AI financial assistant India: answers from your own debits

Two kinds of software call themselves AI money assistants. The first is a general-purpose chatbot: it explains, rephrases and does sums on whatever you type. The second reads your transactions and answers from them, the debits that left, the dates they left on and the merchants behind them. The model inside can be the same. What differs is the input.

That matters most for affordability, which is what people actually bring. 'Can I take this loan', 'can I afford this trip' and 'why is nothing left by the 20th' share one structure: take-home, minus what is already committed, minus what you usually spend, set against the new thing. The rule is simple. Every input is a fact about your accounts, and a chatbot only has the version of those facts you remember.

Which jobs a finance app should do without any typing, with or without AI, is scored on the personal finance app page. This page is about the one job an assistant adds: answering a question you ask, in your own rupees.

The household: ₹95,000 take-home, renting, a car loan, a SIP, the usual subscriptions. They want about ₹4.5 lakh for a home renovation, and a lender's calculator says ₹4,51,613 at an illustrative 12 percent over 36 months is ₹15,000 a month.

The first answer is the rule of thumb, which is what you get when you ask with no other detail: keep EMIs under about 40 percent of take-home. ₹15,000 is under 40 percent of any take-home above ₹37,500. At ₹95,000 it is 15.8 percent. Yes, comfortably. The rule is not wrong; it is a cap on total EMIs, and the generic answer has quietly assumed the new one is the only one.

The second answer comes from a chatbot given what the household remembers: rent ₹24,000, car EMI 'about ₹12,000', SIP ₹10,000, everything else 'about ₹30,000'. A good chatbot does the sums correctly. EMIs after the loan would be ₹27,000, a FOIR of 28.4 percent, with ₹26,000 of room under a 40 percent cap. The month would go from ₹19,000 left to ₹4,000 left. Yes, with a margin.

Every step is sound; the inputs are not. The remembered commitments come to ₹46,000. The actual ones, read from three months of debits, come to ₹69,662. The ₹23,662 between them is five things nobody lied about: a phone bought on a card EMI, an air conditioner on a zero-interest consumer loan, a car EMI ₹601 above 'about ₹12,000', ₹30,000 of term and health premiums that leave once a year and so never feel monthly, and ₹6,000 sent to parents, which the household thinks of as family rather than spending.

An assistant that reads the debits starts from what left the account. EMIs: the car loan at ₹12,601, the phone on the card at ₹5,415, the air conditioner at ₹3,500, ₹21,516 in all. Many lenders also count a notional EMI on a credit card, often 5 percent of the outstanding; on a ₹40,000 statement balance that adds ₹2,000. Obligations today are ₹23,516, a FOIR of 24.8 percent. Add the new ₹15,000 and it becomes 40.5 percent: over a 40 percent cap, under a 50 percent one. Some lenders would sanction it; others would come back with a smaller amount.

The lender's question is not the household's, though. The household needs to know whether the month can pay it. Committed outflows are ₹69,662, 73.3 percent of take-home. Variable spending averaged ₹24,700 over the last three months. What is left is ₹638. The new EMI makes the month ₹14,362 short, and the only line large enough to absorb that is the SIP.

A useful answer does not stop at no. The debits also show which commitments are about to end, because they record how many payments have been made when the history reaches back to the first one. The phone's card EMI has five payments left and the AC loan eight. When both finish, ₹8,915 a month comes back and the surplus becomes ₹9,553. Even then, a ₹15,000 EMI is ₹5,447 too large.

Take it now anyway and the shortfall is ₹14,362 a month for five months, ₹8,947 for the next three and ₹5,447 for the remaining 28. Over 36 months that is ₹2,51,167 that has to come from somewhere: the SIP stopped, or a card balance carried at card rates, which is the expensive version of the same loan.

The honest alternatives all start in month nine, once the two short EMIs have gone.

Affordability is one question. The useful ones share a shape: they are about your own ledger, they have an answer in rupees, and the answer ends in something you can do. A general chatbot can answer none of them unless you type the ledger in, and an assistant that reads transactions can answer them only for the accounts it reads.

Recording those transactions without typing, and a short monthly review built on three questions, is set out on the page on how to track your money. Why most of what a household loses comes from decisions made once, rather than from everyday spending, is argued on the what-is-Unyfy page.

Four limits apply to every AI money assistant, including the one this site makes.

It is not a SEBI-registered adviser. It can tell you what your SIPs cost and whether an EMI fits the month; it should not be picking funds or shares, and a suggestion from an app that earns commission on it deserves a second look.

It cannot see what it does not read. Cash shows up as a withdrawal. A card from a second bank whose alerts it never receives, a spouse's account, a loan paid from another account: all invisible. Leave the phone's card EMI out of the worked household and FOIR with the new loan reads 34.8 percent instead of 40.5, and the surplus reads ₹6,053 instead of ₹638. The error points towards yes.

It can be wrong about what it does see. The same debit can arrive twice, once as an SMS alert and once as an email; count the car EMI twice and FOIR reads 53.8 percent, a no where the honest answer was a maybe. A transfer to yourself can be read as spending, a refund as income, a UPI handle matched to the wrong merchant.

And it answers the question asked. 'Can I afford it' is not 'should I buy it'; the second one stays with you.

The rule for any chatbot, any AI assistant and any caller claiming to be from your bank is the same. Never type or paste an OTP, a UPI PIN, an ATM PIN, a card's CVV, a full card number with its expiry, a net banking or email password, or the answers to your bank's security questions. Your bank does not need them, and no assistant does.

Be careful with uploads too. A bank statement carries your account number, your address and the name of everyone you paid. Uploading one to a general chatbot to 'analyse my spending' hands all of that to a service whose data terms you have probably not read. If you do it, redact the account number first, and check whether conversations are kept, used for training, and deletable.

For an app that reads your transactions, ask three things before connecting it: what exactly it reads, whether that is SMS, email alerts, statements or a consented account aggregator feed; whether it can move money; and how you revoke its access and delete your data yourself.

Unyfy is an AI you can ask about your own money, and it answers from transactions it has already read and categorised rather than from averages. Two capabilities carry the question on this page. On Pro, the Fixed Expenses screen predicts what the coming month is already committed to: the EMIs, SIPs, rent, bills, subscriptions and card bill on a cycle. And it computes a live FOIR, and a blended rate across your loans and cards.

What you see is next month's committed outflows, each named from the lender or merchant on the debit with its amount, beside a FOIR worked from the EMIs actually debited, so a ₹15,000 question starts from what is committed, not from what you remember. It reads bank and card transaction emails and, on Android, transactional SMS, and counts a debit seen through both channels once. It never asks for your bank password or UPI PIN.

Unyfy is not a SEBI-registered adviser and gives no investment advice. The diagnosis is free; Ask Unyfy is free to try, with unlimited questions and voice on Pro at Rs 99 a month. Install Unyfy on Android, or use the web app at app.unyfy.co.in on an iPhone.