The First AI Workflow I Would Build for a Small Business

Most small businesses should start closer to the inbox

When a small business decides to use AI, the first idea is often a chatbot, a custom assistant, or a large internal knowledge base. Those ideas can be useful later. They are rarely the best place to begin.

The first workflow I would usually investigate is simpler:

> When a new enquiry arrives, turn it into a clear, reviewed next action.

That might mean identifying what the person needs, checking whether the request is a good fit, drafting a reply, creating a follow-up task, and recording the information in the right system.

This is not an attempt to remove the human from the relationship. It is a way to stop good enquiries from getting lost in a busy inbox, a contact form, or a collection of disconnected notes.

Why this is a strong first workflow

A useful first AI workflow usually has five characteristics:

1. It happens often enough to matter.

2. It contains information AI can help organise.

3. It still benefits from human judgement.

4. The business can recognise a better result.

5. It can be improved without rebuilding the whole company.

New enquiries often meet all five conditions. They are repetitive, usually text-heavy, and connected to revenue or customer experience. At the same time, the final decision should normally stay with a person who understands the business.

The workflow is valuable even before the AI becomes sophisticated. A clear structure, consistent notes, and reliable follow-up can improve the operation by themselves.

The workflow I would map first

The first version could look like this:

1. Capture the enquiry

The trigger might be a website form, an email inbox, a CRM record, or a message from another channel. The important part is not the tool. It is creating one reliable starting point.

The workflow should capture the original message, the sender, the time received, the source, and any information the customer already provided.

2. Clean and structure the information

Free-form messages are difficult to manage consistently. An AI step can extract a small set of useful fields, such as:

• What the person is asking for.
• Whether they are asking for a project, support, a quote, or general information.
• Their timeline, if they mentioned one.
• The systems or products involved.
• Questions that still need an answer.

This is where a tool such as Claude or another language model can help. It is not deciding the entire relationship. It is turning unstructured language into a format the business can review.

3. Classify the next action

The output should not be “AI thinks this is important”. It should be a practical recommendation:

• Reply with an existing answer.
• Ask for missing information.
• Suggest a discovery call.
• Send the request to a specialist.
• Add a follow-up task.
• Mark it as outside the current offer.

The categories should be defined by the business before the model is connected. Clear categories make the workflow easier to test and easier for a team to trust.

4. Draft, do not blindly send

The system can prepare a reply using the approved tone, service information, and relevant context. A person should review the draft before it is sent, especially when the message includes pricing, commitments, sensitive information, or a difficult customer situation.

This approval step is not a sign that the automation failed. It is the design choice that keeps the business accountable while removing the blank-page work.

5. Log the decision

The final step is often overlooked. The workflow should record what happened: the category, the next action, who approved it, and whether the enquiry moved forward.

That record can live in a CRM, a database, a project tool, or another system the team already uses. The goal is not to create another dashboard. The goal is to make the next step visible.

A simple technical shape

The implementation can be modest. A form or inbox provides the trigger. An orchestration layer such as N8N moves the information between steps. A language model extracts fields and drafts a response. A human approval step controls the outgoing action. The result is saved to the CRM, database, or task system.

The exact stack will depend on the business. I have worked across APIs, WordPress, Salesforce, React, TypeScript, Vercel, Supabase, and automation tools, so I think about the connections as part of the product—not as an afterthought.

The important design questions are:

• Where does the enquiry enter?
• Which information is safe and useful to process?
• What categories does the team actually use?
• Who approves the response?
• Where is the decision recorded?
• What happens when the model is uncertain?

If those questions are unclear, adding another AI tool will not solve the underlying problem.

What I would measure

The first workflow does not need a complicated return-on-investment model. I would start with a small set of operational signals:

• How long it takes to acknowledge a new enquiry.
• How many enquiries receive a clear next action.
• How often a human has to correct the classification.
• How many follow-ups are completed on time.
• Whether the team feels the workflow is easier to use than the old process.

These measures do not prove that AI created revenue by themselves. They show whether the workflow is becoming more reliable. Once the process is stable, the business can connect it to commercial outcomes more carefully.

What I would not automate first

I would not begin by letting an AI agent make final pricing decisions, promise delivery dates, reject valuable customers, or send sensitive messages without review.

I would also avoid automating a process that nobody understands. If the business cannot explain what a good enquiry looks like or what should happen next, the first project is probably process clarification—not AI.

Automation should reduce unnecessary effort, not hide uncertainty inside a faster system.

The principle I keep coming back to

The best first AI workflow is not the most impressive one. It is the one that connects a repeated business problem to a visible next action, with enough human control to earn trust.

For many small businesses, that means starting with the journey from enquiry to follow-up. It is close to the work, close to the customer, and small enough to improve without turning into a giant transformation programme.

If you are unsure where AI belongs in your business, you can book an AI Opportunity Session. We can map the workflow you already have, identify where automation would help, and decide what should stay human.