Your Own Money Admin

The useful half of this course

Three lessons of caution is enough. There is a genuinely useful side to AI and money, and it has nothing to do with markets. It has to do with the pile of financial admin that most of us avoid: the letter with a reference number and no explanation, the statement with charges you cannot account for, the subscription you have meant to cancel since spring, the pension paperwork sitting in a drawer.

The reason this works is precisely the reason the trading claims do not. Language models are strong at explaining, rephrasing, structuring and drafting. Your bank's letter is a language problem, not a prediction problem. Nobody needs to know the future to tell you what "annual equivalent rate" means, or to turn a jumble of complaints into a clear, polite paragraph.

Two boundaries before we start, and they apply to everything in this lesson.

It cannot tell you what to do with your money, and you should not let it try. It has no idea about your circumstances, your obligations, your health, your family or the rules in your country, and a fluent answer about what you "should" do carries no more authority than a fluent answer about a chart. It is for understanding options and preparing questions, not for making the decision.

And it is unreliable at arithmetic. This surprises people, because computers are supposed to be good at sums. A language model is producing text that looks like a plausible calculation, and it will occasionally produce a total that is confidently and completely wrong. Every number that comes out of it needs checking with a calculator or a spreadsheet. Treat the words as helpful and the figures as unverified.

Understanding what you have been sent

This is the single best use. Take a confusing letter or statement and ask for it in plain English.

It is very good at unpacking financial jargon into language a human uses, at explaining what a term means in general, at telling you what a document appears to be asking you to do, and at listing what you would need to check to be sure. Where it is speculating, a well phrased prompt will make it say so.

I have received a financial letter I do not fully understand. Before I share it,
tell me which details I should remove or replace first, such as account numbers,
reference numbers, my address and my full name.
Then, once I paste the redacted version:

Explain in plain English what this document is and who sent it.
Explain every piece of jargon in it, in one sentence each.
Tell me what, if anything, it is asking me to do, and by when.
List anything that is unclear or that I should check with the sender.
Give me three questions to ask them.
Do not tell me what decision to make, and do not calculate anything for me.

Drafting the letter you have been putting off

The other reliably good use is writing things you find unpleasant to write. Complaints, cancellations, chasing a refund, asking a provider to explain a charge, requesting a call be escalated. Many people avoid these purely because starting is horrible, and the cost of avoidance is real money left on the table.

Here it excels, because a good complaint is a structured document: what happened, when, what you were told, what you want, by when. Give it your facts in whatever order they come out and it will return something calm and organised.

Help me write a formal letter. Here are my facts, in no particular order:
[what happened, dates, who I spoke to, what I was told, what I want to happen]
Write it:

Polite, firm, and unemotional.
Structured: what happened, what I was told, what I am asking for, by when.
Under one page.
With a line asking them to confirm receipt in writing.
Leave a clearly marked gap for every reference number rather than inventing one.
Then list anything I should attach as evidence.

Making sense of spending you have already exported

If you can get a list of your transactions out of your bank as a file, you can get real help organising it. The workflow that works is to export it yourself, strip the identifying columns, and then ask for structure rather than sums.

Ask it to propose a set of spending categories that fit your life rather than a generic list. Ask it to sort descriptions into those categories and flag the ones it is unsure about. Ask it to spot payments that look like recurring subscriptions, which is often the single most valuable thing it will do for you, since forgotten subscriptions are quietly common. Ask it to identify which categories vary most from month to month.

Then do the adding up in a spreadsheet, not in the chat. The categorisation is the labour you wanted to avoid. The arithmetic is the part it cannot be trusted with.

I am pasting a list of transaction descriptions and amounts. I have already
removed account numbers, my name and the balance column.

Propose 8 to 12 spending categories that fit this list, not a generic template.
Assign each transaction to one category.
Mark anything you are unsure about as UNCERTAIN rather than guessing.
Separately list anything that looks like a recurring subscription or standing
payment, with how often it appears.
Tell me which categories vary most between months.
Do not total anything. I will do the arithmetic myself in a spreadsheet.
Do not suggest what I should cut or how I should spend.

Building a budget structure

A budget is mostly a structure problem, which is why this works. You can ask for a sensible set of headings for your situation, a monthly template you can fill in yourself, a list of the irregular annual costs people commonly forget, and a short explanation of a budgeting method so you can decide whether it suits you.

What you should not ask is how much to spend on anything. The moment it starts assigning amounts to your categories, it is guessing at a life it knows nothing about. Structure from the tool, numbers from you.

Preparing for a conversation with a professional

If you are going to speak to a regulated adviser, an accountant, your bank or a debt charity, an hour with a chatbot beforehand is genuinely well spent. Not to get the answer, but to arrive able to follow the conversation.

Ask it to explain the vocabulary you expect to encounter. Ask what questions people commonly forget to ask, particularly about fees and how the person is paid. Ask what documents to bring. Ask it to explain, in general terms, how a type of product tends to work, so that the explanation you receive in the meeting is not the first one you have ever heard.

Then verify what matters with the professional in front of you, who is accountable for what they tell you in a way that a chatbot is not.