Turn your unwritten business rules into AI-ready workflows

You run an SME and most of your “procedures” live in people’s heads. New hires learn by watching colleagues, mistakes repeat themselves, and the idea of documenting everything feels exhausting. Yet this is exactly what AI and automation need to be truly useful: simple, clear, actionable rules.

This article shows how to turn your unwritten business rules into explicit, structured rules that an AI can understand and support. The goal is to make your way of working visible and shareable, without bureaucracy or technical deep dives, so you can reduce errors and open the door to practical automations tailored for SMEs.

1. Why everything living “in people’s heads” blocks AI

In most SMEs, “rules” look like this:

• Habits passed on orally
• Exceptions known only by a few people
• Old emails or documents nobody reads anymore
• Spreadsheets that act as the company’s memory

As long as everyone is there, things work “well enough”. But as soon as someone key is on leave or leaves the company, problems show up.

As long as your rules stay implicit, AI cannot really help you: it has no idea what is acceptable, risky or high priority for your business.

1.1. The very concrete risks for an SME

When your rules are not explicit, you face:

• Repeated errors (late invoices, wrong prices, wrong email template…)
• Dependence on a few people who “know how things work here”
• High mental load: everyone has to remember everything
• Difficulties automating: every case is “special”, nothing is standard

When you clarify and structure your rules, you unlock:

• AI assistants that can prepare and check the work
• Simple, reliable automations (reminders, checks, pre-filling)
• Easier onboarding for new hires

1.2. Business rules vs. procedures: you don’t need a 200‑page manual

When leaders hear “formalise your processes”, many imagine a dusty 200‑page quality manual nobody reads. That’s not what you need.

To get value from AI and automation, you mainly need:

• Short rules, focused on the most frequent situations
• Concrete examples: “if this happens, then we do that”
• Operational checklists: 5 to 10 items max

The goal is not to describe everything. The goal is to describe just enough so that humans AND AI know how to behave in 70–80% of standard situations.

2. Start small: define a single micro‑process

Before you formalise anything, narrow your scope to one micro‑process.

A micro‑process could be:

• Handling an incoming customer request by email
• Approving a commercial discount
• Answering a certificate or attestation request
• Handling a leave request
• Checking a supplier invoice before payment

2.1. How to pick your first micro‑process

Ask yourself these five questions:

1. Does this happen at least once a week?
2. Has a mistake on this topic already caused issues?
3. Do several people need to know how to do it?
4. Is there already an informal “way we usually do it”?
5. Would reminders, templates or automatic checks make this easier?

If you get at least three “yes” answers, you have a good pilot for turning your rules into something AI‑ready.

2.2. Sketch the micro‑process in 5 minutes

On a sheet of paper or a slide, simply list:

• Trigger: what starts the action? (e.g. an email “quote request”)
• Key steps: 3 to 7 steps max
• Expected outcome: how do you know it’s done?

You can draw a simple flow like this:

This doesn’t need to be perfect. It’s just a support for discussion with your team before you write down the rules.

3. Turn frontline experience into clear rules

Once you have your micro‑process, the goal is to pull the rules out of your experts’ heads.

3.1. Three simple questions to ask your team

Bring together one to three people who do this work regularly and ask them:

1. “What makes a case easy for you?”
2. “In which cases do you hesitate or ask for someone’s opinion?”
3. “What mistakes do you absolutely want to avoid?”

Capture their answers without worrying about wording yet. You are gathering:

• Decision criteria
• Standard cases vs. sensitive cases
• Frequent pitfalls

3.2. Write your first “if… then…” rules

From these notes, write down rules in plain language:

• If the request comes from an existing customer and the amount is below €2,000, then the maximum discount is 5%.
• If a new customer asks for a very short deadline, then we ask the sales director to validate.
• If the customer mentions a safety issue, then we treat the request as urgent on the same day.

You don’t need to cover every possible case. Focus on:

• The 3 to 5 most frequent situations
• The 3 to 5 riskiest situations

3.3. Separate what AI can do from what must stay human

For each step in your micro‑process, ask:

• Can AI or a tool prepare the work? (e.g. sort requests, pre‑fill an answer)
• Can a simple automation trigger a reminder or a task?
• Is this a decision that must stay fully human? (because it affects customer relationship, people, legal risk…)

You end up with three clear buckets:

• Prepared by AI
• Triggered automatically
• Decided by a human

This separation is key. It prevents over‑automation and reassures your teams that AI is an assistant, not a replacement.

4. Make your rules usable by AI and automation

Once your rules are clear, you can plug them into concrete tools without becoming a tech expert.

4.1. Build an “AI‑ready rule pack”

For each micro‑process, create a simple one‑page document that includes:

• Context: what this is about (e.g. handling simple quote requests)
• Goal: what success looks like (e.g. reply within 24 hours with the right info)
• Decision rules: your “if… then…” statements
• Examples: 2–3 typical cases with the expected response

This becomes the instruction sheet for an AI assistant (for example, a chatbot connected to your tools):

• You feed these rules into the assistant as a “playbook”
• It can then propose draft answers or checks that match the way you work

4.2. Simple automation ideas based on your rules

From your explicit rules, you can implement very simple automations, such as:

• Automatically tagging emails with words like “urgent”, “complaint”, “safety”…
• Creating response templates that follow your rules, to be reviewed by a human
• Generating a checklist for each new case (with a no‑code tool or a shared spreadsheet)
• Setting automatic reminders If an action is not completed within a given delay

In each case, you are not inventing rules for the tool. You are translating rules that already exist in people’s heads into actions tools can execute.

4.3. Test your rules with AI in “simulation mode”

Before connecting anything to live systems, use AI in simulation mode:

1. Give it your rules and a few past (anonymised) examples.
2. Ask: “What would you do in this case?”
3. Compare its suggestions to what your team actually did.

You will quickly see:

• Where your rules are clear and robust
• Where details are missing
• Where you need additional safeguards or escalation rules

This lets you improve your rules before you let AI or automation touch real customers or data.

5. Practical section: 7‑day framework to formalise one micro‑process

Here is a simple, concrete roadmap you can apply this week.

Day 1: pick the micro‑process

• List 3 to 5 recurring activities that generate errors or frustration.
• Choose one as a pilot using the five questions above.

Day 2: map it in 5–7 steps

• Define the trigger, key steps and expected outcome.
• Draw a very simple flow (on paper or in a slide) and share it with the people involved.

Day 3: interview your experts

• Bring together one to three people who know this work well.
• Ask them the three questions: easy cases, hesitation cases, mistakes to avoid.
• Capture everything, even if it feels messy.

Day 4: write 5–10 “if… then…” rules

• Group your notes by theme.
• Write down 5 to 10 rules at most, separating frequent vs. sensitive cases.

Day 5: classify human vs. AI vs. automation

For each step, mark whether it can be:

• Prepared by AI
• Triggered automatically (reminder, task, template)
• Decided only by a human

Day 6: build your one‑page rule pack

• Create a single page with: context, goal, rules, examples.
• Review it with your experts: “Does this reflect how you actually work?”

Day 7: run a simulation with an AI assistant

• Use any mainstream AI assistant with your rules as context.
• Feed a few anonymised past cases and check its suggestions.
• Adjust your rules where needed before you even think about full automation.

In one week, you move from “we kind of know how we do it” to a structured, AI‑ready workflow for one important micro‑process.

Conclusion

Making your business rules explicit is not corporate bureaucracy for large groups; it is the foundation that lets AI and automation create real value in an SME.

• As long as rules live only in people’s heads, AI cannot support you reliably.
• Starting from a single micro‑process keeps things light and delivers quick wins.
• Plain “if… then…” rules are the best building blocks for useful AI assistants.
• Separating what stays human from what can be prepared or triggered by tools keeps your projects safe and acceptable for your teams.
• Testing in simulation mode lets you refine your rules before exposing customers or critical data.

If you want support on this journey, Lyten Agency helps you identify, structure and automate your key processes. Get in touch for a free initial audit.