Use case #0002

Channel selection: how Propensity AI picks WhatsApp, call or field for each account

A propensity score of 78 tells the collections system this account is collectible today. The channel selection model then asks: collectible by what means? A borrower who answered every previous call but has never responded to a WhatsApp message should receive a voice call today. A borrower who pays immediately after every payment link but rarely picks up the phone should receive a WhatsApp today. The propensity score and the channel selection model work together — one tells you who to contact, the other tells you how.

A propensity score of 78 tells the collections system this account is collectible today. The channel selection model then asks: collectible by what means? A borrower who answered every previous call but has never responded to a WhatsApp message should receive a voice call today. A borrower who pays immediately after every payment link but rarely picks up the phone should receive a WhatsApp today. The propensity score and the channel selection model work together — one tells you who to contact, the other tells you how.

Why channel selection is a separate decision from account prioritisation

Account prioritisation answers the question: who do we contact today? Channel selection answers: how? These are related but distinct decisions. A borrower with a propensity score of 82 who receives their contact through the wrong channel produces the same outcome as a borrower with a score of 42 who is contacted through the right channel — no payment. The two decisions together determine the contact economics of a collections operation.

The Propensity Scoring Agent AI resolves both decisions simultaneously for every account in the DPD book: here is the ranked list of accounts to contact today, and here is the recommended channel for each. The collections manager sees a queue that is ordered by propensity and pre-populated with channel decisions — no additional judgement required at the contact level, only at the exception level.

"The right borrower contacted through the wrong channel is a missed opportunity. The Propensity AI scores both decisions simultaneously — who to contact and how — so the team executes, not decides."

The channel decision logic — propensity score × engagement history × preference signals

Score 75–100
High
Propensity
Priority contact — match channel to engagement history

This account is collectible today — use the channel that produced the most engagement historically

High-propensity accounts are the primary focus of the day's collections activity. The channel is selected from the borrower's engagement history: if they have answered the last 3 calls, call. If they have clicked the last 2 WhatsApp payment links, send WhatsApp. If prior contacts have been ineffective and the propensity is driven by a salary credit or legal notice event, call — these are urgency-driven contacts that require conversation, not asynchronous messaging.

Recommended: Voice call if call pick-up >60% · WhatsApp if link click history is stronger · IVR if confirmed payment commitment pending
Score 50–74
Moderate
Propensity
Engaged-but-uncertain — WhatsApp preferred, voice as escalation

Borrower is engaged but payment timing is uncertain — asynchronous contact first

Moderate-propensity accounts have some positive signals (partial payments, engagement) but also uncertainty (broken PTPs, moderate pick-up rate). WhatsApp is the preferred first contact — it is lower-friction, allows the borrower to respond on their schedule, and a link click at this propensity level is a strong indicator to escalate to a voice call the same day. Voice is the escalation, not the first contact.

Recommended: WhatsApp first → if read but no response in 4 hours → Voice escalation
Score 30–49
Low
Propensity
Disengaging — hold for improving signal, or IVR/SMS touch

Borrower is disengaging — voice calls are low-return, a light-touch channel preserves the relationship

Low-propensity accounts have declining engagement signals — falling call pick-up rate, unread WhatsApp messages, no recent payments. Assigning a voice call to this account costs the same as calling a high-propensity account and produces a fraction of the return. A light-touch IVR or SMS contact maintains presence in the borrower's awareness without consuming agent time. The account is re-evaluated daily — if the propensity improves (salary credit, link click), it moves back up the queue.

Recommended: IVR automated touch or SMS reminder · Daily propensity re-check
Score <30
Non-
Responsive
Silent accounts — field visit or legal escalation assessment

Digital channels exhausted — field contact or legal action is the next productive intervention

Very low propensity accounts have stopped responding to all digital channels. Continuing to call and message depletes collections resources for no return. For secured accounts, a field visit (with prior written notice as required by FPC) is the appropriate next intervention. For accounts where the legal threshold has been reached, the Hard Bucket Agent AI takes over the account for formal recovery proceedings. The Propensity AI's role is complete when digital collectibility has been exhausted.

Recommended: Field visit (secured, with notice) or → Hard Bucket AI for formal recovery

The live prioritised queue: what the collections team sees each morning

Collections Priority Queue — DPD 0–90 Book
Nov 14, 2025 · 08:00 · 2,841 accounts · Ordered by propensity score
847Score 75+
Contact today
1,124Score 50–74
WA then call
612Score 30–49
IVR touch
258Score <30
Field or legal
2,841Total accounts
DPD book
Top 7 accounts — ranked by propensity score
#1 Priya Ramachandran
94 Voice · 11 AM DPD 12
#2 Suresh Pillai
88 WhatsApp DPD 8
#3 Suresh Kumar Mehta
78 Voice · 12 PM DPD 58
#4 Meera Krishnan
74 WhatsApp DPD 34
#5 Anand Singh
68 WhatsApp DPD 22
#1,441 Rajesh Nambiar
44 IVR touch DPD 74
#2,710 Vinod Kumar
18 Field visit DPD 88
847Accounts in the top-priority contact tier today — score 75+ across the 2,841-account DPD book
2-decisionQueue presents both the ranked account list and the channel decision — team executes, not decides
4hrsWhatsApp-read-no-response trigger for voice escalation — same-day contact maximises salary window
DailyQueue reordered every morning — yesterday's #1 may be #47 today if their circumstances changed

The oldest account is not the most urgent — the most collectible account is

A DPD 74 account with a propensity score of 44 has not paid in over 70 days and has stopped engaging with contacts. A DPD 22 account with a propensity score of 78 received a salary credit yesterday and clicked a payment link this morning. Every collections instinct says call the older account first. Every data point says the opposite. The Propensity Scoring Agent AI reorders the collections queue every morning based on where today's contact has the highest probability of producing a payment. It does not argue with DPD-order instinct — it replaces it with evidence.

← Back to Propensity Scoring Agent AI