3 ready-to-build workflows

AI agent workflows for Lending & financial services

Take applications, collect KYC, answer eligibility and chase repayments — with every credit decision gated to a human underwriter.

Each recipe below is expressed only in VegaDūta's real workflow building blocks — Trigger, Agent, Knowledge, Condition, Approval, Output, Tool (MCP), Voice, Device, Loop, Code and Parallel — so it maps 1:1 to something you can assemble in the Workflow designer. Consequential actions always pass a human Approval step.

1. Loan application intake + KYC with a completeness gate

A borrower applies and an officer manually captures details, then emails back and forth for KYC documents, income proof and bank statements over days — incomplete files bounce repeatedly and stall the pipeline.

Trigger: WhatsApp / Webhook (loan enquiry)

How the workflow runs

  1. Trigger

    Application opens. A WhatsApp trigger (or a Webhook from your website/LOS) opens the application in the borrower's own language.

  2. Agent

    Capture the application. An Agent node collects loan type, amount, tenure, income and purpose into a clean, structured record.

  3. Output

    Request exactly the right documents. An Output node asks for the specific KYC / income / bank documents and accepts them right in the chat, guiding on format and clarity.

  4. Condition

    Completeness gate. A Condition node won't advance the file until every mandatory KYC and income document is present and legible.

  5. Approval

    Underwriter makes the credit decision. An Approval node hands a complete, structured file to a human underwriter — the credit decision is never automated. This is the non-negotiable control.

  6. Tool (MCP)

    Write to the LOS / core banking. A Tool (MCP) node records the application and the decision in your loan-origination or core-banking system — reached via MCP, not a native connector.

Channels & connectors

  • WhatsApp
  • Webhook
  • Sarvam (multilingual)
  • Approval / HITL
  • Tool (MCP → LOS/core-banking)

Outcome

Applications arrive complete and structured with KYC in hand; underwriters spend time deciding, not chasing documents — and every credit decision is a logged human approval.

Why it helps

Removes the days of manual document chasing from the front of every loan, while the credit decision stays with a licensed underwriter.

Build spec

1 agent — 1 intake agent; the completeness gate, the underwriter Approval and the LOS write are workflow nodes.

System prompt (paste-ready)

You are a loan-application intake assistant. Capture the application as strict JSON {loan_type, amount, tenure, monthly_income, employment, purpose, applicant_details} in the borrower's language, and request the specific KYC, income and bank-statement documents needed, guiding on format and legibility. You do NOT assess creditworthiness, quote an interest rate, or say whether the loan will be approved — the credit decision is a human underwriter's at the approval step. Never promise approval or a disbursal date. Handle identity and financial documents as confidential.

MCP connectors

  • Loan-origination system / core banking — via a Tool (MCP) node (no first-party connector)
  • WhatsApp — via the messaging channel
  • Website/LOS — inbound via Webhook

Built-in tools

  • document parsing (KYC / income / bank-statement PDFs)
  • knowledge_search (required-document checklist by loan type)

Guardrails

  • Never makes or implies a credit decision — completeness only; the underwriter decides at the Approval node, and every decision is logged
  • The file can't advance until every mandatory KYC and income document is present and legible (Condition gate)
  • KYC and financial documents are handled as confidential — captured to the application, never echoed back or shared

Output & delivery

The agent captures the application and collects documents in-chat → the Condition node blocks anything incomplete → the underwriter makes the credit decision at the Approval node → a Tool node records the application and decision in the LOS/core-banking system via MCP.

2. Pre-approval eligibility Q&A (indicative only, gated)

Prospects ask 'will I qualify, and for how much?' at all hours; officers answer the same eligibility questions by hand, and inconsistent replies create false expectations.

Trigger: WhatsApp / Webhook (eligibility enquiry)

How the workflow runs

  1. Trigger

    Eligibility question arrives. A WhatsApp or Webhook trigger starts the flow in the prospect's own language.

  2. Knowledge

    Ground in the product & policy rules. A Knowledge node answers strictly from your current loan products, rate cards and eligibility criteria — with citations, no guessing.

  3. Agent

    Explain indicative eligibility. An Agent node walks the prospect through the criteria and gives an indicative-only view, clearly flagged as not an offer of credit.

  4. Condition

    Route the ready ones forward. A Condition node identifies prospects who look eligible and want to proceed to a real application.

  5. Approval

    Officer confirms any pre-approval. An Approval node routes anything that looks like a firm pre-approval to a human officer — no credit commitment is made by the agent.

  6. Output

    Hand off or invite to apply. An Output node either books an officer callback or sends the borrower into the full application flow.

Channels & connectors

  • WhatsApp
  • Webhook
  • Sarvam (22+ Indian languages)
  • Knowledge base
  • Approval / HITL

Outcome

Prospects get accurate, in-language eligibility answers instantly; officers spend time only on those ready to apply, and no credit commitment is ever made without a human.

Why it helps

Consistent, grounded answers replace ad-hoc manual replies that set false expectations, while any actual pre-approval stays a human decision.

Build spec

1 agent — 1 eligibility agent; the ready-to-proceed Condition, the officer Approval and the hand-off are workflow nodes.

System prompt (paste-ready)

You are a loan eligibility assistant. Answer 'will I qualify and for how much' questions strictly from the current products, rate cards and eligibility criteria in Knowledge, with citations. Walk the prospect through the criteria and give an INDICATIVE view only, stated clearly as 'not an offer of credit and subject to verification and approval'. Never confirm a pre-approval, commit an amount or rate, or say they are approved — anything that looks like a firm pre-approval is routed to a human officer. If the rules don't cover their case, say so and offer a callback.

MCP connectors

  • WhatsApp — via the messaging channel
  • Website — inbound via Webhook

Built-in tools

  • knowledge_search (loan products + rate cards + eligibility criteria)

Guardrails

  • Indicative only: every eligibility answer is flagged as not an offer of credit and subject to approval — the agent makes no credit commitment
  • Anything resembling a firm pre-approval is routed to a human officer at the Approval node
  • Grounded-only: answers come strictly from the current products/rate-cards/criteria with a citation, never an assumed number

Output & delivery

The agent explains indicative eligibility grounded in the product/policy Knowledge → the Condition node identifies prospects ready to proceed → any firm pre-approval is confirmed by an officer at the Approval node → an Output node books an officer callback or sends the prospect into the full application flow.

3. EMI reminders & soft collections outreach

Upcoming and overdue EMIs are chased manually and inconsistently — some borrowers never get a reminder, cash sits uncollected, and the tone is the same whether they're one day or ninety days late.

Trigger: Webhook (scheduled, daily repayment run)

How the workflow runs

  1. Trigger

    Repayment run. A scheduled Webhook trigger evaluates upcoming and overdue EMIs each morning (schedule fed from your LOS/core-banking via Webhook).

  2. Loop

    Per due / overdue account. A Loop node iterates the list so each borrower gets a tailored message.

  3. Agent

    Write a tone-matched nudge. An Agent node tailors tone to how overdue it is — a gentle upcoming-due reminder versus a firmer, still-respectful soft-collections message — and includes the exact amount and a Razorpay UPI pay link.

  4. Output

    Reach out on their channel. An Output node sends the reminder on WhatsApp/SMS/Email in the borrower's language and offers to pay in one tap.

  5. Condition

    Escalate the persistent ones. A Condition node flags accounts past a delinquency threshold for human handling — hard collections and any restructuring are never automated.

  6. Approval

    Collections officer takes over. An Approval node routes escalated, seriously delinquent accounts to a human officer for the call and any decision on the account.

Channels & connectors

  • Webhook
  • Loop
  • WhatsApp
  • SMS
  • Email
  • Razorpay UPI
  • Approval / HITL

Outcome

Every due and overdue EMI gets a timely, tone-appropriate nudge with a one-tap pay link; only the seriously delinquent accounts reach a human, and no account action is automated.

Why it helps

Consistent, respectful follow-up is what pulls repayments in — the manual process skips exactly the accounts that need it, while any real account decision stays with a collections officer.

Build spec

1 agent — 1 messaging agent per account; the per-account Loop, the escalation Condition and the collections-officer Approval are workflow nodes.

System prompt (paste-ready)

You are an EMI-reminder assistant. For each account {borrower, emi_amount, due_date, days_overdue, language}, write a message whose tone matches the stage — a gentle upcoming-due reminder, or a firmer but always respectful soft-collections note — stating the exact amount and including a Razorpay UPI pay link. Follow fair-collections conduct: never threaten, harass, shame, or contact at unreasonable hours, and never disclose the debt to anyone but the borrower. You do NOT restructure, waive, settle, or make any decision on the account — seriously delinquent accounts go to a human officer.

MCP connectors

  • Loan-origination system / core banking — repayment schedule fed inbound via Webhook (no first-party connector)
  • Razorpay UPI — pay-link for one-tap repayment
  • WhatsApp + SMS + Email — via the messaging channels

Built-in tools

  • knowledge_search (collections communication + conduct policy)

Guardrails

  • Fair-conduct: respectful tone only, no threats/harassment/shaming, no unreasonable-hour contact, and the debt is disclosed to no one but the borrower
  • Hard collections, restructuring, settlement and any account action are never automated — seriously delinquent accounts escalate to the collections officer at the Approval node
  • Payment is only ever collected through the Razorpay UPI link; the agent never handles card or account details

Output & delivery

A Loop node runs the agent per due/overdue account → an Output node sends the tone-matched reminder with a UPI pay link in the borrower's language → the Condition node flags accounts past the delinquency threshold → those go to a collections officer at the Approval node for the call and any account decision.

Related industries

Build one of these in minutes

The sandbox gives you a live agent workspace — no account, no card. Or head back to the full catalogue and compare patterns across every industry.

Try this in the sandbox — no signupBrowse all industries
See how VegaDūta compares to n8n, Dify and BotpressWhy VegaDūta's architectureEstimate your WhatsApp API costs