The answer
Partly. AI can turn a recording of someone doing the job into a usable first draft, add screenshots and answer staff questions from the library. In our model about 60% of SOP time can go to software. What stays human is doing the process well on camera and checking the draft with the person who owns the job.
The tasks, and how much of each is automatable today
Six tasks make up SOP documentation. Drafting from a recording and formatting are 75 to 80% automatable. Finding outdated SOPs is about 60% automatable. Answering staff questions is about half automatable with an AI assistant over the library. Recording and reviewing for accuracy stay largely human, because the SOP is only as good as the process shown.
| Task | Min / week | Automatable |
|---|---|---|
| Capture the process being doneScreen recorders capture every click automatically. Someone still has to do the job properly and explain the why. | 60 | |
| Write the first draftAI turns a narrated recording into clear numbered steps. It cannot add the judgement calls nobody said out loud. | 180 | |
| Add screenshots and format the documentScreenshots can be pulled from the recording at each step. Cropping and redacting customer data still needs a check. | 60 | |
| Review the draft with the person who owns the jobThis is where wrong steps and missing exceptions get caught. Do not skip it to save time. | 60 | |
| Find SOPs that are out of dateTagging SOPs by tool means a tool change can flag affected documents. Process changes that happen quietly still need a person to notice. | 45 | |
| Answer staff questions the SOP should answerAn AI assistant over the SOP library answers 'how do I' questions. Questions about exceptions still go to the ops manager. | 90 | |
| Whole process | 495 |
Hours saved per week
For a 20-person service business writing or updating about two SOPs a week and fielding process questions, SOP work takes around 8.5 hours a week by hand. Automating the tasks above returns about 5 hours a week. The rest is recording and reviewing, which is where an SOP gets its accuracy.
Modelled on: A 20-person service business maintaining around 30 SOPs, writing or updating about two a week, with an operations manager doing it alongside their main job.
- Two SOPs a week, each needing about 30 minutes of recording or walkthrough.
- About 90 minutes per SOP to write a clean draft by hand, plus 30 minutes for screenshots and layout.
- About 30 minutes per SOP for the process owner to review.
- About 90 minutes a week answering staff questions that an SOP should have answered.
Tools that do the job
You need a way to capture the process, an AI step that writes structured steps from the capture, and a library people actually search. Record a screen walkthrough, transcribe it, and have an AI agent draft the SOP into Notion. n8n or Make can flag SOPs that mention a tool whenever that tool changes.
- AI agent
- Drafts SOPs from transcripts into a fixed template and answers staff questions from the library.
- Notion
- The SOP library, with owner, last-reviewed date and tools-touched tags on each page.
- Fireflies
- Transcribes walkthrough calls where the expert explains the process live.
- n8n
- Creates review tasks when an SOP passes its review date or a tagged tool changes.
- Slack
- Where staff ask the AI assistant process questions.
A simple workflow
The workflow starts with a recording of the person who does the task best, narrating as they go. The transcript and screenshots go to an AI step that drafts the SOP in a fixed template. The process owner reviews and approves. The SOP is tagged with every tool it touches, so tool changes trigger a review.
- CaptureThe best person at the task records themselves doing it, narrating the why as well as the what.
- DraftAI writes purpose, trigger, steps, exceptions and owner into the standard template, with screenshots.
- ReviewThe process owner corrects the draft and approves it. Only approved SOPs are published.
- TagEach SOP gets an owner, a review date and the tools it touches.
- MaintainReview dates and tool changes create review tasks for the owner automatically.
Keep these human
- Showing the process. The person recorded sets the standard, so pick the best one, not the available one.
- Approval by the process owner. An AI-written SOP nobody checked spreads mistakes faster than no SOP.
Where it goes wrong
- SOPs written once and never updated. Without owners and review dates the library goes stale within months. Put both on every page.
- Recording the wrong person. Capturing someone's workaround turns a bad habit into the official process.
- A library nobody searches. If staff ask colleagues instead, put the AI assistant where questions are asked, usually Slack.
What most guides miss
Most guides focus on writing SOPs faster. Writing is not where SOPs fail. They fail by going stale when a tool or step changes. Tag each SOP with every tool it touches and its owner, then trigger a review whenever that tool changes. That turns maintenance from a yearly clean-up into a small, routine task.
The evidence
- 62% of people say they struggle with too much time spent searching for information at work.Microsoft, Work Trend Index 2023: Will AI Fix Work? (2023)
- Applying expertise has a technical automation potential of only 18%, which is why reviewing an SOP stays human.McKinsey Global Institute, A future that works: Automation, employment, and productivity (2017)
- Generative AI raised McKinsey's estimate of how much data processing can be automated from 73% to about 90%.McKinsey & Company, The economic potential of generative AI: The next productivity frontier (2023)
“Automation applied to an inefficient operation will magnify the inefficiency.”