"The employee who handled accounting on their own fell ill and had to take an extended leave. Nobody knew how the monthly close was being run." We received this consultation from the administrative department of a wholesale business. The issue was that because the employee was so capable, everyone else never needed to learn the operational details. Work that only one person can perform does not look like a problem while things run smoothly. It is only when operations stop that a company realizes how heavily it depended on that single point of failure.
Everyone knows standard operating procedures and manuals should be prepared. Yet progress stalls because those capable of writing them are busy, lack the time, and continually push the task aside. This is where generative AI helps. However, claiming "AI makes creation faster" is only half the story. Real progress in eliminating person-dependent silos requires a complete operational system: defining what to document, who refines it, and how updates are sustained. This article organizes the process from the perspective of an operational buyer.
When "We Can't Operate If That Person Leaves" Catches Up with a Company
Person-dependent work is not inherently evil. Having an experienced individual run tasks alone is often faster, and the overhead of documenting everything can outweigh the benefits. The problem arises when this state becomes entrenched in specific operations and no backup exists.
Specifically, the risk surfaces in three ways. First is the risk of operational disruption. When a team member departs due to resignation, unexpected medical leave, or overlapping busy periods, that specific workflow comes to a complete standstill. Second is training costs. If procedures exist only in someone's head, handovers require dedicated, hands-on tutoring, delaying how quickly new hires become independent. Third is inconsistent quality. When methods vary by person and decision criteria are not shared, deliverables change whenever assignments shift.
In SMBs, where individuals routinely wear multiple hats, these impacts are felt acutely. That is why the process must begin by identifying which operations would cause the most damage if halted. Attempting to document everything all at once inevitably fails, making prioritization the essential first move.
What to Document: Not Everything
The biggest reason manual creation efforts stall is taking on too broad a scope with the mindset of "documenting every single workflow." The operations that truly warrant documentation meet specific criteria:
- Concentrated on a single owner: Workflows where only one person knows the steps
- High impact if halted: Tasks like invoicing, payments, or order processing where delays affect clients or cash flow
- Recurring: Operations run regularly (monthly, weekly, etc.) with generally fixed procedures
Conversely, low-frequency tasks with shifting conditions or work where situational judgment is paramount will see documentation quickly become obsolete, making the effort counterproductive. This filtering follows the same logic as prioritizing which workflows to systematize first. For a framework on auditing overall operations to determine where to start, refer to our guide on how to decide which operations to automate first.
Start by selecting just a few tasks that meet these criteria within a single department. Starting with the workflow whose disruption would cause the most harm delivers visible impact and builds internal buy-in.
Dividing Labor: AI Builds Drafts, Humans Refine
Once target operations are selected, manual creation begins. Here, generative AI's role is not writing a finished product from scratch, but rather converting fragmented notes into a structured draft.
Specifically, you feed AI "scattered information or tacit knowledge"—transcripts of verbal explanations, rough notes, or system operation logs—and instruct it to structure them into an SOP format. Even from bullet points or voice memos, structured drafts emerge in minutes. This relieves domain practitioners from the burden of writing from scratch, allowing them to focus on verification and refinement.
However, an indispensable premise must be kept in mind: AI does not know your company's unique internal rules or exceptions. Tacit rules such as "this client has a different closing date" or "this action requires managerial approval" will naturally be missing from AI drafts. That is why subject-matter experts must review the draft and add exceptions, unique policies, and warnings. Viewing this division of labor as "AI doing 80% and humans doing the final 20%" dramatically lowers the psychological barrier.
External case studies report that by having veteran employees dictate key points into voice memos and using AI to convert them into SOPs, the time needed to draft a manual dropped from half a day to approximately 1.5 hours, enabling the production of 15 manuals in three months (SME AI Training and Education Institute). Multiple reports also cite up to a 70% reduction in manual creation hours. While these are third-party examples whose results vary based on workflow complexity and raw material quality, the structural takeaway matters more than the exact numbers: the model shifts from humans writing everything to AI drafting and humans refining.
Note that since company data is fed into AI tools, internal guidelines regarding what data can be entered into which tools must be decided beforehand. Establishing boundaries based on our guide to creating AI usage policies for SMBs ensures safe execution.
Don't Do It All at Once: Build a Template Before Scaling Horizontally
Once a single department completes a few SOPs, the next step is not immediate company-wide rollout. It is creating reusable templates and frameworks. If you skip this step and let each department create manuals on their own, you will end up with an unmanageable sprawl of inconsistent documents that go unused.
This framework consists of two core elements. The first is a template—a standardized layout defining fields such as purpose, target audience, step-by-step procedures, exceptions/cautions, and revision date. The second is a prompt—a standard instruction template such as: "Based on these operation logs and bullet points, generate an SOP draft following this template." Preserving successful prompts used during early iterations creates reusable assets for horizontal expansion. For prompt construction principles, see our guide on prompting fundamentals for business.
Phasing the rollout makes progress much easier to manage.
| Stage | Estimated duration | Scope |
|---|---|---|
| Launch | First month | Create a few high-priority SOPs in one department, solidifying templates and prompt formats |
| Retention | Next 1–2 months | Expand volume within the same department and test update maintenance workflows |
| Horizontal rollout | Subsequent period | Distribute established frameworks to other departments so each team can operate autonomously |
Rather than launching a company-wide mandate from day one, build templates and prove value in one department before expanding. This sequence is key to avoiding mid-project stalling.
Don't Stop at Creation: Updates and Evolving into Q&A
An SOP begins aging the moment it is created. If operational procedures change and documentation fails to follow, employees will find that following the manual leads to mistakes, and they will stop consulting it altogether. Ultimately, overcoming person-dependent workflows depends less on initial creation and far more on whether the update cycle functions.
Sustaining updates requires systematization. Establish operational rules during the initial adoption phase: clearly state revision dates and owners on each manual, incorporate updating documentation directly into workflow changes, and run periodic audits to refresh outdated material. Relying on "whoever notices should fix it" never lasts; designating who reviews what and when is crucial.
Furthermore, well-structured manuals can serve as knowledge bases for search and internal Q&A. Recently, organizations are loading internal SOPs into AI notebook tools (like Gemini Notebook, formerly NotebookLM) to answer employee questions such as "Who approves this transaction?" complete with citations referencing the exact source manual. Because citations are displayed, users can cross-check source material even if an AI response is imprecise, substantially easing internal inquiry workloads. At this stage, manuals shift from mere onboarding handoffs into an accessible company-wide knowledge base.
As context, the AI adoption rate among Japanese SMBs stands at 20.4%, with 87.0% of adopting companies citing "operational efficiency / reducing work hours" as their goal (Organization for Small & Medium Enterprises and Regional Innovation, March 2026 Survey). While non-adopters remain the majority, those taking action focus squarely on operational efficiency. Resolving person-dependent silos is one area where efficiency gains are most immediate.
Creating SOPs is not technically difficult once underway. The real challenge lies in structural design: selecting the right targets, dividing labor, and maintaining continuous update cycles. If you are looking to clarify which operations to document first, embed AI drafting workflows into your organization, or build an internal Q&A knowledge base from finished SOPs, please reach out to GleamHub's Development, AI & Automation consultation desk. We will audit your operations and help design a roadmap starting with mission-critical workflows. Contact us via our inquiry page (custom estimates provided based on scope).
Sources
- Survey on the Utilization of AI and Related Technologies in SMEs (March 2026) | Organization for Small & Medium Enterprises and Regional Innovation
- Streamlining SME Business Manual Creation with AI: Practical Steps to Eliminate Person-Dependency and Cut Creation Hours by 70% | SME AI Training and Education Institute
- How to Create SOPs and Manuals with AI | Kaminashi
- How to Eliminate Person-Dependent Workflows with Generative AI: Case Studies, Benefits, and Key Considerations | Smart Style TECH BLOG
