A new way to read your own records
SpendixAI can connect to AI assistants through MCP, so questions about spending can be asked in plain language. The assistant reads the records and responds. This is convenient, but the answers are only as good as the question and the underlying data. A few well-framed questions on food spending make the tool useful, and a habit of checking the replies keeps it reliable.
Question one: where did the increase come from?
Ask for a comparison of this month’s food spending with the last, listing the three merchants or categories that changed the most. The reply should give amounts, not just impressions. Compare them with the dashboard to confirm.
If the assistant names a merchant you do not recognise, that is a prompt to check aliases.
Question two: how often were orders placed?
Ask how many food delivery orders were placed in each of the last three months, and what the average amount was. Counts are often more revealing than totals, and the assistant can produce them quickly.
Question three: what is a typical grocery week?
Ask for the average weekly grocery spending over the last twelve weeks, excluding the highest week. The exclusion guards against a stock-up skewing the figure. This gives a realistic baseline for the weekly budget.
Question four: are there duplicates or odd charges?
Ask for food-related transactions of the same amount on the same day. The reply highlights candidates for duplicate review. Verify each against your order history before concluding anything.
Question five: what would a modest change save?
Ask what a year would save if delivery orders fell by two a month at the average amount. The assistant will multiply, and the answer is arithmetic on your data. It is a prompt for a decision, not a prediction of behaviour.
| Question type | What to check in the reply |
|---|---|
| Comparison | Totals match the dashboard |
| Counts | Orders match app history |
| Averages | Excluded items are as asked |
| Duplicates | Each matches a real order |
| Scenarios | The assumptions are stated |
Check the answers
Treat the assistant’s replies as drafts. Confirm key numbers against the app. If something is off, a common cause is a category mismatch or an uncategorised merchant, which is a cue to tidy the data. Assistants can also misunderstand a question, so rephrase if the answer seems unrelated.
Keep decisions human
The assistant can summarise and calculate, but decisions about food, family and money are yours. Use the replies to prepare for the monthly review, not to replace it. The best use is to reduce the time spent finding numbers so that more time goes to deciding what to do with them.
Protecting your privacy
Spending data reveals a lot about a household: its routines, health needs and location. Before connecting any assistant, read what it will be able to see and how long it keeps the data. Use the permission controls in the app to limit scope, and revisit them from time to time. If a reply ever contains information that does not look right, treat it with caution and verify before acting.
Never paste passwords, card numbers or one-time codes into a conversation with any assistant.
Cleaning up old connections
Review the list of connected apps in your account settings every few months. Remove connections you no longer use, and check that those remaining have only the permissions they need. A tidy set of connections reduces risk and keeps the experience simple. If an assistant ever suggests an action that moves money or changes settings, stop and check; the safest default is to treat the assistant as a reader of records, not a decision-maker.
Useful ideas start with a clearer picture of your own spending.
Explore your reports


