when to keep manual

Not every follow-up belongs in this workflow. Use this decision rule:

Automate the draft when:

• The lead is new, inbound, and not yet contacted.
• The inbound message has enough detail to write a specific reply.
• The response window matters (minutes to hours, not days).
• You can define one clear next step for the lead.

Keep manual when:

• The lead is an existing client with history — context is too rich for a cold template.
• The inbound message is vague ("just looking") and needs judgment, not a draft.
• The deal is high-value or regulated and every word carries risk.
• The owner has a relationship and a personal reply will land better.

If you automate the manual cases, you send generic replies to people who deserve a human one. That costs more than the time you saved. A bad first impression on a warm lead is worse than a slow one.

Common mistakes

• No approval gate. If the assistant sends directly, one bad draft reaches a real lead. Always draft-first.
• Generic templates. If the draft does not reference what the lead asked, it reads like spam. The whole point is specificity.
• Expanding too early. Running two follow-up workflows before the first one is boring means neither runs well. Nail one, then add.
• No stop conditions. Without them, the assistant drafts duplicates or contacts flagged leads. Stop conditions are not optional.
• Measuring volume, not accepted drafts. A workflow that produces 50 drafts with 40 rejections is not working. Track approval rate.

Next step

Start with one trigger, one source, and one approval channel. Run the runbook above in draft-first mode for two weeks. Track how many drafts the owner accepts with light edits versus rejects. When the acceptance rate is high and the workflow is boring, add a second trigger or a second lead segment.

If you want a worker that runs this runbook with memory, context, and approval gates built in, see the AI CRM Follow-Up Agent page. For the broader setup pattern, read AI Workflow Automation That Saves Owner Time and Markdown Runbooks for AI Agents.

<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{
"@type": "Question",
"name": "When should you automate CRM follow-up?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Automate CRM follow-up when the lead is new, inbound, and not yet contacted, the inbound message has enough detail to draft a specific reply, the response window is hours not days, and you can define one clear next step. Keep follow-up manual for existing clients, vague inquiries, high-value or regulated deals, and leads where a personal reply will land better than a drafted one."
}
},
{
"@type": "Question",
"name": "What does an AI CRM follow-up workflow include?",
"acceptedAnswer": {
"@type": "Answer",
"text": "An AI CRM follow-up workflow includes a trigger (form submission or missed call), read-only sources (CRM record and original inbound message), a draft output with specific context, an approval gate where the owner reviews before send, stop conditions to prevent duplicates or flagged contacts, and a receipt that logs what happened for each run."
}
}
]
}
</script>