A Delhi fintech replaced 14 spreadsheets with one warehouse and three dashboards
A founder was spending 6 hours every Monday rebuilding the same weekly business review from spreadsheets pulled out of five different systems. We set up a BigQuery warehouse, dbt models, and three dashboards. He got his Mondays back.
The challenge
The founder was CFO-by-necessity. Every Monday morning he would pull CSVs from Razorpay, Cashfree, their CRM, Zoho Books, Mixpanel, and a Postgres replica; paste them into a master spreadsheet; reconcile differences; and rebuild the same 12-slide weekly business review.
The process took 5-6 hours, broke every time someone upstream renamed a column, and meant the leadership team never saw numbers more frequently than weekly. Fraud cases in particular were being caught 5-7 days late.
Our approach
We stood up a BigQuery warehouse in ap-south-1, with ingestion via Airbyte from the five SaaS sources and a logical replication slot from Postgres for the transactional data.
Transformation was done in dbt, with one staging model per source and three core fact tables (payments, customers, support_tickets). Tests on every primary key, foreign key, and business invariant (e.g. settlement amount = sum of txn amounts minus fees).
Three dashboards were built in Metabase: (1) an executive dashboard for the founder with revenue, CAC/LTV, active users, and fraud flags; (2) a support dashboard with SLA, ticket backlog, and first-response time; (3) a payments ops dashboard with settlement lag, chargeback rate, and gateway health.
We set up scheduled alerts in Metabase so the founder gets a Slack ping if revenue drops more than 10% day-over-day or if fraud-flagged transactions exceed a threshold — instead of finding out in Monday's review.
Knowledge transfer: two training sessions with the ops team and a dbt docs site they can navigate themselves.
The results
Monday morning data work went from 5-6 hours to zero. The dashboards are live by 9am every day, auto-refreshing every 5 minutes.
Fraud detection latency went from 5-7 days to real-time alerting. Three fraud rings were identified and shut down within the first two months that would have cost an estimated ₹42L in chargebacks if caught at the old cadence.
The leadership team now makes decisions against daily numbers rather than weekly. Response time to business anomalies is measured in hours, not weeks.
Takeaways
- A founder pulling CSVs is a warehouse project waiting to happen. Every hour they spend there is unpaid data-engineer time.
- Three dashboards beat fifty. Solve the top use cases cleanly before you build a library of one-off reports.
- Alerts > dashboards for anomalies. Nobody looks at a dashboard at 2am; Slack pings do the watching for you.
Thinking about a similar project? Let's talk specifics.
Book a free 45-minute call. Bring your current numbers; we'll bring an honest opinion on what would actually move them.
Book ConsultationA Delhi fintech replaced 14 spreadsheets with one warehouse and three dashboards — FAQ
Questions we usually get about projects like this one.
The outcome above is specific to this client's baseline, market, and constraints — we don't promise identical numbers. What we can replicate is the method: the same diagnostic, the same attribution discipline, and the same hand-coded delivery approach used on this analytics engagement. During the discovery call we will map your starting point against this case and tell you what parts of the playbook realistically transfer and which would need to be adjusted.
For analytics engagements we default to a lean, well-supported stack: React + Vite for custom frontends, Node.js or serverless functions on Vercel for any backend, Shopify / WooCommerce where commerce is in scope, GA4 + PostHog + server-side tagging for analytics, Meta Ads / Google Ads / LinkedIn Ads for paid, n8n or native Zapier for workflow automation, and Tally / Zoho Books for accounting integrations. We pick the simplest tool that clears the bar — we don't sell stack complexity.
This specific engagement ran for the period referenced above. Typical analytics projects of comparable scope take between 4 and 12 weeks from kickoff to launch, with a further 4 weeks of post-launch iteration. Fixed-scope work is priced upfront; open-scope work is run on a monthly retainer with weekly written updates so there are no invoice surprises.
Pricing depends on scope, integrations, and whether we are starting from scratch or improving an existing setup. We publish transparent price bands per dimension on our pricing page — engagements comparable to this case study typically sit in the mid-tier band for Analytics. You can see exact ranges at /pricing, and we confirm a fixed fee (or retainer range) before any work begins.
You do. All code is delivered into your GitHub / GitLab organisation. Ad accounts, analytics properties, domains, and hosting sit in your name — not ours — from day one. On handoff we transfer admin access, document every integration, and record a 30-minute video walkthrough of the stack so your next hire or incoming agency can pick it up without us being a single point of failure.
More case studies
How a Bangalore D2C skincare brand cut bounce rate 41% with a hand-coded rebuild
A skincare brand was losing mobile shoppers at the homepage because their Shopify theme was shipping 2.1 MB of JavaScript. We rebuilt the storefront on a custom stack, kept the Shopify backend, and cut time-to-interactive from 5.4 seconds to 1.9 seconds.
Read case studyFrom ₹1.2 L CAC to ₹38K CAC: scaling a B2B SaaS on LinkedIn Ads
A founder-led SaaS was paying ₹1.2L to acquire each paying customer on LinkedIn Ads. We rebuilt their funnel, targeting, and creative, and brought CAC down to ₹38K inside one quarter — with more customers than before, not fewer.
Read case studyLet's talk about your business.
Tell us what you're working on and where you want to go. We'll put together a plan. No obligation, no sales pitch.
- Free 30-minute call
- A plan built around your goals
- No obligation, no pressure
- Your own account manager