6 min read
SOPs that keep up with how we actually work
When we change how we work, a Claude skill turns the meeting into a draft SOP in seconds. Our SOPs stay current, and we can find gaps, missing handoffs and every process a tool change would affect.
When we change how we work at Growth Advisory, we talk it through in a meeting first. Since this summer, we’ve been using AI to turn the transcript of that meeting into the first draft of the new or revised standard operating procedure (SOP). Specifically, a Claude skill takes the information from the meeting and writes it into our SOP template and reports back questions for everything the meeting left open. This blog post describes what goes into one of our SOPs, how the AI skill writes it, and the questions we can now ask across all SOPs. We close with lessons learned from working with SOPs saved in Google Docs and how we plan to improve our SOP management in the future.
What goes into one of our SOPs
An SOP describes one process so that anyone trained can run it and get the same result. A founder working alone may not need one: they deliver the whole service themselves and hold the standard in their head. Once a second person delivers part of it, the standard has to be written down. Then the SOPs show what each new hire has to learn and how to check the work. A clear benefit for the company is that it keeps running when someone leaves and the owner can step back from the day-to-day.
At Growth Advisory, we teach our clients to write their core processes this way too. Our test: write an SOP for someone who knows 80% less than you do. For this, the SOP template follows a consistent format: a short header, the context a new person needs, such as the goal and who does the work, and then the process as numbered steps, starting with a trigger such as “a request arrives via email.”
We make sure these things are consistent. In fact, the very first SOP we wrote was how to write an SOP.
Google Docs let our SOPs drift apart
Until this summer, we had a very simple but manual system. We used a shared Google Drive folder with a subfolder per department, and a Google Doc per SOP. We picked it because everyone already had Drive and knew how to edit a Doc, so we needed no new software. It had three problems:
- Writing a new SOP was slow. Someone had to copy the template and fill in every field by hand.
- SOPs drifted apart. Some got far more detailed than others, and people changed the template in their copy, because Google Docs doesn’t enforce the template.
- Nobody could see the set as a whole, for example which SOPs were overdue for a review or where gaps existed between processes.
With the help of AI, we could keep our simple system but manage it smarter.
An AI skill writes the SOP from the meeting
An AI skill is a set of written instructions that Claude, Anthropic’s AI assistant, loads for one kind of task. Ours holds the template field by field, the department list, the naming rules, and what our SOP training teaches. We trained the AI skill by letting the AI go through our existing SOPs and learn their rules: how much detail, how to phrase a step, and where one step ends and the next begins.
For a new or revised SOP, we now use a meeting transcript as the input the AI skill works with. We talk processes through all the time, like when we train a new colleague or when we decide to change how something works. The skill reads that meeting transcript and writes the SOP in seconds. It takes the last decision on each point, not the first, because we change our minds during meetings as we consider different aspects of a process. Anything that was only suggested becomes a question that we answer when finishing the written SOP.
We also ran the finished AI skill over the SOPs we already had to review them. It applied the template consistently and drafted new parts that had been missing in the SOPs that deviated from the template. Now all of our SOPs are updated and follow the template, as they should.
Questions we can now ask across all SOPs
With every SOP in the same shape and in one place, Claude can read all of them at once. We can ask it things like:
- Which SOPs haven’t been reviewed in the last six months?
- Which SOP contradicts another one?
- Where is a handoff missing, because one SOP ends and the next doesn’t pick up?
- What is a given role responsible for, across every SOP that names it?
- Which processes have to change if we retire a tool or replace a piece of software?
- At what point in a client engagement do we collect their phone number, in case we need to reach them?
Especially in a growing company, it becomes challenging to keep the whole system in mind while reviewing SOPs, and these questions could not be answered easily. AI doesn’t break a sweat doing this for us.
Markdown in Google Drive replaced our sync
A challenge we faced with using Claude as our AI tool was that Claude works poorly with Google Docs. However, Claude works very well with Markdown, which is plain text with a few symbols for headings and lists, so any text editor can open it. So we built a sync. It converted each Google Doc to a Markdown file on the computer running Claude Code, the version of Claude that works directly with files, and converted the result back after every change. The round trip lost formatting and broke in ways that were easy to miss.
In October, Google Drive started opening and editing Markdown files directly. We deleted the sync, about 4,000 lines of code responsible for the conversion logic. Our SOPs are now Markdown files in a shared drive and don’t need to be converted anymore. The team opens them in Google Docs as before, and Google Drive for Desktop puts the same files on the computer where Claude Code works on them. – Thank you, Google!
Comparing the SOPs with how we actually work
Potwin, the software we’re building that keeps the transcript of every meeting, adds the last piece to our ideal way of working with SOPs. We can ask Claude Code to compare what we said in recent meetings with what the SOPs say and point out where they differ. When our actual way of working deviates from the documented SOP, we decide: either we train the team on the SOP and remind them how we want them to work, or the way we work has improved and the SOP gets updated.
One recent example is from this October. We noticed that our own hiring SOP described an older process than the one we actually run. We fixed the SOP in the way described in this blog post: the skill drafts the process from the meeting where we discussed it, and the colleague who runs it confirms the draft. Our Chief Operating Officer approves the change.
Our future plan is building SOP writing and review directly into Potwin itself. Then we can part-automate the review and revision cycle on a predictable schedule. Detecting changes in how we work from meeting transcripts can then proactively alert us. Until then, we will continue using Claude Code and Google Drive, which are proving to be a good combination.