Trust rules for the payees you know and AI for the long tail you do not. A rule does exactly what you told it, every time, but only for the text it matches. AI can deal with merchants you have never listed, but it can be wrong and is harder to predict. The strongest setup uses both: rules first, then AI for what is left, and you review the leftovers.
How do rules behave?
A rule says: if the narration contains this text, use this category. It is predictable and easy to explain, and you can see why a transaction landed where it did. The cost is upkeep. Every new merchant, a changed payment handle or a differently spelled narration needs a new rule or an edit. Rules that are too broad can also catch the wrong thing, such as a word that appears in two unrelated shops.
How does AI behave?
AI generalises. It can place a merchant you have never classified by comparing it with similar ones, and it adapts to messy text without anyone writing a rule. The cost is that its answer is a judgement, not a promise. Two similar narrations may be treated differently, and it will not know that a payment to a particular person is your rent unless history tells it.
Rules or AI: how do they compare?
| Question | Rules | AI |
|---|---|---|
| Predictable? | Yes, same input gives same result | Mostly, not always |
| New merchants | Need a new rule | Handled as a best guess |
| Upkeep | Ongoing as merchants change | Low |
| Risk of a wrong category | When a rule is too broad | When the merchant is ambiguous |
| Easy to explain | Yes | Less so |
| Best for | Rent, EMIs, salary, regular bills | The long tail of small shops |
What is a sensible combined setup?
- Write rules for recurring payees where the answer never changes: rent to a named person, an EMI, a subscription, the same grocer.
- Let AI take the remaining transactions, if you are comfortable with that.
- Review everything the AI placed, starting with the larger amounts.
- When you correct a transaction, turn the correction into a rule if you expect it to repeat.
The goal is a shrinking pile: each month fewer lines need AI, because your rules cover more.
Here is a quick illustration with invented figures. Say forty regular payees cover most of your monthly lines. A few rules for those take an hour to write and then work every month. The rest is a scatter of one-time shops, where writing rules would never end. That split is why rules for the known and AI for the rest tends to work, and why the leftover pile matters: if it stays large, add rules for the repeat items inside it.
Review the AI's choices for the largest amounts first; a small mistake on a ₹120 snack matters less than one on a ₹12,000 payment.
How does SpendixAI order these steps?
SpendixAI applies your own rules first, then your history and similar merchants, and only then, if AI is on, an AI model for what remains. That model sees the normalised merchant name and an amount band, not your name, account numbers, full statement text or balances. You can switch AI off, cap it with a monthly budget, or hold unmatched items for review. Corrections become history, which behaves like rules you did not have to write.
When should you prefer rules only?
If your spending is concentrated in a few payees, or the categories are tied to tax or accounting treatment, rules give you control and a clear audit trail. If your spending is scattered across many small merchants, AI saves time, provided you spot-check it. Many people start rules-only, see how large the leftover pile is, and then decide about AI.
Useful ideas start with a clearer picture of your own spending.
Explore your reports


