Personalisation Agent AI

Invoked when: a campaign is dispatched — product recommendation is computed fresh for each borrower in the send list

• –Evaluates each borrower against 30 signals spanning repayment behaviour (EMI track record, DPD history, bounce frequency), product holding profile (current loan type, tenure stage, outstanding balance), bureau trajectory (score change since origination), app and digital engagement (login frequency, feature usage), and campaign response history (which product categories they have clicked on in prior communications). The 30 signals are weighted by predictive importance for each product category — the weight of "tenure stage" is higher for top-up recommendations than for gold loan recommendations, where asset ownership signals are more predictive.
• –Matches each borrower to the product recommendation most consistent with their signal profile: top-up on existing loan (for performing borrowers at or past the 6-EMI milestone), personal loan for a new purpose (for borrowers whose engagement signals suggest a life event — salary credit increase, new dependent indicator from co-applicant data), gold loan (for borrowers in geographies with high gold loan penetration who do not already hold one), or business loan (for borrowers whose salary credits have been replaced by irregular business-type credits). The recommendation is specific — not "you might like a loan" but "a ₹1.5 lakh top-up at your current rate, with 12 EMIs".
• –Flags low-confidence recommendations — cases where the signal data is insufficient or contradictory to produce a reliable product match — and falls back to the segment-level default recommendation for those borrowers rather than surfacing a personalised recommendation that could be wrong. A confident wrong recommendation (a gold loan offer to a borrower who lives in a city apartment and has no asset signals) damages trust more than a generic segment-level message.

Invoked when: a campaign message is assembled for dispatch — dynamic content slots are populated from the borrower's signal profile

• –Populates dynamic content slots in campaign templates at send time — subject lines, opening lines, product name, offer amount, rate, and CTA text — with the values most relevant to each borrower. An email campaign for a top-up offer sends a different subject line to a borrower who has consistently paid early ("Your excellent repayment record unlocks a new offer") versus one who pays on time but at the last day of the grace period ("A new offer based on your loan performance") — the same campaign, two different subject lines calibrated to two different repayment profiles. Personalised subject lines increase open rates significantly in financial services email, and open rate is the prerequisite for all downstream conversion.
• –Adjusts message framing by borrower segment and life stage: MSME borrowers receive business-need framing ("working capital for your next season"); retail borrowers receive personal-need framing ("funds for what matters next"); borrowers approaching loan closure receive renewal framing ("your next chapter with LendingIQ"). Segment framing is applied at the campaign template level; individual signal data adjusts the specific values within the frame — the agent does not generate free-form message copy, it selects and populates from pre-approved content building blocks.
• –Selects the CTA variant most consistent with the borrower's digital engagement profile: borrowers with high app engagement receive an in-app deep link CTA; borrowers who consistently engage via WhatsApp receive a WhatsApp reply CTA; borrowers with low digital engagement receive a call-back request CTA routing to a relationship manager. The CTA is the conversion mechanism — sending a digital-first CTA to a borrower who has never clicked a digital link is wasted personalisation effort.

Invoked when: a borrower is confirmed eligible by the Pre-Approval Offer AI — personalisation layer adds the framing and channel match to the offer parameters

• –Receives the offer parameters generated by the Pre-Approval Offer AI (amount, rate, tenure, product) and adds the personalisation layer: which specific aspect of the offer is most likely to be the borrower's primary decision driver. For borrowers whose signal profile shows rate sensitivity (who compared rates at origination, who engaged with rate-related content), the offer is framed rate-first. For borrowers whose signal profile shows amount sensitivity (who applied for the maximum available at origination), the offer is framed amount-first. For borrowers who completed quickly at origination (low friction tolerance), the offer is framed simplicity-first ("3 steps, same-day"). The offer parameters are fixed by the credit workflow; the personalisation layer determines how to present them.
• –Tracks offer matching conversion rates by signal profile — which combinations of borrower signals produce the highest acceptance rates, and which personalisation framings outperform the segment baseline. This feedback loop refines the signal weights in the recommendation engine over time, improving offer matching accuracy as more acceptance and rejection data accumulates. Where a signal combination consistently underperforms, the agent flags it for data science review rather than continuing to apply a failing model.