Iro AIWhat you'll be able to do
- Write a spreadsheet formula from a plain description — and get it explained so you can adapt it
- Decode an inherited formula you didn't build and understand exactly what it does
- Summarize a piece of guidance or a reg into a high-level plain-English brief
- Draft a clear, professional client email in a fraction of the time
- Clean and restructure messy exports without hand-editing every row
Inside the path
A focused set of five-minute lessons — each one ends with a hands-on exercise, not a quiz you can guess.
Formulas, written and explained 6 min
Describe what you need in plain words and get a working formula plus an explanation you can adapt and trust.
Decode an inherited spreadsheet 5 min
Paste a nested formula you didn't build and have AI walk you through what each part does.
Summarize guidance at a high level 6 min
Turn dense standards and regs into a plain-English overview — as orientation, not authority.
Client emails that land 5 min
Draft clear, professional client and colleague emails by giving the model tone and the key points.
Clean data without the guesswork 5 min
Standardize names, split columns, and spot outliers — then verify the totals tie out.
Try a sample exercise
This is the kind of card you'd practice inside Iro — you do the thinking, then get feedback.
◆ Sample exercise · Prompt practice
You have a workbook where column B holds client names, column C holds invoice dates, and column D holds amounts. You need a formula that totals one client's Q1 invoices — and you don't want a number you can't trust.
Your task: Pick the prompt that gets a usable, verifiable formula.
- "Write a formula to add up my sales."
- "In Excel, column B has client names, column C has invoice dates, column D has amounts. Write a formula that sums column D where B = \"Acme\" and C falls in Q1 2026. Here are 3 sample rows: [paste]. Explain how the formula works so I can adapt it, and note edge cases like blank cells or text in column D. I'll test it against a total I already know before trusting it."
- "What was Acme's total spend in Q1? Just give me the number."
- "Do the client's Q1 books and tell me if the numbers look right."
See why the second prompt wins
The winning prompt gives the model everything it needs and keeps you in control: it supplies context (what each column holds), includes a sample of the data so the formula fits your real layout, asks for an explanation so you can adapt it instead of pasting blindly, requests edge-case handling (blank cells, stray text), and ends with a verification step — testing against a known total. The vague "add up my sales" gives you a generic formula that may not match your columns. "Just give me the number" is the trap: the model can't see your data reliably and will confidently return a made-up figure. And "do the books" hands professional judgment to a tool that has none. In Iro you'd write your own version and get feedback on context, samples, and where to verify.
Language and logic, not arithmetic
The fastest way to get burned by AI in accounting is to treat it like a calculator. Language models don't compute — they predict text — so when you ask for a total or a variance, they'll often return a number that looks right and isn't. That confidence is the danger: a wrong figure delivered in a clean sentence is easy to miss.
Point AI at language and logic instead and it becomes genuinely useful. It writes the formula that does the math inside your spreadsheet, explains a nested function you inherited, summarizes a piece of guidance so you know where to look, drafts the client email, and cleans up a messy export. In every one of those, the actual numbers are still produced and checked by you or your tools — the model just does the reading, writing, and structuring around them.
A verify-first workflow for accountants
The habit that makes AI safe here is simple: never let a number leave the model unchecked. Build the check into how you prompt:
- Give context and a sample: tell it what your columns hold and paste a few real rows so a formula fits your data.
- Ask for an explanation: a formula you understand is one you can adapt and trust; a black box is a liability.
- Test against a known total: run any new formula on a figure you already know before relying on it.
- Treat guidance summaries as orientation: use AI's plain-English overview of a standard or reg to know where to look, then confirm the detail in the authoritative source.
Do that and AI clears the busywork off your desk without ever putting an unverified number in front of a client. This is a productivity tool — the responsibility for accuracy stays with you.