AI Agents & Workflow Design for Finance
Move from individual requests to repeatable work
You may already use AI to draft an email or explain a spreadsheet. The next step is getting it to follow a process: use the right files, complete the checks, prepare the output and identify what still needs your attention.
First, try three finance examples with supplied files and worked answers. Then choose one project and develop it through the remaining lessons. You will write the instructions, save them in ChatGPT, test what happens when information changes and prepare a handover.
What you will be able to do
- ✓Choose a recurring task with a clear purpose, manageable scope and a useful output.
- ✓Define the sources, steps, calculations and review points before building.
- ✓Create a reusable ChatGPT skill using written instructions, without writing code.
- ✓Test missing information, changed balances, policy limits and new reporting periods.
- ✓Set up and review a scheduled test where your account supports it.
- ✓Prepare a handover so the workflow can be reviewed, maintained and used again.
Practise with work you recognise
Monthly reporting
Prepare a variance table and management summary. Check the calculations and distinguish supported explanations from questions that still need an answer.
Expense review
Apply a supplied policy, retain every claim and identify missing receipts, approval gaps and possible duplicates for the reviewer.
Customer follow-up
Review overdue balances and prepare reminder drafts while accounting for part payments, disputes and invoices that are not yet due.
A short client document exercise introduces the method. For the final project, choose reporting, expense review or customer follow-up and save your workflow as a reusable skill.
Who this course is for
Designed for accountants, finance managers, financial controllers, analysts, advisers and business professionals working in firms or in-house teams.
You should be comfortable using AI for basic tasks, uploading files and reviewing business outputs. You do not need programming experience. Course 1 or Course 2 can provide a useful foundation, but equivalent practical experience is sufficient.
What you will learn
▸Module 1: See how an agent handles a finance taskComplete a short document check and understand the steps behind it.
- ✓Start with a task you recognise
- ✓What does the agent do?
- ✓Complete your first task in ChatGPT
▸Module 2: Try three finance workflowsSee what reporting, expense review and customer follow-up look like in practice.
- ✓Prepare a monthly management report
- ✓Prepare an expense review register
- ✓Prepare customer follow-ups without chasing the wrong invoice
▸Module 3: Plan your chosen projectChoose one of the examples and write the instructions for your own workflow.
- ✓Choose the project you will build
- ✓Write the instructions for your project
- ✓Decide what should happen when information is missing
▸Module 4: Save your workflow in ChatGPTSave your reviewed instructions as a skill, then use it in a new chat.
- ✓Check your instructions before saving them
- ✓Save your instructions as a skill
- ✓Use the saved skill in a new chat
▸Module 5: Check and improve your workflowTest the same project with missing or changed information and correct any errors.
- ✓Try missing files and changed figures
- ✓Check the figures and source records
- ✓Correct an error and check it again
▸Module 6: Repeat the work and hand it overUse updated files, try a scheduled task where available and prepare the handover.
- ✓Use updated information for the same task
- ✓Try running the task at a scheduled time
- ✓Hand over the method and assess whether it helps
The final project check brings together work completed during the lessons. You then take the final knowledge check. The six-hour estimate includes time to build, test and review.
What you take away
- ✓A reusable finance workflow you have built and tested in ChatGPT.
- ✓An Excel practice workbook, equivalent CSV files and supporting policies and evidence.
- ✓Starter briefs and worked examples for three final project options.
- ✓Templates for your workflow brief, test record and handover.
- ✓A certificate of completion after completing the course requirements.
Which ChatGPT plan do you need?
For individual study, we recommend ChatGPT Plus. An existing Pro account or an organisation-approved Business or Enterprise workspace is also suitable when the required features are enabled. Enterprise is not required.
Use the ChatGPT desktop app with Work, file uploads and personal Skills available. Work is the mode used to carry out the tasks; a skill saves the instructions for a task you want to repeat. Check access before enrolling. Software costs and any extra usage charges are separate from the course fee.
▸Other plans and access settings
OpenAI also lists Work on Free and Go. Their usage allowance may interrupt repeated practice, so we recommend Plus for individual study. In a workplace account, your administrator may control access to Work and Skills. Check Scheduled separately when you reach Module 6.
Product guidance: OpenAI: plans and usage , OpenAI: build skills , OpenAI: Work access and workspace controls .
Scheduled is used where enabled. A local scheduled task needs the app open and the computer awake; a cloud task needs access to the files from where it runs. If scheduling is disabled, complete the design and manual test and record that the schedule remains untested.
Product guidance: OpenAI: scheduled tasks .
Practise with fictional data
All exercises and projects use the supplied fictional files and fictional test changes. Do not upload, paste or connect real client data, employee personal information or confidential business records. This applies to every plan, including Business and Enterprise. No connection to live finance systems is required.
Created with a finance professional's perspective
Angela Shi
Founder & CEO, Empathetic AI · Former fintech CFO · Finance educator
Angela brings more than 15 years of experience across financial services and FinTech, alongside her work developing AI applications and teaching business and finance professionals.
The course reflects the questions finance teams face when adopting AI: what work can be delegated, what evidence supports the result and who remains responsible for the final decision.
Developed by Empathetic AI, an Australian AI company working at the intersection of finance expertise, practical implementation and education.
Common questions
▸How is this different from learning to prompt?
You move beyond a single request. You define a multi-step process, save it as a reusable skill, test it with changed inputs and plan how to repeat and review the work.
▸Do I have to use one finance example throughout?
You first try reporting, expense review and customer follow-up. Then you choose one project to develop through the build, testing and handover lessons. Other examples continue to illustrate the method, but you keep one project for your own practical work.
▸Will I build an AI agent?
You will configure ChatGPT Work to carry out a finance workflow and save the method as a reusable skill. The course focuses on practical business configuration, testing and review. It does not require you to train a model or develop a software application.
▸Do I need Microsoft Copilot or a Claude subscription?
No. The full practical build uses ChatGPT Work. A short comparison explains how the workflow method can transfer to Claude and Microsoft environments. Separate subscriptions and builds on those platforms are not required.
▸Do I need ChatGPT Enterprise?
No. We recommend Plus for individual learners, or an existing Pro, Business or Enterprise account with the required features enabled. All course work uses fictional data, regardless of plan.
▸How do I earn the certificate?
Complete the lessons and practical activities, finish your chosen project and score at least 80% in the final knowledge check. The certificate is issued by Empathetic AI Academy.
▸Is the course suitable for a team?
Yes. The examples suit both client work and internal finance work. Teams can use the same brief, test record and handover structure to discuss which workflows to adopt.
Make your next recurring task easier to prepare and review.
Build the method, test the result and leave with a workflow you understand.