How to Track Expenses

What tracking is actually for

Most people start tracking with a vague goal: to be better with money. That goal cannot tell you when to stop, so tracking becomes an open-ended obligation, and open-ended obligations get dropped.

Tracking does three specific jobs. It shows you what happened, which is the awareness job. It tells you whether a plan survived contact with the month, which is the budget job. And it produces a record you can hand to someone else, which matters for tax, for a business, or for splitting costs with a partner or a flatmate.

There is something to be honest about here, which is what tracking does not do on its own. In a randomised trial in Austria, Stefan Angel tested three financial tools including a budgeting app, and found that app users checked their account balance significantly more often than the control group. What the study did not find was evidence that any of the tools improved financial behaviour by themselves. The authors concluded the results do not support relying on standalone digital tools of that design.

Read that as a warning against the wrong expectation rather than an argument for giving up. Tracking buys you information. Information only becomes money when it feeds a decision you were already trying to make. So pick the decision first.

For most people it is one of four: whether to cancel something, whether a category is out of proportion, whether income covers outgoings before payday, or how much is genuinely spare each month.

The four ways people track

There are four methods in common use. None of them is the right answer for everyone, and the trade-off is always the same one: how much work you do versus how much the system does.

Pen and paper, or a notes app. Still the most common method among people who budget at all. In Debt.com’s July 2026 survey of just over a thousand US adults, 37 percent of budgeters used pen and paper, ahead of spreadsheets at 27 percent and mobile apps at 22 percent. It is free, it starts in ten seconds, and writing something by hand makes you notice it. What it cannot do is add up reliably, search, or answer a question about last March. Good for noticing, bad for analysis.

A spreadsheet. In Britain the ranking flips. A February 2026 YouGov survey of 2,087 adults found that among people with a budget, 39 percent used spreadsheets or similar tools, against 9 percent using a budgeting app. The same survey found 36 percent using no specific tool at all, which is where a lot of abandoned spreadsheets quietly end up. A spreadsheet is a powerful thing, it costs nothing, and the file stays yours forever. The price is that you are the data entry clerk, permanently, and there is nobody auditing your formulas. Research into business spreadsheets by Raymond Panko found errors in a substantial share of the spreadsheets audited, usually a formula that had stopped covering the range it was supposed to cover. Those were business spreadsheets with more at stake than yours, but the failure mode transfers: a total that stopped counting December looks exactly like a total that works.

An app with automatic categorisation. Less typing, and the app does the arithmetic. Two honest caveats. Categorisation rules need maintenance, because merchant names are messy and one badly-named vendor can quietly send six months of spending into the wrong bucket. And in Auritrack’s case specifically, there are no live bank connections yet, so data gets in by describing it, entering it, photographing a receipt, or uploading a statement.

AI-assisted tracking. You describe what happened in ordinary language and the software does the filing. Spent £14 on lunch at the place near work. It reads the amount, the vendor, and the date, picks a category, and files it. It handles “last Friday” without a date picker, and several transactions in one message. The cost of this method is real and should be stated plainly: it is a paid feature, and it will occasionally file something wrong, so you have to be willing to correct it. With Auritrack the free tier covers manual tracking, budgets, and storage, while the AI features run on a plan or on pay-as-you-go Auricoins that do not expire.

If you want the long version of why spreadsheets in particular tend to die, track expenses without spreadsheets goes through it properly.

Picking categories that earn their place

The most common way a tracking system dies is not laziness. It is forty categories.

Someone sets up Groceries, Restaurants, Coffee, Takeaway, Snacks, and Work Lunches, and then every logged expense requires a small judgement call. A sandwich eaten at your desk from a supermarket is which of those? The judgement is tiny but you make it several times a day, and the friction compounds until logging feels like admin.

A category earns its place only if you would do something different based on its number. If Coffee and Restaurants would produce the same action from you, which is probably “eat out less”, they are one category.

Start with six to eight. Something like housing, food, transport, bills and subscriptions, health, and one bucket for everything else. Split later, and only when a number is too big to act on. If Food comes back at €640 for the month and you have no idea what to do with that, then splitting out delivery is worth it, because “€300 of that was delivery” is a fact you can act on and “you spent €640 on food” is not.

The everything-else bucket does real work here. Without it, every unclassifiable purchase stalls the whole system while you decide where it goes. With it, the awkward ones land somewhere and you keep moving.

Receipts and statements

Two things reliably jam the machinery: the pile of paper, and the fact that your bank’s version of events arrives late and badly labelled.

The receipt problem is one of timing. A receipt is only useful in the moment it is handed to you, because that is the last point at which you remember what the purchase was for. A week later it is a slip of thermal paper with a total and a merchant code, and you are guessing. Photograph receipts at the point of sale, or do not keep them at all. The pile on the kitchen counter feels like a record but functions as a stack of postponed decisions, which is why it keeps growing.

The statement problem is different. Your statement is the source of truth, since nothing gets in or out of the account without appearing on it, but it shows up weeks after the fact, in a layout designed for a bank rather than a person, with merchant names like SQ *A4TRDNG that mean nothing. Statements are for reconciliation, not for tracking. Once a month, check that your records and the statement agree, and expect to find two or three things you never logged. That check is the point.

In Auritrack, the receipt route is a photograph and the AI reads the total. For a statement, upload it as PDF, CSV, or Excel up to 1MB, and the AI extracts the transactions, separates income from expenses, and proposes categories. You review and edit the list before anything saves, so nothing lands in your records unchecked. Text-based PDFs parse well, photographs of printed statements much less so. Each import costs 10 Auricoins.

Clearing a backlog

Everyone who starts tracking properly wants to begin by reconstructing the past year. Almost nobody finishes.

The instinct makes sense, because a year of history sounds like it would tell you more than a month. In practice reconstruction is the single hardest task in expense tracking, and putting it first means the first thing you do is the thing most likely to make you quit. It is also the task your last attempt probably died on.

There are two honest options.

Start from today. Log nothing before this morning. You will have a usable picture in four weeks and a good one in three months, and the cost is that you cannot answer questions about last year. For most people that cost is imaginary, since they were not going to answer those questions anyway.

Or import the recent past. Sixty to ninety days is the sweet spot: recent enough that you still recognise the transactions when you review them, long enough to catch quarterly and annual charges. Pull the statements, import them, and review the categories in one sitting.

What you should not do is reconstruct from memory. Say you sit down to rebuild March and you remember rent at ₦450,000, transport at roughly ₦60,000, and food at “maybe ₦120,000”. The rent is right because it is fixed and memorable. The other two are guesses shaped by what you expected to spend, which is exactly the bias the subscription research measured. A reconstructed month is not data. It is your assumptions, written down in a format that looks like data, which is worse than having nothing because now you trust it.

How to tell if it is working

Tracking systems tend to be judged by the wrong measure. People count consecutive days logged and treat a gap as failure, which turns a tool into a test you can fail.

Better to ask three questions.

The first is whether you can answer a question about your own money in under a minute. Pick one: what did I spend on transport last month? If finding out takes twenty minutes of scrolling, the system is storing data without returning any, and something needs to change, usually the categories.

The second is whether you are still being surprised. Early on, most months contain at least one real surprise. As tracking does its job those surprises should shrink, not because you are spending less but because your expectations have moved to match reality. Once you can predict your own monthly total within roughly ten percent, tracking has delivered most of what it has to offer.

The third is whether one decision has changed. Not your whole financial life. One decision: a cancelled subscription, a different weekly shop, something you did not buy because you already knew what the month looked like.

Subscriptions are the fastest test of all three, because they are where the gap between belief and reality is widest and most measurable. Citizens Advice found that unused subscriptions cost UK consumers £688 million in a single year, more than double the figure from two years earlier, and that 26 percent of UK adults had accidentally taken out a subscription in the previous twelve months. Forty percent of those said it auto-renewed without them realising. Very little of that is carelessness on the customer’s side. Economists working with real card data have shown that inattention to subscriptions is worth a great deal of money to the companies selling them, which means nobody on the other side of the transaction has any reason to remind you.

Getting past week two

Week two is where tracking dies, and it is worth knowing that this is normal rather than personal.

The popular claim that a habit takes twenty-one days has no real evidence behind it. When researchers at UCL actually measured it, following 96 people performing a daily behaviour for twelve weeks, the median time to reach something like automatic was 66 days, with a modelled range running from 18 days to well over 200. A 2024 systematic review of 20 studies covering 2,601 people put the median at 59 to 66 days, with individual times ranging from 4 to 335 days. Both looked at health behaviours rather than money, so treat the numbers as the shape of the thing rather than a promise.

The shape is what matters: at week two you are somewhere near the beginning. It feels like effort because it is still supposed to feel like effort.

Two design principles follow.

Lower the cost per entry until it is nearly nothing. Anything that requires opening a file, finding the right row, and choosing from a dropdown will lose to the thing that takes one sentence at the moment of payment. This is the strongest argument for AI-assisted tracking and the honest reason spreadsheets lose: not that they are worse tools, but that they cost more per use, and cost per use is what determines survival.

Design for your worst week, not your best one. Everyone builds a system for the version of themselves with a free Sunday morning. You need a version that survives the week you are ill, travelling, or busy. That usually means batching is allowed: four days logged in one message on Sunday is a complete success, not a partial one. Missed days are not debt. If you skip a week, start again on the current day and let the gap stay a gap, because going back to fill it is the reconstruction trap wearing a different coat.

Your first week

Nothing here needs a fresh start, a migration, or a system. It needs about four minutes today.

Today. Run one receipt through the free receipt scanner. No account needed. This is the habit in miniature, and it tells you whether the four-second version of tracking feels different to the version that made you quit last time.

This week. If it did feel different, create an account or get the app on Google Play or the App Store, and log one expense by typing what you spent. Just the one, without building a category structure or a plan around it first.

This month. Run the subscription tracker and cancel one thing. Then set the ten-minute monthly review.

If you want a plan to compare the numbers against once you have a few weeks of them, the budget planner is the next tool along, and the AI bookkeeping page explains what the assistant actually does with a transaction once it has one.

The subscription study at the top of this guide is the argument for all of it. Those 2,500 people were not careless with money. They just had no mechanism for knowing what they spent, and the study showed that concentrating harder does not substitute for one.