AI Workflows for Small Business: Five You Can Run on a Workflow Tool and a Model API
Five AI workflows a small business can run on a low-code tool and a model API: enquiry triage, follow-up, invoice extraction, review replies and a document assistant. Each with the rule steps, the model steps and the review point.
On this page
AI workflows for small business are automations built on a low-code workflow tool, with a model API on the steps a rule cannot do, that remove the reading and drafting work from the handful of processes a small team runs every day. The five below are the ones that pay first in almost every small business I look at. Each is mapped into rule steps, model steps and a review point, because that map is the difference between a workflow that runs for a year and one that gets switched off after a bad week.
Direct Answer
The five AI workflows worth building first in a small business are: enquiry triage and reply drafting, follow-up on quiet threads, invoice and receipt extraction, review reply drafting, and a document assistant for internal questions. All five run on a workflow tool such as n8n, Make or Zapier plus a model provider’s API, cost tens of euros a month combined, and keep a person on the review point before anything is sent or paid. Start with enquiry triage; it has the clearest baseline and the most reversible actions.
Key Takeaways
- Five workflows, each mostly rules with a model on one or two steps.
- Every workflow has a review point before an irreversible action.
- The tools are pay-as-you-go. The investment is the design and the first weeks of review.
- Money out, complaints and legal matters stay manual.
Which five AI workflows should a small business build?
| Workflow | Rule steps | Model steps | Review point | Baseline |
|---|---|---|---|---|
| Enquiry triage and reply | Trigger on form or email; create record; send after approval | Classify the enquiry; draft the reply | Owner approves the draft | Hours to first reply |
| Follow-up on quiet threads | Detect no reply after N days; schedule | Decide whether a follow-up fits; draft it | Approve before sending | Share of threads followed up |
| Invoice and receipt extraction | Trigger on attachment; write to accounting tool | Extract supplier, date, total, VAT from any layout | Bookkeeper checks flagged ones | Minutes per invoice |
| Review reply drafting | Trigger on new review; publish after approval | Draft a specific reply from the review text | Owner edits and approves | Share of reviews answered |
| Document assistant | Index documents; log questions | Answer from documents with the source | Owner corrects the document when wrong | Interruptions per week |
IBM’s intelligent automation overview describes the general split between rules and models. What the table adds is the review point, which is where most small-business automations are missing a step: the model drafts and the workflow sends, and the first wrong message ends the project.
How do you build one in a workflow tool?
Take enquiry triage. In a tool like n8n, whose documentation shows the building blocks, the workflow is six nodes. A trigger watches the inbox or the form. A model node receives the message text and returns a classification and a draft reply. A rule node creates the CRM record with the classification. A notification node sends the draft to the owner’s phone with approve and edit buttons. A wait node holds until the owner responds. A send node sends the approved text and logs it.
The model node is where the provider’s API is called. For workflows where the model must look something up before drafting, such as the customer’s history, the developer pattern is tool-calling; OpenAI’s function calling guide is the reference. For the five workflows here, the low-code model node is usually enough.
How to start with AI workflows in a small business
- Pick enquiry triage. Count enquiries for a month and note the time each waited for a reply.
- Write the classification categories and a reply template per category. In the owner’s words.
- Build the six nodes. Trigger, model, record, notify, wait, send.
- Run it with the owner approving every draft for two weeks. Corrections go into the template.
- Add follow-up. Same tool, same model node, a detection rule for silence.
- Add invoice extraction when the bookkeeping backlog hurts. Flag anything the model is unsure of for the bookkeeper.
- Add review replies and the document assistant last. They pay, but the first two pay more.
- Measure each against its baseline after a month.
What AI workflows for small business are not
They are not autonomous. Every one sends or pays only after a person has approved.
They are not a platform decision. Any of the three workflow tools handles these five. The comparison in n8n vs Make matters for the sixth workflow and beyond, not the first.
And they are not a replacement for writing down how the work is done. The classification categories and the reply templates are the work. The tool just runs them. The general definition is in what is AI automation.
Summary
Five AI workflows pay first in a small business: enquiry triage and reply, follow-up, invoice extraction, review replies and a document assistant. Each is a handful of rule steps with a model on one or two, and a review point before anything irreversible. Build them on a workflow tool plus a model API, start with enquiry triage, review for two weeks, and measure against the baseline. The wider library is under AI automation; the general small-business argument is in AI for small businesses. If you want the first workflow designed with you, that is AI prototypes and pilots.
Frequently asked questions
What is the easiest AI workflow for a small business to build?+
Enquiry triage: when a form or email arrives, a model classifies it and drafts a reply, and the owner approves from a phone. It is a handful of steps in a workflow tool and pays from the first week.
Which tools do I need?+
A workflow tool such as n8n, Make or Zapier for the rules and connections, and a model provider's API for the judgment steps. Both are pay-as-you-go at small-business volume.
Do these workflows need a developer?+
The five here can be built in a low-code tool by someone comfortable with it. When the workflow must write to several systems with permissions, or handle money, bring a developer in.
How much do AI workflows cost a small business?+
Tool subscriptions plus model usage, typically tens of euros a month for all five combined at small-business volume. The real cost is the days spent designing and reviewing.
What should stay manual?+
Anything involving money out, complaints, legal matters, and any message that carries the relationship, such as a first reply to a big prospect. The workflow prepares; a person sends.
Sources
- n8n documentation: n8n
- What is intelligent automation?: IBM
- Function calling: OpenAI
