Write it down before you automate it: SOPs that make AI workWrite it down before you automate it: SOPs that make AI work

AI gives its best results in companies that have written down how they work. Before any agent, chatbot or automation goes live, the processes it will touch need to exist on paper, with an owner, a goal and clear steps. This guide walks through a seven-step method for building that foundation, from your first KPI review to a shared company brain your team and your AI both read.

Why AI tools so often stall

Many teams recognize this cycle. The company buys an AI tool. The team tries it once or twice. Work carries on the old way, and the subscription quietly renews until someone buys the next tool.

The usual explanation is "my team doesn't like change" or "the AI just isn't good enough." Both miss the real cause. When a tool lands in a company where processes live only in people's heads, it has nothing solid to work from. It guesses, the guesses are wrong, and people stop trusting it.

Two principles help here:

• AI enhances people's work. It makes a good process faster and more consistent. It doesn't decide what the process should be.
• AI multiplies what it finds. Clear processes become faster. Unclear processes become unclear at higher speed.

So the first job is making your processes clear. The tool for that is the standard operating procedure.

What an SOP is, and what's new about it

A standard operating procedure (SOP) is a documented set of step-by-step instructions that helps workers perform routine or complex tasks consistently, safely and efficiently.

That classic definition needs one addition today. An SOP now serves two readers: your people and your AI. The same page that trains a new hire also tells an agent what to do, in what order, and when to stop and ask a human.

Writing for both readers changes how you write. Plain structure, one owner per task, explicit conditions and checkable results suit a person skimming on a Monday morning. They also suit a model reading the page as instructions.

The method runs in seven steps. Steps 0 to 3 are entirely human work. AI enters only at step 4, once there is something clear for it to work with.

1. Step 0: Map your KPIs and read the data
2. Step 1: Sort processes into acquisition, delivery and support
3. Step 2: Map each process as a flow and find the friction
4. Step 3: Write the SOPs
5. Step 4: Bring in AI as a reviewer
6. Step 5: Work with AI as an assistant
7. Step 6: Connect everything in a company brain

Step 0: Map your KPIs and read the data

Start with numbers, because they show where the real bottlenecks are. Opinions about what's slow are useful hypotheses; the data confirms or corrects them.

List the key performance indicators for each part of the business, following the customer's journey from first contact to repeat purchase. Typical examples:

• Marketing: visitors, leads per month, cost per lead
• Sales: lead-to-call rate, close rate, time from inquiry to signed deal
• Delivery: time to onboard, projects delivered on time, upsell rate
• Support: first response time, tickets per customer, churn rate

Then look at the trends. Where do leads drop off? Where does work wait? Which step generates complaints? The answers tell you which process to document first and where AI will pay back fastest.

Step 1: Sort every process into three categories

Every business, whatever it sells, runs on three groups of processes. Sorting your work into them gives each process a clear purpose and a clear measure of success.

• Acquisition answers "How do you find and sign new customers?" Its goal is a growing customer base. Marketing and sales usually own it.
• Delivery answers "How do you deliver your product or service?" Its goal is keeping customers, growing revenue per customer and earning referrals. Product and operations usually own it.
• Support answers "How do you handle questions and issues after the sale?" Its goal is keeping churn low. Customer service usually owns it.

Most companies pour their energy into acquisition. Delivery deserves equal attention. A customer who is delighted with what they received buys again, upgrades and recommends you, and that revenue usually costs far less than a new lead.

Step 2: Map each process as a flow

Draw how work actually moves today, one diagram per category: acquisition, delivery, support. Boxes for steps, diamonds for decisions, arrows for hand-offs. A whiteboard or a simple flowchart tool is enough.

A visual flow makes two kinds of friction easy to spot:

• Time sinks: steps that need a lot of manual effort, waiting or copying between tools.
• Quality risks: steps where mistakes happen often, such as data typed twice, approvals that get skipped, or hand-offs where context gets lost.

Mark each one on the map. Time sinks are your best candidates for automation. Quality risks are where a written checklist, and later an AI check, will protect you most.

Step 3: Write the SOPs

With the flow mapped, turn it into text. A good SOP answers three questions.

• Who: one named person or role is responsible. Never "the team" or "marketing." When a task belongs to a group, everyone assumes a colleague has it.
• Why: one sentence describing what "done" achieves. If that sentence is true at the end, the procedure worked.
• How: the steps, as numbered lists and checklists.

The document structure

Pairing the text with a flow diagram helps both readers: people grasp the diagram at a glance, while AI reads the text precisely. A complete SOP is built top-down:

1. Title: the name of the procedure.
2. Purpose: what the document is for, in one or two sentences.
3. Before starting:

• • Contacts: everyone involved, including freelancers and outside agencies, so anyone can reach them in one click.
  • Tools: the apps, files and accounts the work needs.

4. Procedure: the work itself, split into phases.
5. Metrics and success criteria (optional): the KPIs from step 0 this procedure should move, so you can see whether it works.

The phase structure

Each phase follows the same pattern:

• Title, verb first: "Send the proposal," not "Proposal sending." Readers know immediately what to do.
• Responsible: one person or role, who has the final word.
• People involved: everyone else taking part.
• Goal: one sentence on what this phase achieves.
• Prerequisites: a green light. When every condition is met, the phase can start. This keeps people from fixing problems that belong to an earlier phase or a different SOP.
• Emergency procedure: the condition that means "stop here" and the single person to contact. One name, never a department, so a problem never waits for someone to claim it.
• Steps: numbered, one action per step, verb first.
• Quality check: a short yes/no list. Every answer must be yes before moving on; a single no means going back to fix it.
• Q&A: the questions every new hire asks, answered once. It also spares the colleague on holiday a phone call.

A worked example

Here is one phase from a small real estate agency's sales SOP, covering a property viewing and its follow-up.

Phase 2: Show the property and log the follow-up

• Responsible: Ana (sales lead)
• People involved: Marko (sales agent)
• Goal: every viewing ends with notes and a follow-up date in the CRM the same day.
• Prerequisites: the viewing is confirmed in the shared calendar, and the client's contact details are in the CRM.
• Emergency procedure: stop when the owner cancels access or the client reports a complaint. Contact Ana.

Steps:

1. Message the client on WhatsApp one hour before the viewing.
2. Meet the client in front of the property and show it.
3. Back at the office, add viewing notes to the client's CRM record.
4. Set the follow-up date in the CRM before closing the record.

Quality check:

• Was the reminder sent one hour before?
• Are the viewing notes in the CRM?
• Is a follow-up date set?

Q&A:

• The client didn't show up. What now? Log "no-show" in the CRM and set a follow-up for the next working day.

Notice step 4 and the third quality check. The team's most common mistake was forgetting the follow-up date, so the SOP builds that check in rather than relying on memory.

Step 4: Bring in AI as a reviewer

Once the human work is done, AI's first job is quality control. Write your house standard for SOPs once, then let AI check every new or edited procedure against it. The review returns three lists: what to fix, what to improve and what is already fine.

A person can do this review too, but every person reviews differently. AI applies the same standard every time, across every department. The result is a library where every SOP has the same shape, so anyone opening an unfamiliar procedure knows exactly where to look.

Step 5: Work with AI as an assistant

Next, give your team an AI assistant that knows the SOP library. It can:

• Write new SOPs in your house format, interviewing the owner about steps, tools and common mistakes.
• Edit existing SOPs when a process changes, preparing the change for a person to approve.
• Answer questions like "How do I issue a refund?" in seconds, with a link to the source procedure.

That last capability changes daily work. People stop interrupting colleagues or digging through shared drives, and they can check the answer against the linked source. The repetitive writing and searching moves to the machine, and people spend their time on judgment, relationships and creative work.

Step 6: Connect everything in a company brain

The final step brings all procedures and company knowledge into one place, so your team and your AI read the same page. Richer context means better answers: good input gives good output.

A useful company brain is simple, with three parts:

• Knowledge base: the procedures and reference material, organized in sections.
• Human actions: uploading new documents, reviewing pages a colleague or the AI drafted, and acting on review dates.
• Chat: the assistant from step 5, working from the knowledge base.

A typical knowledge base holds:

• Company: what you do and sell, your ideal customers, value proposition, positioning and team contacts
• Brand: colors, fonts, tone of voice and the keywords you use
• Departments: administration (invoicing, billing, taxes), sales, delivery and support, each with its SOPs
• Restricted areas: sections only certain roles can open, such as a founders-only space

Access layers matter. Not everything belongs in front of everyone, and a good company brain lets you set who sees what by role.

Practical advice for getting started

Start with quick wins

Resist building the whole system at once. Put candidate processes on an impact-effort matrix and begin with the high-impact, low-effort ones. Larger bets can follow once the first results are in.

Small wins also bring your team along. People worry that AI threatens their jobs, and headlines rarely help. Ask the team where the friction is, fix that problem first, and let them feel the improvement in their own day. Adoption follows experience far more reliably than announcements.

Stay independent of any single model

Keep your knowledge in formats you own, such as plain Markdown files on your own storage, so no provider holds it hostage. Think of the model as the mind and the application around it, often called the harness, as the body. Choose a harness that can switch between models, so a price increase or a better competitor is a configuration change rather than a migration. Local models are an increasingly practical option for sensitive data.

Already recorded your processes on video?

Videos are an excellent starting point, especially for onboarding. Transcribe them and give the transcript to an AI assistant with your SOP template. You'll have a structured first draft in minutes, ready for the process owner to check.

The foundation comes first

The companies getting real value from AI rarely have the most expensive tools. They have written down how they work. Steps 0 to 3 are human work: measure, sort, map and write. Steps 4 to 6 let AI review, assist and connect, building on the clarity your team has already created.

Pick one process this week. Name its owner, write its goal in one sentence, and list its steps. That single page is the first piece of the foundation everything else stands on.

At AIgenica, documenting a client's processes is where every automation project begins. If you'd like help mapping your first processes and finding where AI will pay back fastest, tell our agent about your business and we'll come back with a tailored proposal.