8.2 Recommended stack (Fast MVP, Scalable, Enterprise)
Three reference stacks
Section titled “Three reference stacks”Stack A — Fast MVP (months 1–9, book up to ₹30 Cr)
Section titled “Stack A — Fast MVP (months 1–9, book up to ₹30 Cr)”Goal: ship live lending in 6 – 9 months on ₹3 – 5 Cr of platform investment.
| Layer | Choice |
|---|---|
| LOS + LMS core | Buy from CloudBankin / Finezza / Biz2X (whichever fits cost + speed) |
| Underwriting rule engine | Use vendor’s BRE; build only top 20 – 30 rules |
| Decisioning + manual review | Vendor + own light supervisor UI |
| KYC + KYB | Karza / Signzy / IDfy (aggregator) |
| V-CIP | IDfy / Signzy |
| BSA | Perfios (or FinBox) |
| AA | Setu / FinBox |
| GST | Cygnet / Webtel / Vayana |
| eSign | Leegality / Digio |
| eStamp | Same |
| NACH + UPI AutoPay | Digio + Razorpay |
| Payment gateway / payouts | Razorpay / Cashfree |
| SMS / WhatsApp | Gupshup / Kaleyra |
| Amazon SES | |
| Dialer | Exotel / Knowlarity |
| Co-lending allocation | Build light layer on top of vendor (first partner only) |
| Borrower portal | Build light (React + own BFF) |
| Admin console | Use vendor admin + own light additions |
| Analytics | Metabase on warehouse |
| Warehouse | PostgreSQL replica + dbt (lightweight) |
| Hosting | Cloud (AWS Mumbai); managed services preferred |
Total platform spend year 1: ~₹3 – 6 Cr (vendor + cloud + tooling), excluding engineering team.
Stack B — Scalable (months 9–24, book ₹30 – 200 Cr)
Section titled “Stack B — Scalable (months 9–24, book ₹30 – 200 Cr)”Goal: replace as much vendor “core” as possible with own IP; keep primitives.
| Layer | Choice |
|---|---|
| LOS + LMS core | Build own progressively (start with LMS; LOS next) |
| Underwriting rule engine | Drools or decision-tables in DB + build rule-admin tooling |
| Decisioning | Build |
| Manual review | Build rich case-management |
| KYC + KYB | Continue buying (multi-vendor) |
| V-CIP | Continue buying |
| BSA | Continue buying (Perfios primary, FinBox secondary) |
| AA | Continue buying (multi-AA via Setu / FinBox) |
| GST | Continue buying (multi-GSP) |
| eSign + eStamp | Continue buying |
| NACH + UPI AutoPay | Continue buying |
| Payment + payouts | Continue buying |
| SMS / WhatsApp / Email | Continue buying |
| Dialer + collections workflow | Buy dialer; build collections workflow |
| Field-agent app | Build |
| Co-lending — multi-partner | Build allocation + settlement engine |
| Borrower portal | Build (full) |
| Admin console | Build (full, replace vendor) |
| Analytics + warehouse | Snowflake / ClickHouse + dbt + Metabase |
| ML / model registry | (start at year 2 if ready) |
| Workflow engine | Camunda 7 / Temporal |
| Hosting | Cloud, multi-AZ |
Total platform spend year 2: ~₹6 – 12 Cr (vendor + cloud + tooling), excluding team.
Stack C — Enterprise (year 2+, book ₹200 – 1,000 Cr)
Section titled “Stack C — Enterprise (year 2+, book ₹200 – 1,000 Cr)”Goal: become a platform that powers own + partner books at scale, with optional SaaS productisation.
| Layer | Choice |
|---|---|
| LOS + LMS | Own |
| Underwriting engine | Own rule engine + ML scorecards |
| Decisioning | Own |
| Co-lending — 3+ partners + DLG | Own end-to-end |
| KYC / KYB / V-CIP | Multi-vendor (best primitives + reliability) |
| BSA / AA / GST | Multi-vendor |
| eSign / eStamp | Multi-vendor with intelligent routing |
| Mandates + repayments | Multi-rail (NACH + eNACH + UPI AutoPay + BBPS) |
| Communication | Multi-vendor |
| Dialer + collections | Best-in-class |
| Field-agent app | Own, possibly white-label-able |
| Borrower / DSA / CA / partner / anchor portals | Own |
| Admin console + multi-tenant | Own |
| Analytics + warehouse | Snowflake / Databricks + dbt + Looker / Superset |
| ML — feature store + model registry + monitoring | Feast + MLflow + own deployment |
| Workflow engine | Temporal at scale |
| Hosting | Cloud, multi-AZ + DR multi-region |
| Productisation | Own, optional SaaS for external customers |
What to build internally (independent of stack)
Section titled “What to build internally (independent of stack)”These are competitive moats and should never be vendor-dependent:
- Policy rule library — your credit IP.
- Co-lending allocation + settlement — operational moat.
- CA / Tally distribution — channel UX that vendors don’t ship.
- Repeat-borrower scorecard — internal performance data is your asset.
- Cohort + portfolio analytics — your view, your warehouse.
- DLG monitoring + invocation — regulatory + operational discipline.
What to never build first
Section titled “What to never build first”These are mature commodities that vendors do better:
- KYC primitives — PAN, Aadhaar, OVD, V-CIP.
- BSA parser.
- AA TSP integration / FIU SDK.
- GST APIs (require GSP licence).
- eSign / eStamp (require partner licence).
- NACH activation rail.
- SMS / WhatsApp / Email rails.
- Sponsor-bank API banking primitives (the bank itself is the rail).
Custom-build roadmap (suggested)
Section titled “Custom-build roadmap (suggested)”| Time | Build |
|---|---|
| Month 0–3 | Acquisition, application, admin, integrations layer, light decisioning |
| Month 3–6 | KYC orchestration, ingestion pipeline, basic underwriting rules, sanction + docs |
| Month 6–9 | Disbursement, LMS (if not buying), collections basics, MVP analytics |
| Month 9–12 | Co-lending v1 (single partner) — allocation + settlement |
| Month 12–18 | LMS own-build (if started on vendor), enrich UW rules, fraud signals, field-agent app |
| Month 18–24 | Multi-partner co-lending, advanced analytics, ML challenger, custom warehouse |
| Month 24+ | Multi-tenancy / productisation, advanced ML, expansion products |
Build vs buy decision template
Section titled “Build vs buy decision template”For each module, ask:
- Is there a vendor that does this well? → Y/N.
- Is the vendor’s cost reasonable for our volume? → Y/N.
- Is this our competitive moat? → Y/N.
- Will we need deep customisation? → Y/N.
- Can we afford the engineering cost + risk + delay to build? → Y/N.
- What’s the switching cost if we buy now and build later?
The recommendation per module is the honest answer, not a default to either side.