AI SOP Template: Sales Outreach Process
ZarifSales teams that bolt AI onto outbound without a real SOP end up with thousands of identical low-quality emails and no replies. The teams winning in 2026 are running structured AI-augmented outreach SOPs where each step has a clear owner, a defined AI prompt, and a human checkpoint before anything ships. This is the playbook, written so you can copy it into a Notion doc and start running it on Monday.
An AI sales outreach SOP is a documented sequence of repeatable steps that combines AI tools (research, drafting, scoring) with human judgment (review, send, follow up) to produce personalized outbound at scale.
TL;DR
- A working SOP cuts outreach drafting time from 25 minutes per prospect to under 4 minutes
- Reply rates climb from 1 to 2 percent (templated) to 8 to 12 percent (AI-augmented with human review)
- Each role (rep, RevOps, manager) has explicit ownership over named steps and prompts
- Run the cycle weekly, not daily, so you can analyze and iterate on what is converting
- Never let AI send unattended; the human checkpoint before send is the single highest-ROI step
Why generic AI outbound is failing in 2026
Inboxes are saturated. Buyers can spot a templated AI-generated email in two seconds: the over-polished hook, the bolted-on personalization line, the closing CTA that sounds like it came out of a textbook. Reply rates on pure-AI cold outbound have collapsed below 1 percent in most categories. The teams beating this are using AI as a research and drafting accelerant inside a tightly-scoped SOP, not as an autonomous send-bot.
The SOP below assumes a 5-person sales pod with reps owning their named accounts, a RevOps function (or one operations-savvy rep) maintaining the playbook, and a manager who reviews outcomes weekly. It scales down to a solo founder and up to a 50-person team with minor adjustments.
SOP roles and ownership
Three roles, three responsibilities. Pin this in the playbook so nobody is guessing who runs what.
- Sales rep: Owns the account list, runs the AI prompts for research and drafting, conducts human review, hits send, manages replies and follow-ups.
- RevOps lead: Maintains the prompt library, the ICP definition, the outreach templates, and the AI tool stack. Audits a sample of sent emails weekly.
- Sales manager: Sets the weekly KPI targets, reviews reply and meeting-booked metrics, runs the retro, approves prompt changes that affect everyone.
Step 1: Account research and prioritization
Owner: Sales rep | Frequency: Weekly, every Monday morning | Time: 30 minutes for 25 accounts
The rep pulls the 25 highest-priority accounts from the CRM. Run each through an AI research prompt that returns a structured summary. The Clay or Apollo enrichment route works for firmographics; ChatGPT or Claude with web search handles the qualitative signals.
Prompt to run on each account:
"Research [COMPANY NAME] and return a structured summary with: 1) headcount and growth trajectory in last 12 months, 2) current tech stack signals from job posts, 3) any leadership changes in last 6 months, 4) recent funding or earnings news, 5) one specific business pain that our [PRODUCT] would solve based on what they publicly say. Cite sources."
The rep ranks the 25 accounts into A (clear pain match, recent trigger event), B (good fit, no trigger), and C (cold). Only A and B accounts proceed to drafting.
Step 2: ICP and buyer-persona match
Owner: Sales rep | Frequency: Per account, once it enters the active list | Time: 5 minutes per account
For each A and B account, identify two contacts: an economic buyer (VP-level, owns budget) and a champion (director-level, feels the pain). Pull from LinkedIn Sales Navigator or Apollo. Run a second AI prompt to summarize each contact's recent activity:
"Summarize [CONTACT NAME]'s LinkedIn activity over the last 90 days: posts, comments, articles shared, job changes. Identify three topics they care about and one recent post I could authentically reference."
Save the output in the CRM contact record. This is the raw material for personalization.
Never paste raw CRM data, internal pricing, or customer lists into a public AI tool. Use Claude Enterprise, ChatGPT Team, or a self-hosted model when the prompt contains anything sensitive. Leakage of one prospect list to a public model is a fireable offense in most enterprise sales orgs.
Step 3: AI-drafted email sequence
Owner: Sales rep | Frequency: Per contact, after Step 2 | Time: 4 minutes per email
The rep runs a drafting prompt that produces a 4-email sequence (initial, breakup attempt 1, value-add resource, final). The prompt template lives in the team's prompt library and gets versioned by RevOps.
Drafting prompt:
"Write a 4-email outbound sequence to [CONTACT NAME] at [COMPANY]. Their pain: [PAIN FROM STEP 1]. Recent post: [POST FROM STEP 2]. Our product: [ONE-LINE VALUE PROP]. Tone: peer-to-peer, no jargon, 60 to 80 words per email. Email 1 must reference the recent post in a non-cringe way. Email 4 is a true breakup, no CTA. Output as plain text with subject lines."
The rep reviews each draft in under 60 seconds. Flag for rewrite if: the personalization line sounds shoehorned, the value prop reads as generic, or the CTA is overly aggressive.
Step 4: Human review and send
Owner: Sales rep | Frequency: Daily during active campaign | Time: 60 to 90 seconds per email
This is the single highest-ROI step in the entire SOP. Do not skip it, do not delegate it to AI, do not batch-approve. The rep reads each email out loud (yes, out loud) before sending. If it does not sound like something a human would actually write to another human, rewrite the offending line.
Send through your sequencing tool (Outreach, Salesloft, Apollo, Lemlist) on a Mon/Tue/Thu schedule with 3 to 4 day spacing between touches. Cap at 50 sends per rep per day to maintain deliverability.
Step 5: Reply triage and response
Owner: Sales rep | Frequency: Twice daily, 9am and 3pm | Time: 15 minutes per session
Replies go to one of four buckets: positive (book the meeting), neutral curiosity (send a tighter value pitch), objection (handle directly), or out-of-office / wrong contact (re-route). Use an AI assistant to triage the inbox at the start of each session.
Triage prompt:
"Categorize each of these inbound replies into POSITIVE, CURIOUS, OBJECTION, or REROUTE. For each, suggest a one-sentence next action. Reply text: [PASTE]"
Never let AI auto-respond to replies. The moment a prospect engages, every word from the rep needs to be human-authored. The cost of a bot-feeling reply at this stage is killing a live opportunity.
Step 6: Weekly review and prompt iteration
Owner: Sales manager and RevOps lead | Frequency: Friday afternoon, 45 minutes | Time: 45 minutes
Pull the week's metrics from the sequencing tool: emails sent, open rate, reply rate, meeting booked rate, meeting-to-opportunity rate. Review a sample of 10 sent emails per rep for quality. Identify the top-performing email by reply rate and the worst, then update the prompt templates accordingly.
Critical KPI targets for a tuned SOP:
- Open rate: above 50 percent
- Reply rate: 8 to 12 percent on A-tier accounts
- Meeting booked: 2 to 4 percent of sent emails
- Time per prospect end-to-end: under 15 minutes
If reply rate falls below 5 percent for two weeks in a row, the prompts have drifted into generic AI-speak and need a reset.
Tool stack for this SOP
The SOP works on any combination of these tools. Pick what your team already uses and build the prompt library inside it.
- AI drafting: ChatGPT Team ($30/user/month), Claude Pro ($20/month), or Gemini Advanced ($20/month)
- Account research: Clay, Apollo, ZoomInfo, or Common Room
- Sequencing: Outreach, Salesloft, Lemlist, Smartlead, or Apollo
- CRM: HubSpot, Salesforce, Close, or Pipedrive
- Prompt library: Notion, Coda, or PromptLayer
Total tooling cost for a 5-rep pod runs $400 to $800 per month all-in, dwarfed by the salary cost of the reps themselves.
Common SOP failure modes
Three patterns kill AI outbound SOPs. First, skipping the human review step "just this once" until it becomes never. Second, letting prompts age without iteration so every rep ends up sending the same flavor of AI-polished email. Third, treating AI-generated personalization as a substitute for actually understanding the buyer's business.
A well-run SOP feels lighter every week as the prompt library matures. If it feels heavier, you are not iterating on the prompts, you are just adding steps.
FAQs
How long does it take to set up this SOP from scratch?
Two weeks of focused work. Week one is documenting the playbook, building the prompt library, and training the team. Week two is running it live with manager review on every email so the team calibrates. By week three the cycle should run smoothly with only the Friday review touchpoint required.
Can a solo founder run this SOP without a RevOps function?
Yes. The solo version compresses the three roles into one and runs at a smaller scale (10 to 15 accounts per week instead of 25 per rep). The prompt library still matters; just maintain it in a single Notion page and version it manually.
Which AI tool produces the best cold email drafts in 2026?
Claude Sonnet 4.5 produces the most natural-sounding sales drafts in 2026, followed by GPT-5.4. Both win against generic outbound tools because they preserve voice better. Avoid using inbox-integrated AI features in tools like Outreach for the actual drafting step; their fine-tuned models tend toward formulaic output.
What reply rate should I expect from a properly run AI outbound SOP?
8 to 12 percent on A-tier accounts where the personalization is grounded in real research. B-tier accounts run 4 to 6 percent. Anything below 4 percent on A-tier means the prompts are producing generic output or the ICP definition is too loose.
Is it ethical to use AI for cold outreach?
Yes, when the AI is researching, drafting, and accelerating the rep, not impersonating one. The line is whether a real human reviewed and approved every word that hits the prospect's inbox. Auto-sent AI emails with no human in the loop are increasingly seen as spam by both buyers and email providers, and they tank deliverability fast.
How do I keep prospect data secure when using AI tools?
Use enterprise tiers (ChatGPT Team or Enterprise, Claude for Work, Gemini Workspace) which include zero-data-retention guarantees. Never paste raw exports of your CRM, customer lists, or pricing into consumer-tier AI tools. Anonymize any data point you cannot avoid sharing.
