A product recommendation is only as good as the model behind it. An institution that recommends a personal loan to an MSME proprietor, or a home loan top-up to a borrower whose last EMI bounced, has not personalised the offer — it has labelled a generic offer with the borrower's name. True personalisation requires a model that reads the borrower's current financial position, behavioural signals, lifecycle stage, and product eligibility simultaneously, and surfaces the one product that fits all four. The Personalisation Agent AI builds and continuously updates a 30-signal profile for every borrower in the portfolio, and scores every eligible product against that profile before any offer is generated.
Why 30 signals — and why fewer is not enough
A recommendation built on 5 signals — say, income, current product, CIBIL score, DPD history, and loan vintage — will be correct for the majority of borrowers whose profile is straightforward. It will be wrong for the borrower whose income is flat but whose property has appreciated significantly, creating a LAP opportunity the income signal alone would miss. It will be wrong for the MSME borrower whose GST revenue growth makes them a working capital top-up candidate regardless of their CIBIL score. It will be wrong for the salaried borrower whose portal behaviour in the last 72 hours shows active EMI calculation — a real-time intent signal that a static profile cannot capture.
Thirty signals are not arbitrary precision — they represent the minimum set of data dimensions required to distinguish between borrowers who look identical on 5 signals but who have materially different optimal product recommendations. Each additional signal added above 5 increases the recommendation precision incrementally; the relationship between signal count and recommendation accuracy is not linear, and the returns diminish significantly above 30 in the context of a lending portfolio. The 30 signals below represent the empirically productive range for an Indian lending context.
The 30 signals across 6 categories
Category 1 — Repayment behaviour (6 signals)
02DPD history — worst-ever classificationHas the borrower ever been DPD 30+? DPD 60+? NPA? The worst-ever classification is a permanent signal — it limits certain product recommendations regardless of current behaviour.Weight: 8pts
03NACH bounce frequency (last 12M)Number of NACH bounces in the last 12 months, regardless of recovery. Even recovered bounces indicate cash-flow fragility at specific calendar points.Weight: 6pts
04Voluntary prepayment historyHas the borrower made prepayments? How many, and of what total value? Prepayments signal surplus cash and active financial engagement — two strong indicators of expansion appetite.Weight: 5pts
05Outstanding balance as % of originalBelow 60%: strong equity position, eligible for top-up or LAP. Below 40%: approaching maturity, next-product conversation relevant.Weight: 4pts
06EMI-to-income ratio trend (FOIR trajectory)Is the borrower's FOIR improving (income growing faster than obligations) or deteriorating (new obligations appearing)? Trajectory matters more than the current ratio.Weight: 5pts
Category 2 — Income and financial capacity (6 signals)
08Income growth YoY (12-month comparison)Income growing 20%+ YoY unlocks higher product amounts than current income alone would support — the trajectory justifies a forward-looking offer.Weight: 6pts
09Available FOIR headroomThe gap between current FOIR and the product-specific FOIR ceiling. Determines the maximum additional EMI — and therefore the maximum offer amount for each eligible product.Weight: 9pts
10End-of-month balance trend (3M)Is the borrower accumulating cash month-on-month or depleting it? A growing end-of-month balance indicates a borrower building financial buffer — improving capacity confidence.Weight: 4pts
11External EMI obligations (new NACH debits)Any NACH debit that does not belong to this institution — visible in bank statement analysis. Each new external obligation reduces available FOIR headroom and constrains the offer amount.Weight: 6pts
12MSME: GST outward supply (last 4 quarters)For MSME borrowers: the primary income signal. GST outward supply is more reliable than bank statement credits for proprietors who mix personal and business accounts.Weight: 8pts (MSME only)
Category 3 — Asset and collateral signals (5 signals)
14Current LTV (loan-to-value)The ratio of outstanding loan to current property value. As both principal reduces and property appreciates, LTV falls — creating headroom for a top-up or LAP that may not have existed at origination.Weight: 6pts
15Additional unencumbered property (from profile)If the borrower declared additional property at origination (beyond the mortgaged property), this may be available as collateral for a new LAP — a product the existing loan relationship does not touch.Weight: 5pts
16MSME: business equipment / machinery valueFor MSME borrowers: equipment and machinery on the balance sheet may be available as security for an equipment top-up or business term loan not currently covered by the existing facility.Weight: 4pts (MSME only)
17CERSAI charge status on existing securityWhether the existing security has a clear CERSAI charge (confirming the institution's priority claim). Critical for determining whether a top-up can be secured against the same property.Weight: 3pts
Category 4 — Credit bureau signals (4 signals)
19CIBIL score trajectory (vs origination)A borrower whose score has improved 50+ points since origination is now a meaningfully better credit risk — eligible for better rates, larger amounts, or products not available at origination.Weight: 5pts
20Bureau: new enquiries (last 6M)Multiple recent bureau enquiries suggest the borrower is actively shopping for credit. This is a readiness signal — but also a flag that a competing institution may get there first.Weight: 5pts
21Bureau: existing loans with other institutionsLoans at other institutions that appear in the bureau but not in the institution's CBS. Each represents an FOIR obligation that constrains the offer amount, and a relationship to potentially consolidate.Weight: 4pts
Category 5 — Engagement and intent (5 signals)
23Outstanding balance check frequency (14 days)Repeated balance checks suggest active thinking about the loan relationship — either prepayment planning or top-up consideration. Three or more checks in 14 days: elevated signal.Weight: 6pts
24Communication open and click historyHas the borrower opened and clicked product-related communications? Which product categories? The pattern across the last 6 months reveals latent product interest.Weight: 4pts
25Portal pages visited (product section)Which product pages has the borrower visited on the institution's portal or website? A borrower who visited the LAP page twice in the last month is a LAP prospect, not a generic prospect.Weight: 5pts
26WhatsApp / SMS response rate (last 6M)A borrower who consistently responds to institution communications is more likely to respond to a personalised offer. Non-responders need a different channel strategy, not a higher-frequency message.Weight: 3pts
Category 6 — Lifecycle and context (4 signals)
28Income event proximity (salary or GST credit)Confirmed salary or GST credit in the last 72 hours puts the borrower in peak financial optimism. Offers sent within this window have measurably higher acceptance rates across all product types.Weight: 5pts
29MSME: business expansion signals (GST new state, MCA)A new GST state registration or MCA director addition indicates business growth — a business actively expanding is a business that needs capital, often before it knows exactly how much.Weight: 6pts (MSME only)
30Borrower segment — salaried / SE / MSME / agriThe foundational segment classification that determines which signals are weighted most heavily and which product families are eligible. MSME signals dominate for MSME; salary signals dominate for salaried. Segment is not a single signal — it is the weight matrix for all others.Weight: matrix modifier
A live 30-signal profile: Kaveri Constructions · MSME
30 signals do not make the recommendation complex — they make it precise
The institution's relationship manager reviewing Kaveri Constructions' profile does not need to read all 30 signals. They need to know the recommendation (MSME WC top-up, ₹17–20 lakh) and the top 3 reasons (EMI calculator use, TN GST expansion, 34% revenue growth). The 30-signal model does the work of synthesising those reasons from the full picture — including the 22 signals that were checked and found neutral, confirming that there are no hidden disqualifying factors. The recommendation that arrives at the RM's desk is not the output of a keyword match or a single threshold check — it is the output of a complete financial profile assessment that no human analyst could run for 48,000 borrowers every morning.