Go-to-market — dogfood, then productize

The business case · every claim links to the screen that proves it

Phase 1 — Dogfood (internal)
now

Reliance runs Nirnay across Jio, Retail & O2C — hundreds of GSTINs on one ontology. Proves scale, controls and audit-readiness on the hardest tenant first.

Honest scope: this proves it works at scale — not that outsiders will pay. That's Phase 2.

Group Control Tower Autonomous Close
Phase 2 — Productize (Jio distributes)
next

Jio sells the same engine to Indian enterprises. Gate before scaling: 5–10 design partners, 3 paid pilots, 2 ERP environments — with measured lift.

Onboarding is a sales diagnostic, not self-serve — an instant risk scan → assisted rollout.

Onboarding diagnostic
The motion
Beachhead

₹500Cr+ turnover · 5+ · 5,000+ invoices/month. Manufacturing / auto / pharma first — high input-GST, heavy .

Channel

Assisted-first via Jio enterprise accounts + ERP/CA partners. Sellers get quota credit; CAs own config & review; partners paid on activated GSTINs.

Expansion

Start with one company's monthly close. Grow into group-wide oversight, then nearby jobs (notices, TDS) on the same connected data.

Pricing
Platform
Annual fee + tiered active GSTINs, banded by invoice volume
Modules
Autonomous Close & Group Control Tower as paid add-ons
Implementation
Connectors + rollout priced separately; paid pilot credited to year 1
Not priced as a % of ITC “recovered” — eligibility and timing are disputable, it encourages unsafe claims, and procurement can't budget it. ITC protected is the ROI story, not the invoice.
What's copyable, what isn't — the honest edge
Start with the truth: the matching is commodity — ClearTax, Cygnet and IRIS already do it — and a rival could rebuild these screens in a few months. No single feature is the moat. The edge is where the decision sits: in front of the payment, across the whole group, with a person's approval on every call — proven inside Reliance first, then sold through Jio.
In front of the payment
Incumbents report the mismatch after the invoice is approved. Nirnay holds the cash before treasury runs it — on the AP path, not in a tax silo.
Across the whole group
A supplier's total risk across every company — which a single-company tool simply can't see.
Automation with hard limits
A person sets the rules; the system acts within them; certain cases can never be auto-approved.
Where the decision record fits — and why it isn't the moatshow

The recorded decision history defends you when a notice lands — a real asset that compounds monthly, but be honest: a rival could partly reconstruct it from a client's past filings. It's a supporting edge, not the whole story. Running the AI on Jio's own cloud answers data-security worries — a nice-to-have, not the main edge.

“Where's the AI? This is just subtraction.”
What plain code does vs what the AI does — and the honest testshow

Correct — and that's the point. The maths is not AI, and never should be: the matching, the money at risk, the 18% interest and the safety checks are all plain, predictable code. We deliberately keep AI out of every number. Holding it back is what makes this trustworthy. The AI does the judgment and the writing that today eat an analyst's month:

Pure code — no AI (~86%)
  • 3-way match: your books × supplier's filing × goods receipt
  • Money at risk, credit at stake, 18% interest
  • Safety checks + auto-clear rules (nothing clears unless it's proven safe)
  • Clean matches — accepted in bulk, never touched by AI
Genuinely AI — judgment & language
  • The non-obvious call: Annapurna — accept the honest bill, but flag a 17(5) reversal. A dropdown gets this wrong.
  • Reads the free-text ASMT-10 notice and pulls out each point being questioned
  • Writes the reasoning, vendor-chase emails & the ASMT-11 reply — per case, not templated
  • Triages 70 exceptions → what a senior must actually see; escalates when unsure
The honest test: delete the AI and the numbers still compute — but you lose the reasoning, the citations, the notice-reading, the drafted replies and the triage. You're back to a spreadsheet of mismatches and a human doing 70 legal judgments by hand before the 14th. That gap is the product.
vs ClearTax / Cygnet / IRIS: they match and stop at “these two numbers differ” — a report that reaches the tax team after the invoice is approved. Nirnay puts the call in front of the payment: it tells you what to do under the law, a person approves, and the money is held before it leaves. It also sees the supplier across every company and remembers every decision for when a notice lands. Same match everyone has — but acted on, before cash moves, not filed after.
The three hardest questions — and the honest answers
Why does internal deployment prove external product-market fit?show
It doesn't. Dogfooding proves scale and auditability — not willingness to pay. Phase 2 gates on paid design partners first.
Why will Jio's salesforce or CA partners actually sell this?show
Access isn't distribution. Jio teams get quota credit on renewals; CAs own config & audit so it augments their billing; partners paid on activated GSTINs.
Why would a CFO replace a working compliance stack for an AI agent?show
They won't — so Nirnay sits above the incumbent: govern exceptions, record every decision, produce the evidence pack. Earn system-of-record status later.