Overview

You have a process that's been running for years. It lives in someone's head, scattered across emails, and in a Word doc that hasn't been updated since 2019. It needs to be formalized as an SOP. Or you need to create a new SOP from scratch. You could spend three days interviewing people and writing it. Or you could spend 45 minutes using AI to draft it, then two hours refining it. The difference is knowing how to use AI for SOP work specifically.

SOPs are one of the highest-leverage uses of AI in operations. They're tedious to write, they're critical to get right, and they have a clear structure that AI can help with. This lesson teaches you the exact prompts, the verification process, and the common pitfalls when using AI for SOP work.

Why AI is Good at SOPs (And Where It Falls Short)

AI is good at SOPs because SOPs have a predictable structure: step, owner, input, action, output. AI can generate this structure quickly. SOPs also benefit from clear, plain-language writing, that's an AI strength.

AI falls short when the SOP is industry-specific, has compliance requirements, or has unspoken complexities baked into the current process. An AI-generated SOP for a generic process might be 90% ready. An AI-generated SOP for a regulated or complex process might be 50% ready.

Strategy: Use AI for drafting and structure. Use humans for context, compliance, and verification.

The SOP Prompt: Getting It Right the First Time

Writing an SOP from scratch (vs. updating an existing one) requires a detailed prompt that captures the current process. Here's the structure:

Role: You are a business process analyst with experience writing standard operating procedures.

Context:
- Organization: [Size, industry, structure]
- Process: [Name of process]
- Current state: [How it's done today]
- Pain points: [What's broken or inefficient]
- Who's involved: [Departments/roles]
- Key constraints: [Budget, tools, staffing]

Task: Write a complete SOP for [process name]. The SOP should be suitable for [audience/use case].

Format:
- Title and purpose statement
- Numbered steps (target: X-Y steps)
- For each step: Step # | Owner | Input | Action | Output | Timeline
- Include decision points where relevant
- End with troubleshooting section

Constraints:
- Use only existing tools/systems (don't suggest new software)
- Assume [skill level of person executing]
- Include all compliance requirements
- Don't assume additional headcount

Here's a complete real example:

Role: You are a process documentation specialist for a professional services firm. You understand how to write clear, detailed procedures for consulting operations.

Context:
- Organization: 80-person management consulting firm with 6 partners, 12 senior consultants, and 30 junior consultants
- Process: Proposal development and approval
- Current state: When a client asks for a proposal, the Partner handling the relationship asks a Business Development Manager to draft it. The BDM interviews the Partner, researches the scope, drafts a 2-5 page proposal in Word. They email it to the Partner. Partner reviews it (usually takes 3-5 days because they're busy). Partner gives feedback. BDM revises. Partner approves. BDM sends to client.
- Pain points: This takes 2-3 weeks. Clients often complain about slow response. Sometimes proposals don't include important scope items because the BDM didn't understand them. Sometimes the price is wrong because nobody verified the budget against our rates.
- Who's involved: Business Development Manager (drafting), Partner (review/approval), Project Manager (validation that we can resource the work)
- Key constraints: We use Word for documents, Excel for pricing, email for communication. We can't add headcount.

Task: Write an SOP for proposal development. The SOP should speed up the process and ensure we don't miss critical details.

Format:
- Title and purpose
- 8-12 numbered steps
- For each step: Step # | Owner | Input | Action | Output | Timeline
- Include decision points for different proposal types (small vs. large projects)
- Include a checklist for what gets reviewed before approval

Constraints:
- Use only Word, Excel, and email
- Target timeline is 5 business days from client request to proposal delivery
- Assume BDMs have 2-3 years experience but may not know consulting terminology
- Include internal validation by Project Manager before Partner approval

That's a solid prompt. It gives the AI enough context to draft something real. Now what does it generate?

Real Example: From Messy Process to Clean SOP

The messy process (what you're starting with):

"We handle customer complaints via email and phone. The person who gets the complaint asks the service team if there's an issue. If there is, we schedule a call with the customer. If not, we tell them we didn't see a problem. Sometimes this takes days. Sometimes customers call back angry because we didn't solve it. The VP of Customer Success oversees this but isn't directly involved."

AI-generated SOP (first pass):

CUSTOMER COMPLAINT RESOLUTION SOP

Purpose: Ensure customer complaints are addressed promptly and resolved to customer satisfaction.

Step 1: Receive Complaint
- Owner: Customer Success team
- Input: Email or phone call from customer
- Action: Log complaint in tracking system with customer name, issue description, and contact info
- Output: Complaint ticket created in system
- Timeline: Same day

Step 2: Acknowledge Complaint
- Owner: Customer Success team
- Input: Complaint ticket
- Action: Send customer an acknowledgment email within 4 business hours confirming receipt
- Output: Acknowledgment email sent
- Timeline: 4 business hours

Step 3: Investigate Issue
- Owner: Relevant service team (Support, Engineering, Billing)
- Input: Complaint details
- Action: Review customer account/service to understand the issue. Document findings.
- Output: Investigation notes in ticket
- Timeline: 24 business hours

Step 4: Determine Resolution
- Owner: Service team lead
- Input: Investigation findings
- Action: Determine if the complaint is valid. If valid, propose solution. If not, prepare explanation.
- Output: Resolution plan
- Timeline: 24 business hours

Step 5: Contact Customer
- Owner: Customer Success team
- Input: Resolution plan
- Action: Call customer (or email if preference noted) to explain resolution and next steps
- Output: Customer communication logged
- Timeline: Within 2 business days of investigation complete

Step 6: Implement Solution
- Owner: Relevant service team
- Input: Agreed-upon resolution
- Action: Implement the fix/refund/credit/replacement
- Output: Solution implemented, documented in ticket
- Timeline: 2-5 business days depending on solution type

Step 7: Verify Resolution
- Owner: Customer Success team
- Input: Customer feedback on solution
- Action: Follow up with customer to confirm the issue is resolved
- Output: Resolution confirmation in ticket
- Timeline: 3-5 business days after solution implemented

Step 8: Close Ticket
- Owner: Customer Success team
- Input: Resolution confirmed
- Action: Update ticket status to "Closed" and archive
- Output: Ticket closed
- Timeline: Same day as confirmation

Your assessment: This is 70% done. The structure is good. But it's missing some things:

• What if the customer complains about the same issue multiple times? (No aggregation logic)
  - What's the definition of "valid"? (Missing decision criteria)
  - Who escalates to VP of Customer Success and when? (Missing escalation)
  - What if investigation takes longer than 24 hours? (No exception handling)

Your iteration request:

Good structure. I need you to add:

1. A decision point in Step 3: "If the same issue has been reported by 3+ customers in the past 30 days, flag this as a systemic issue and notify the VP of Product." Add this as Step 3b.
2. In Step 4, clarify the "valid" determination: A complaint is valid if it's about our service, not customer misuse. Valid examples: system downtime, missing feature, billing error. Invalid examples: customer didn't read the documentation, they're using the product wrong, it's working as designed.
3. Add an escalation rule: If resolution takes more than 5 business days, or if customer has complained about the same issue before, flag to VP of Customer Success for direct involvement.
4. In Step 5, add: "If customer is upset, offer [discount/credit/extension] as goodwill. See pricing guidelines in Appendix A."

AI-updated SOP (second pass, with your additions):

CUSTOMER COMPLAINT RESOLUTION SOP

Purpose: Ensure customer complaints are addressed promptly and resolved to customer satisfaction. Escalate systemic issues and ensure customer retention.

Step 1: Receive Complaint
- Owner: Customer Success team
- Input: Email or phone call from customer
- Action: Log complaint in tracking system. Document: customer name, issue description, contact info, and whether they've complained about this before
- Output: Complaint ticket created with full context
- Timeline: Same day

Step 2: Acknowledge Complaint
- Owner: Customer Success team
- Input: Complaint ticket
- Action: Send acknowledgment email within 4 business hours. Thank customer, confirm we're investigating, give expected timeline.
- Output: Acknowledgment email sent
- Timeline: 4 business hours

Step 3: Investigate Issue
- Owner: Relevant service team (Support, Engineering, Billing)
- Input: Complaint details
- Action: Review customer account/service to understand the issue. Document findings with specific evidence (log entries, screenshots, etc.)
- Output: Investigation notes in ticket
- Timeline: 24 business hours

Step 3b: Check for Systemic Issues
- Owner: Customer Success team
- Input: Issue description
- Action: Search ticket system for same issue reported by other customers in past 30 days. Count occurrences.
- Decision: If 3+ customers reported this, mark as "Systemic Issue" and notify VP of Product
- Output: Systemic flag in ticket (if applicable)
- Timeline: Concurrent with Step 3

Step 4: Determine Resolution
- Owner: Service team lead or VP of Customer Success (if escalated)
- Input: Investigation findings
- Action: Determine if complaint is valid:
- VALID: Service failed, system downtime, missing feature, billing error, unintended product behavior
- INVALID: Customer misuse, failure to read documentation, feature working as designed, using product outside scope
If valid, propose solution. If invalid, prepare explanation for customer.
- Output: Resolution plan with justification
- Timeline: 24 business hours

Step 5: Determine Escalation
- Owner: Customer Success manager
- Input: Ticket with investigation and resolution plan
- Decision:
- Escalate to VP of Customer Success if: (a) resolution will take >5 business days, (b) customer has complained about same issue before, (c) customer is high-value account (>$10K MRR)
- Output: Escalation flag and VP assignment (if applicable)
- Timeline: Same day as Step 4 complete

Step 6: Contact Customer
- Owner: Customer Success team or VP (if escalated)
- Input: Resolution plan
- Action: Call customer (preferred if urgent) or email. Explain issue, acknowledge impact, share resolution plan. If customer is upset, offer goodwill credit/extension per Appendix A guidelines.
- Output: Customer communication logged in ticket
- Timeline: Within 2 business days of investigation complete

Step 7: Implement Solution
- Owner: Relevant service team
- Input: Agreed-upon resolution
- Action: Implement the fix, refund, credit, or replacement. Document what was done and when.
- Output: Solution implemented, documented in ticket with completion timestamp
- Timeline: 2-5 business days depending on solution type

Step 8: Verify Resolution
- Owner: Customer Success team
- Input: Follow-up with customer
- Action: Contact customer to confirm issue is resolved and they're satisfied. Document their response.
- Output: Resolution confirmation (or new complaint if not resolved)
- Timeline: 3-5 business days after solution implemented

Step 9: Close Ticket
- Owner: Customer Success team
- Input: Resolution confirmed
- Action: Update ticket status to "Closed". Archive documentation.
- Output: Ticket closed in system
- Timeline: Same day as confirmation

Now it's much better. It has decision logic. It has escalation. It captures systemic issues. Is it perfect? Not yet. You'd still need to verify:

• Is the 24-hour investigation timeline realistic for your team?
  - Are the "valid/invalid" definitions actually correct for your product?
  - Are the goodwill guidelines in Appendix A clear and sufficient?
  - Will the VP of Customer Success be able to handle the escalations?

But this is now a working draft. You can pilot it with the team, get feedback, refine again.

The SOP verification loop

After you get the AI draft: (1) Read it through. (2) Ask your team: Does this match how we actually work? (3) Note gaps. (4) Ask AI to address gaps. (5) Pilot with small group. (6) Refine based on pilot feedback. (7) Roll out. That's the whole cycle.

Updating Existing SOPs with AI

You probably have SOPs that are outdated. Maybe they're 5 years old. Maybe the process changed and the SOP didn't. AI can help you update them efficiently.

The update prompt structure:

Role: You are a process improvement analyst. You update SOPs to reflect current practices.

Context:
- Here's our current SOP for [process]: [PASTE THE OLD SOP]
- What's changed: [Describe what's different now]
- New tools/requirements: [What's new in your environment]
- What should stay the same: [What still works]

Task: Update this SOP to reflect current practice. Don't remove sections that still work. Add new steps where practice has changed. Update timelines if they've changed.

Format: [Same as original SOP]

Constraints:
- Keep the original structure and tone
- Flag any steps that seem outdated but I should verify (don't assume)
- Don't remove detail, only add what's changed

Real example of updating an SOP:

Role: You are a process improvement analyst. You update SOPs to reflect current practices.

Context:
Here's our current SOP for Candidate Placement:

CANDIDATE PLACEMENT SOP (Last updated 2021)
Step 1: Receive job request from client
Step 2: Search resume database for matches
Step 3: Phone screen candidate if interested
Step 4: Submit resume to client
Step 5: Client schedules interview
Step 6: Candidate interviews
Step 7: Offer is made
Step 8: Placement confirmed

What's changed:
- We now use a new applicant tracking system (Lever) instead of just the resume database
- We added a video interview step before phone screening to save time
- We implemented a candidate feedback survey after rejection
- The process used to take 4 weeks; now it should take 2 weeks
- We have new compliance requirements for background check documentation

New tools/requirements:
- Lever (ATS)
- Video interview tool (Wistia)
- Background check vendor (Checkr)
- Compliance requirement: background clearance before offer can be made

What should stay the same:
- Client relationship approach (we're consultative, not transactional)
- Quality bar (we're selective about who we place)
- Phone screening is still the primary candidate vetting method

Task: Update the SOP to reflect current practice. Include the video interview step, the new systems, the background check requirement, and tighten the timeline.

Format: Keep the original format but update steps, owners, and timelines. Add new steps where needed.

Constraints:
- Keep it to 8-10 steps (don't add busywork)
- Flag any compliance aspects that need legal review

What AI would generate: A revised SOP that includes Lever, the video interview, background checks, and compressed timelines. Then you review it, make sure the timelines are realistic, and verify compliance aspects with your legal person.

The SOP Verification Checklist

After you get an AI-generated SOP, before you roll it out, run this checklist:

Completeness:

• Does every step have an owner (a specific role/person)?
  - Does every step have a clear input and output?
  - Are decision points included where the process branches?
  - Are exceptions/error cases handled or at least mentioned?
  - Is there an escalation path for problems?

Accuracy:

• Did you walk through this with the person/team who actually does the work? (Not the ideal version, the actual version?)
  - Are the timelines realistic for your organization?
  - Are the tools/systems named correctly?
  - Are compliance or approval steps included if relevant?

Usability:

• Is the language clear enough that someone new could follow it?
  - Are there undefined terms or jargon that needs glossary entries?
  - Is the format appropriate for your use (printed handbook vs. wiki vs. training)?
  - Are there appendices or references needed?

Risk:

• Are there compliance or safety-critical steps that need audit/legal review?
  - Could someone following this SOP exactly create a problem for the organization?
  - Is there a step where judgment is required but the SOP doesn't explain how to make that judgment?

Use this checklist before you share the SOP with your team. If you're getting "no" answers, fix those things before rollout.

Try This Now: The SOP Draft-Refine Cycle

Let's walk through a complete example where I show the prompt, the output, the issues found, and the refinement.

Scenario: You're the Operations Manager at a 30-person recruiting firm. You need an SOP for candidate offer negotiations. Currently, the process is: Recruiter talks to candidate, gets general sense of expectations, then calls hiring manager, they discuss salary range, recruiter calls candidate back with offer. But there are frequent miscommunications and the process stalls. You want to systematize it.

Your prompt:

Role: You are a recruiting operations specialist with experience in offer negotiation processes.

Context:
- Organization: 30-person recruiting firm. We place mid-level professionals (program managers, marketing managers, operations analysts) with corporate clients.
- Current process: Recruiter gets sense of candidate salary expectations. Recruiter calls hiring manager to discuss budget. Recruiter calls candidate back with offer. Sometimes offer is rejected because expectations weren't clear. Sometimes there's back-and-forth that takes days.
- Target process: We want to structure the negotiation so it's clear upfront what's negotiable, what's not, and we avoid surprises.
- Who's involved: Recruiter (owns process), Hiring Manager (makes the decision), Candidate
- Current tools: Email, phone, spreadsheet with standard salary bands by role

Task: Create an SOP for candidate offer negotiation. Start from the moment the candidate passes final interview until the offer is accepted or rejected.

Format: Numbered steps with Owner | Input | Action | Output | Timeline format. Target 8-10 steps. Include a decision tree for common negotiation scenarios.

Constraints:
- Assume candidates are at executive level (salary expectations clear but maybe negotiable)
- Assume hiring managers have a set budget range they're working from
- Don't assume we have legal team. Keep it simple
- All communication should be documented in our email system
- Timeline should not exceed 5 business days from "passed interview" to "offer accepted or rejected"

AI output (first pass):

CANDIDATE OFFER NEGOTIATION SOP

Purpose: Standardize the offer negotiation process to increase acceptance rates and reduce back-and-forth delays.

Step 1: Confirm Candidate Interest
- Owner: Recruiter
- Input: Candidate passed final interview
- Action: Call candidate same day. Confirm interest in role and ask about salary expectations/must-haves
- Output: Notes on candidate expectations documented in email to recruiter and hiring manager
- Timeline: Same day as final interview

Step 2: Gather Hiring Manager Input
- Owner: Recruiting Manager
- Input: Candidate expectations from Step 1
- Action: Email hiring manager asking: What's your budget? What's non-negotiable (title, start date, benefits)? What's flexible?
- Output: Budget and negotiation parameters documented
- Timeline: 1 business day

Step 3: Determine Offer
- Owner: Recruiting Manager and Hiring Manager
- Input: Candidate expectations + Manager budget
- Action: Compare expectations to budget. If within range, prepare offer. If not, schedule 15-min call to discuss how much flexibility exists.
- Output: Offer package agreed upon (salary, title, start date, benefits)
- Timeline: 1-2 business days

Step 4: Present Offer
- Owner: Recruiter
- Input: Approved offer package
- Action: Call candidate (phone preferred). Present offer. Explain each component. Ask for feedback. Let them know decision timeline (usually 24 hours).
- Output: Offer verbally presented, candidate response documented
- Timeline: Same day as offer is determined

Step 5: Candidate Response
- Owner: Candidate
- Input: Offer presented
- Action: Candidate reviews offer and responds
- Output: Candidate accepts, counter-offers, or declines
- Timeline: Up to 48 hours (per conversation)

Step 6a (IF ACCEPTED): Send Offer Letter
- Owner: Recruiter
- Input: Acceptance from candidate
- Action: Prepare offer letter in writing and send via email. Include all terms discussed. Ask for signed confirmation.
- Output: Written offer letter sent
- Timeline: Same day as acceptance

Step 6b (IF COUNTER-OFFER): Negotiate
- Owner: Recruiter and Hiring Manager
- Input: Counter-offer from candidate
- Action: Call hiring manager immediately. Discuss if counter is acceptable or if we can counter-counter. Respond to candidate within 24 hours.
- Output: Negotiation outcome (accept counter, modify, decline)
- Timeline: 1 business day max

Step 7: Offer Closure
- Owner: Recruiter
- Input: Signed offer letter or negotiation resolution
- Action: Confirm candidate start date with hiring manager. Update your CRM. Notify HR to begin background check and onboarding prep.
- Output: Placement confirmed in system
- Timeline: 1 business day

Your review: Good start. But missing:

• What if the hiring manager's budget doesn't match candidate expectations? The decision point in Step 3 is vague.
  - What if the candidate counter-offers on something that's non-negotiable (title, start date)? How do you handle that?
  - No mention of benefits negotiation. We have set benefits, are those negotiable?
  - What if the candidate accepts verbally but then asks for more money after the offer letter is sent?

Your refinement prompt:

Good framework. Three additions needed:

1. In Step 3, clarify what happens when expectations don't match budget. Add a decision tree:
   - If candidate wants $150K and budget is $140K: We typically go $145K as compromise
   - If difference is >$10K: Schedule call between recruiter, manager, and candidate to find middle ground
   - If difference is >$20K: Likely a bad fit, discuss with manager if we even move forward
2. In Step 6b, add guidance for non-negotiables: "If candidate counters on title or start date (marked non-negotiable), explain: 'We can't adjust this, but we can adjust [salary/other element].'"
3. Add a note about benefits: "Benefits are not negotiable. They're standard across the firm. Explain this upfront in Step 1."
4. Add a new Step 8: "If candidate asks for additional concessions after offer letter is sent, respond same day that all terms were agreed and documented. Don't reopen negotiation."

Updated SOP (refined version): [I'll skip the full output for brevity, but it would include all the additions you requested, structured clearly with decision criteria and non-negotiable boundaries.]

That's the cycle: Draft → Review → Refine → Verify → Pilot → Roll out.

What to Do Monday Morning

• Pick one process that needs an SOP. Write a detailed prompt using the template in this lesson (role, context, task, format, constraints).
  - Send the prompt to AI and get a draft. Don't iterate yet, just get the first pass.
  - Walk through the draft with the person/team who actually does this work. Ask: "Is this accurate? What's missing? What's wrong?"
  - Make a list of refinements. Be specific: "Step 4 doesn't mention what happens if X." "Timeline for Step 5 is too short."
  - Send refinement requests to AI. Iterate 1-2 more times until the SOP is structurally sound.
  - Run it through the verification checklist before rollout. Check completeness, accuracy, usability, risk.

Key Takeaways

• AI drafts SOPs quickly. The structure, formatting, and general flow are AI's strength. Use that to save time.
  - Your job is verification and context. You verify accuracy, completeness, and compliance. You add context AI doesn't have.
  - A detailed prompt saves iteration. Spend 15 minutes writing a comprehensive prompt. Get 80% of the way there in the first draft.
  - Walk it with the people who do the work. They'll catch accuracy issues immediately. That's not AI's job; it's theirs.
  - Iterate on structure, not wording. If the logic is wrong, ask AI to fix it. If the wording is awkward, fix it yourself.
  - Decision points and exception handling are critical. AI often misses these. Watch for them in your verification.
  - Compliance and safety steps need review. Before you roll out an SOP, have relevant experts (legal, compliance, safety) review if needed.

FAQs

How detailed should my prompt be?

Detailed enough that a new hire could read it and understand your situation. If you're explaining verbally after they read the prompt, add more detail. Aim for 200-300 words of context.

Should I include the old SOP when updating?

Yes, paste it into the context. AI can see what changed more easily if it has the old version. Just tell it: "Here's what we had, here's what changed, update it."

How many iterations is normal?

Usually 1-2 refinement rounds. If you're iterating more than 3 times, the AI might not understand your situation well enough. At that point, build it yourself using AI as a brainstorming tool.

Should I share the AI-generated SOP with my team as-is?

No. Walk it with them first. Get feedback. Verify it's accurate. Then refine. Then share. Never ask people to follow an SOP you haven't verified with them.

What if my team says the AI-generated SOP is too rigid?

SOPs should be guidelines, not rules. Add language: "Follow this process unless circumstances require adjustment. If you deviate, document why and notify your manager." That gives flexibility without removing structure.