Onboarding Drop-Off Agent AI

Invoked when: a borrower's journey session ends without completing the current step, and a configurable inactivity threshold is crossed without resumption

• –Reads the journey analytics events for each in-progress application — session end without step completion, time-on-step exceeding the configured threshold before exit, and application not resumed within the configured re-engagement window — and classifies the abandonment by step: KYC step, document upload step, liveness check step, income verification step, consent step, or final submission. The step classification determines which re-engagement template is sent, because the friction at each step is different and the message must address the actual friction.
• –Distinguishes active abandonment (the borrower navigated away or closed the app) from passive stall (the borrower is still on the page but has not progressed for a long time — possibly struggling with a step rather than having abandoned). Passive stalls get an in-app contextual help prompt rather than an outbound re-engagement message, because the borrower is still present and an outbound WhatsApp while they are actively struggling with the app creates confusion.
• –Logs every abandonment event with the step, the session duration at that step (how long did they try before leaving), and the device type — because long session duration before abandonment suggests the borrower was actively trying but could not complete the step, whereas short session duration suggests they looked at the step and decided not to proceed. Different diagnoses require different responses.

Invoked after each configured wait period following detected abandonment

• –Attempt 1 (4 hours after abandonment): A step-specific message that names the step where the borrower stopped and offers one concrete piece of help — for a liveness check abandonment: "We noticed your video check didn't complete. Here are three things that help it work: good lighting, a plain background, and holding the camera at eye level." For a document upload abandonment: "You need your last 3 months' bank statement. You can download it from your bank's app under Statements, or link your account directly instead — no upload needed." The message is specific, practical, and immediately actionable.
• –Attempt 2 (24 hours after abandonment): Offers an alternative pathway. For borrowers who struggled with the digital self-serve journey, this message introduces the assisted option — a link to schedule a call with an Account Executive who will guide them through the completion process. The alternative pathway removes the friction entirely rather than trying to help the borrower through it.
• –Attempt 3 (72 hours after abandonment): A final re-engagement with a clear "we'll keep your application for 7 days" framing, providing the deep link back to exactly the step where they stopped. After this, if there is no response, the application is flagged for human review of whether to close or retain, and the abandonment is logged as final with the reason code "no response to 3 re-engagement attempts."

Invoked when: a borrower has been on a specific step for longer than the configured time threshold without progressing — indicating active struggle rather than abandonment

• –Triggers an in-app contextual help overlay — not an outbound message, but a prompt within the journey itself — that offers step-specific guidance: at the liveness check step, instructions for optimal lighting and camera positioning; at the AA consent step, a plain-language explanation of what Account Aggregator consent means and what data will be accessed; at the income document step, the list of acceptable documents with instructions for downloading each from common bank apps.
• –Where the in-app help does not resolve the stall — the borrower continues on the step without progressing after the help is shown — escalates to an offer of live chat or a callback with an Account Executive AI or human RM. The escalation is an offer, not an automatic route — the borrower selects whether they want help.
• –Logs every step assist event with the step, the help content shown, and whether the borrower progressed after seeing the help or subsequently abandoned. This outcome data tells the CX team which help content is effective and which is not.