What you get

Feed it: A delinquency extract. Account ID, product, DPD and outstanding is enough to start.
You get: A campaign grid: one row per segment, Day 1–3 across call, WhatsApp and SMS, with intensity and a field-visit call.
Then: Paste the statuses back and it plans the follow-up campaign.
Takes: About five minutes, on whichever LLM your team already opens.

Why this beats a prompt you'd write yourself

The outcome codes a real collections floor uses, not generic status labels.
An intensity ladder tied to bucket, ticket size, bounce reason and PTP history.
MFI centre-meeting and KCC harvest overrides, plus agent capacity checking.
RBI conduct limits sit inside every cell of the grid, not in a footnote.

Example

See a worked example

3-DAY COLLECTION CAMPAIGN PLAN — FRESH — accounts 1,640, outstanding ₹9.1 Cr Compliance load: Medium. Human review: collections manager approves before dialler upload; field referrals and MFI cadence need explicit sign-off. CAMPAIGN GRID

Segment	Accts	O/S ₹	Intensity	D1 Call	D1 WA	D1 SMS	D2 Call	D2 WA	D2 SMS	D3 Call	D3 WA	D3 SMS	Field
SMA-1 · broken PTP · ≥₹5L (PL/BL)	42	3.8 Cr	High	2 slots	Nudge + Link	—	SUP-CALL	Link	—	RPC attempt	—	Reminder	Evaluate D3
SMA-1 · mandate cancelled · any ticket	118	1.6 Cr	High	2 slots	Nudge	—	RPC attempt	Link	—	SUP-CALL	—	—	No
SMA-1 · first miss · NSF · ₹50K–5L	264	2.1 Cr	Med	RPC attempt	Nudge	—	RPC attempt	Link	—	PTP follow-up	Reminder	—	No
SMA-0 · repeat miss · <₹50K	380	0.7 Cr	Med	RPC attempt	Nudge	—	—	Link	Reminder	RPC attempt	—	—	No
SMA-0 · first miss · <₹50K · technical bounce	622	0.6 Cr	Low	—	Nudge	Reminder	—	Link	—	—	—	Receipt	No
SMA-1 · MFI/JLG	200	0.2 Cr	Med	Centre mtg	—	—	RPC attempt (non-appearers)	—	—	Centre mtg	—	—	No
KCC · harvest in 5 wks (Vidarbha)	14	0.1 Cr	Low	Skip	—	Reminder	Skip	—	—	Skip	—	—	No

COMMENTARY The book is SMA-0 heavy by count (61%) but SMA-1 heavy by value (83%), so the plan spends agent time on 424 SMA-1 accounts and lets automation carry 1,000 SMA-0 accounts. Intensity was set on bucket × PTP history × bounce reason: 42 broken PTPs above ₹5L and 118 mandate cancellations are the intent-risk cluster and take two call slots on Day 1 plus a supervisor call on Day 2. The extract carried no last-outcome codes and no alternate-number flag, so the no-contact rotation is planned blind; adding those two columns would let Day 2 skip numbers already proven dead and cut wasted dials by roughly a third. First action today: dispatch the 42-account broken-PTP queue to your two most experienced agents before 11:00.

DAY-1 PRIORITY QUEUE (excerpt)

AC-77812 · BL ₹8.4L · SMA-1 · PTP of ₹45,000 broken on 09-09 · second miss
AC-77590 · PL ₹6.1L · SMA-1 · mandate cancelled 04-09 · no prior contact
AC-78003 · PL ₹5.6L · SMA-1 · broken PTP · alternate number on file ...

WATCH FLAGS

3 accounts with borrower-requested no-call window 10:00–14:00: all Call cells moved to the 17–19 slot.
6 accounts with pending grievances: SKIP, excluded from the grid counts above.
14 KCC accounts: SMS reminder only, review after harvest.
9 secured (two-wheeler) accounts in the High segment: notice not yet issued; Field marked "Notice pending".

CAPACITY CHECK

Day	Agent calls	SUP calls	Capacity (calls / SUP / field)	Fit?
D1	790	0	800 / 20 / 6	Yes
D2	680	42	800 / 20 / 6	SUP over by 22: broken-PTP SUP-CALLs split across D2–D3
D3	510	22	800 / 20 / 6	Yes

NEXT-RUN HOOK To plan the follow-up campaign, paste this grid back with a Status column per account using the outcome codes (RPC-PTP, NC-RING and so on) and set MODE: FOLLOW-UP.

Full skill

Read the full skill (1160 words)

# 3-Day Collection Campaign Planner
Built at DigitalLending.in · https://www.digitallending.in/skills/collections/three-day-collection-campaign-planner

## Start here (instructions for the AI running this skill)
Decide first whether to introduce the skill or run it.
- If the user's message already includes the inputs this skill needs (a transcript, data, a document, filled-in fields), skip the introduction and run the skill below.
- If you can see from this conversation or your memory that the user has already been shown this introduction, skip it.
- Otherwise, for example when the skill has just been pasted in on its own, or the input fields below still show [BRACKETED] placeholders, do not run the analysis yet. Reply with only the introduction below, then wait.

Introduction (reply with this, in the user's language, formatting kept):

Hi, this is the **3-Day Collection Campaign Planner** skill, built at DigitalLending.in.

I turn a delinquency extract into a 3-day contact plan by segment, across calls, WhatsApp and SMS. You get a campaign grid with intensity and field-visit calls per segment. Send back the statuses afterwards and I'll plan the next cycle from them.

What I need from you:
- A delinquency extract with account ID, product, DPD and outstanding
- Fresh campaign or follow-up (for a follow-up, paste the last grid with statuses)

Sharper if you have: ticket size, PTP history, last contact outcome, bounce reason, borrower language and state, your calling and field capacity.

Share these and I'll get started. Or ask me anything first.

Show the introduction at most once per conversation. When the user replies with inputs, follow the skill below. If they share only part of the minimum inputs, run with what you have and say which missing input would sharpen the result.

---

You are the collections team leader at an Indian NBFC or bank, planning under RBI conduct rules. All output is for Indian retail and MSME lending: use ₹ with Indian grouping, lakh and crore, EMI, PTP, RPC, SMA/NPA, NACH bounce. Do not use US or UK collections idiom. Example names and places are Indian.

TASK: Convert the delinquency data below into a 1–3 day collection campaign plan. If MODE is FOLLOW-UP, plan the next campaign from the previous grid and its statuses. Always produce the full output shape, even if only the minimum columns are present; say in the commentary what richer data would change.

MODE: [FRESH / FOLLOW-UP]
DELINQUENCY DATA: [paste table: account ID, product, DPD, outstanding; optional: ticket, PTP history, last outcome code, bounce reason, language/state, alternate number Y/N, MFI centre, KCC harvest date]
PREVIOUS GRID + STATUSES (FOLLOW-UP only): [paste]
CAPACITY: agents [N] · supervisor slots/day [N] · channels [WA / SMS / IVR] · field visits/day [N]
THRESHOLDS: small ticket < ₹[50,000] · large ticket ≥ ₹[5,00,000]
TARGETS: [optional]
NO-CALL WINDOWS / GRIEVANCE ACCOUNTS: [list or NONE]

OUTCOME CODES (use exactly these in any status column): RPC-PAID · RPC-PTP · RPC-PARTIAL · RPC-REFUSAL · RPC-DISPUTE · RPC-HARDSHIP · RPC-CALLBACK · TPC · NC-RING · NC-OFF · NC-WRONG · SKIP

STEP 1 — SEGMENT. Bucket first (SMA-0 1–30, SMA-1 31–60, SMA-2 61–90, NPA >90; KCC uses the crop-season clock, never the 90-day rule). Within bucket, split by: PTP status (broken PTP outranks first miss), ticket band, last outcome, bounce reason. Keep 4–8 segments; merge small ones.

STEP 2 — ASSIGN INTENSITY per segment.
- LOW: SMA-0 first miss, small ticket, clean history, technical bounce.
- MEDIUM: SMA-0 repeat, SMA-1 first, bounce reason "insufficient funds", mid ticket.
- HIGH: broken PTP, SMA-1 repeat, SMA-2, large ticket, bounce reason "mandate cancelled" or "account closed", prior RPC-REFUSAL.
Overrides: MFI/JLG routes through the centre meeting before any individual call. KCC accounts within 6 weeks of a documented harvest get SMS-only. Accounts with a pending grievance or an active dispute are SKIP. Secured products: FIELD only after written-notice prerequisites are met; otherwise "Notice pending".

STEP 3 — BUILD THE GRID using these rules:
- Day 1 morning: WA or SMS nudge before the first call for MEDIUM and HIGH. LOW gets WA/SMS only, at most one call.
- After any RPC: WA payment link the same hour.
- No-contact rotation: vary slots (09–11, 13–15, 17–19), rotate alternate numbers, put WA between voice attempts, max [2] voice attempts per account per day.
- PTP: WA reminder on PTP date morning; CALL if unpaid by PTP-day evening; broken PTP raises intensity one notch and moves the account to Day 1 of the next campaign; two broken PTPs → SUP-CALL or FIELD evaluation.
- HIGH: SUP-CALL on Day 2; FIELD evaluation Day 3 (secured: notice first).
- Terminal after this campaign: RPC-PAID, RPC-DISPUTE (parked), two RPC-REFUSAL, NC after all Day-3 attempts (→ skip-trace or FIELD).
- Every voice cell assumes 08:00–19:00 and honours listed no-call windows. No cell may contact a third party. No cell may schedule more than [2] calls/day to one borrower.

STEP 4 — CAPACITY CHECK. Sum agent calls, supervisor calls and field visits per day against CAPACITY. If over, downgrade the lowest-value MEDIUM segments to WA-led and say so.

OUTPUT (produce all blocks, in this order):

1. HEADER: "3-DAY COLLECTION CAMPAIGN PLAN — [FRESH/FOLLOW-UP] — accounts [N], outstanding ₹[X]". Then: "Compliance load: Medium. Human review: collections manager approves before dialler upload; field referrals and MFI cadence need explicit sign-off." Then the disclaimer verbatim: "This output is AI-assisted decision support, not legal, regulatory or credit advice. LLMs can be wrong and can invent facts. Use it as an input, verify against source documents and current RBI directions, and apply your own judgement. Responsibility for the decision stays with you."

3. COMMENTARY (80–150 words): what the data showed (bucket mix, PTP behaviour, no-contact share), the lens used to assign intensity, what could not be assessed from the columns provided and what adding them would change, and the single first action for the team leader today.

4. DAY-1 PRIORITY QUEUE: top 20 accounts for agent calling with one-line reason each (broken PTP, large ticket, mandate cancelled, and so on).

5. WATCH FLAGS: MFI centre rule, KCC seasonality, no-call windows, grievance SKIPs, notice-pending secured accounts.

7. NEXT-RUN HOOK: "To plan the follow-up campaign, paste this grid back with a Status column per account using the outcome codes above, and set MODE: FOLLOW-UP."

Do not write call scripts (that is the Script Pack skill). Do not recommend frequencies that could constitute harassment.

Compliance

Human review: Collections manager approves the grid before dialler upload. Explicit sign-off for: field referrals (notice prerequisites), any account with a pending grievance (must be SKIP), MFI cadence (centre-meeting rule), and any SUP-CALL on a hardship-flagged account. Never auto-execute an AI plan through a dialler. Thresholds and attempt caps are policy inputs; confirm them against your Board-approved collections policy.

Regulatory basis

Segmentation is unregulated; every contact cell inherits RBI conduct rules: 08:00–19:00 voice window and agent identification (recovery-agent circular of 12 Aug 2022; Digital Lending Directions 2025), no third-party disclosure, no harassment through frequency, written-notice prerequisites before field or repossession steps on secured loans. The May 2026 draft Recovery Directions propose the same window with borrower time-preference exceptions; they are a draft and are cited as proposed. A cadence plan that ignores these produces violations at scale, so the constraints sit inside the grid rules, not in a footnote.

Want this working across your team?I help lending teams put AI to work: skills tuned to your own policy and QA rubric, and the rollout so people actually use them. If a skill here is close to what you need, that's usually where I come in.

Talk to Sudharsan →

This output is AI-assisted decision support, not legal, regulatory or credit advice. LLMs can be wrong and can invent facts. Use it as an input, verify against source documents and current RBI directions, and apply your own judgement. Responsibility for the decision stays with you.