Email & SMS Campaign Agent AI

Invoked when: a lifecycle trigger is received from the Lifecycle Campaign Manager AI, or a scheduled batch campaign send time is reached

• –Maintains the trigger library — the set of borrower events that fire automated campaign sends: disbursement (welcome email), first EMI (repayment confirmation SMS), EMI bounce (payment reminder email within 2 hours), loan closure (thank-you email with NPS survey link), and NPS survey completion (acknowledgment within 24 hours). Each trigger has a pre-approved campaign mapped to it; the agent matches the incoming event to the correct campaign and dispatches in under 15 minutes.
• –Manages campaign frequency capping — no borrower receives more than 3 commercial email campaigns in a 7-day period, and no borrower receives more than 2 promotional SMS messages per week, regardless of how many trigger events fire in that window. Transactional messages (EMI confirmation, payment receipt, disbursement confirmation) are excluded from frequency caps — these are service communications, not commercial campaigns.
• –Tracks delivery metrics per send: delivery rate, open rate (for email), click rate, and bounce categorisation (hard bounce — invalid address, permanent suppression required; soft bounce — temporary failure, retry after 24 hours). Hard bounces are added to the permanent suppression list within the same dispatch cycle.

Invoked when: a campaign has two approved variants loaded — A/B test begins with the next scheduled or triggered send for that campaign

• –Runs A/B tests across subject lines (the primary determinant of email open rate), send timing (the day of week and hour that maximises open rates for the target segment), and CTA text (the specific wording of the primary call-to-action). Tests one variable at a time — simultaneous multi-variable tests cannot isolate which variable drove the performance difference. Each test is pre-specified with the primary metric and minimum detectable effect before the test begins.
• –Allocates traffic 50/50 between variants A and B for the duration of the test. Where the send list for a campaign is smaller than the minimum sample required to reach 95% significance with the expected effect size, the agent flags the test as underpowered and accumulates data across multiple send cycles before declaring a result.
• –Declares the winning variant at 95% statistical significance and updates the campaign library. Test results are logged in the campaign knowledge base — the accumulated record of which subject line approaches, timing windows, and CTA framings have outperformed for each segment. This knowledge base informs the marketing team's content decisions for new campaigns.

Invoked when: any outbound SMS or commercial email is about to be dispatched — compliance check runs before the send is released to the gateway

• –Validates every outbound SMS against the TRAI DND registry before dispatch. The agent correctly categorises each SMS type before the DND check — promotional SMS (commercial messages, offers, campaigns) may not be sent to DND-registered numbers; transactional SMS (disbursement confirmation, EMI receipt, OTP) are exempt. A misclassification that sends a promotional SMS to a DND-registered number is a TRAI compliance violation regardless of the borrower's prior consent to marketing.
• –Validates every outbound email and SMS against the borrower's DPDP consent record — confirming that the borrower has consented to receive marketing communications through the channel being used. Where consent is absent or has been withdrawn, the send is suppressed for that channel. The agent does not substitute an alternative channel — if a borrower has consented to email but not SMS, an SMS campaign is suppressed; the email send proceeds.
• –Verifies that every SMS uses a TRAI-registered sender ID and a DLT-registered message template before dispatch. Unregistered sender IDs and templates are flagged for the technical team to resolve before the send is released. The agent checks registration status at send time, not at campaign setup time — a registration that lapses after setup is caught at dispatch.