An AI workflow connects a useful task to a repeatable process: collect the inputs, check them, let AI help with a defined step, and review the result before it is used. Start with work that already has an owner and a clear definition of "done".
At a glance
- Start with client reporting and fictional data.
- The first version creates a draft for review.
- Use the invoice guide for payment follow-up.
Start with client reporting
For: small agencies and consultancies that prepare recurring client updates.
The workflow: approved figures → calculated changes → AI draft → human review → client-ready report.
The output: a short draft summary with links back to its source figures. The first version produces a draft; a person decides whether to send it.
Follow the reporting walkthrough · Get the free prompt, sample data and checklist
Two more workflows to scope
Turn an enquiry into a draft response
Use a new enquiry as the trigger. Validate the contact details, extract the request, check it against your services and draft a reply for the account owner. Keep uncertain or incomplete requests in a review queue. The system should not invent availability, promise a price or make a booking without the rules and approvals you have set.
A useful first test: can it recognise an enquiry that asks for something you do not sell? Read how to build an AI agent for implementation routes.
Prepare an invoice follow-up
Start with an approved list of overdue invoices. Check payment status and any dispute notes, then draft a polite reminder for review. A human should resolve disputed balances and sensitive customer situations before contact.
A useful first test: does it stop when an invoice has already been paid? Use our existing invoice-chasing guide for the next step.
What a useful workflow guide should tell you
- The input, output and person responsible.
- The software and access it needs.
- Which steps use fixed rules and which use AI.
- What to do when data is missing, ambiguous or out of date.
- The cost of running and reviewing it.
- The tests to run before relying on it.
The reporting example on this site is a design walkthrough using fictional data. It is not a claim that a particular platform or model has passed production testing. Check each guide's evidence note before using it.
Choose a platform after you choose a task
Use our tool-selection guide to choose a route. If your shortlist includes automation platforms, read Make versus n8n. If you will ask somebody to build it, prepare an implementation brief first.