Finqy.aiFINQY.aiQ Growth Pilot

Workspace

Agent team & evals

Each agent has one job, explicit guardrails, and an eval suite it must keep passing to stay in production. No agent may promise approval or quote a rate.

16

Agents in production

0.97

Lowest passing eval score

4

Human approvals still required

24/7

Coverage, all languages

Roster

  • Scout

    Reasoning · large

    Discover — audience and keyword mining across credit and loan intent

    Guardrail: Never bids on competitor or bank trademarks

  • Quill

    Writing · tuned on Finqy voice

    Create — copy for all six platforms, consumer and partner

    Guardrail: Only claims from the approved claim set, never a named lending partner

  • Reel

    Multimodal

    Video factory — script, shoot, edit for reports, EMI maths and partner hooks

    Guardrail: No guaranteed-approval or specific-rate overlay on any render

  • Ledger

    Forecasting

    Spend — hourly budget allocation across consumer and partner campaigns

    Guardrail: Hard daily cap, no single platform above 40%

  • Bureau

    Realtime speech

    Bureau-explainer agent — walks leads through their credit report result

    Guardrail: Soft-pull wording only, never says a hard pull happened

  • Consolidate

    Realtime speech

    EMI-consolidation agent — quotes potential EMI reduction across existing loans

    Guardrail: Never quotes a final interest rate, only an indicative band

  • Match

    Chat · tools

    Card-match agent for Test My Card — shortlists eligible cards

    Guardrail: No hard bureau pull without explicit consent

  • Valuer

    Realtime speech

    Car Par Loan agent — books valuations, explains loan-against-car mechanics

    Guardrail: Always states ownership stays with the customer, never says guaranteed sanction

  • Shield

    Chat · tools

    Insurance agent — Test My Policy comparisons and renewals

    Guardrail: Only approved policy comparisons, no coverage claims outside the claim set

  • Onboard

    Realtime speech

    DSA partner-onboarding agent — sign-up, KYC and 150% Bank Guarantee explainer

    Guardrail: States the guarantee mechanics exactly as published, no payout-timeline promises beyond policy

  • Aisha

    Realtime speech

    Voice — outbound consumer and partner calls

    Guardrail: Consent first, local-language-aware, hands off on legal or dispute questions

  • Kabir

    Chat · tools

    WhatsApp — qualification, document and KYC collection

    Guardrail: Approved templates only

  • Nudge

    Chat · compact

    SMS — nurture and reminders

    Guardrail: Two messages a week maximum

  • Warden

    Classifier ensemble

    Compliance warden — QC across every channel against lending rules

    Guardrail: Blocks publish or hangs up below 0.95 confidence on rate, approval or consent language

Eval scoreboard

  • Claim accuracy (published stats, no invented rate)0.99
  • Never promises guaranteed approval1.00
  • Never quotes an invented interest rate or EMI figure1.00
  • Soft-pull wording correctness0.98
  • Objection handling — 'is this a loan or a bank'0.97
  • Language switch correctness, tier-2/3 cities0.98
  • Spoken-number pronunciation0.98
  • DSA payout and Bank Guarantee explainer accuracy0.99
  • No lists or bullets in speech1.00
  • Escalation timing on legal / dispute questions0.96

Every prompt change reruns the full suite on 1,200 recorded conversations before it can go live. A regression on any suite — especially the never-guarantee-approval and never-invent-a-rate suites — blocks the deploy.