Investors

The execution layer between an SME's books and its business. 

India has 60 million-plus small businesses. They run on Tally, WhatsApp and memory, and they learn about a cash crunch, a missed statutory date or a customer who has stopped paying only once it has cost them. Cortex reads their own numbers every day, names the specific thing that is about to cost them money — the customer, the amount, the date — and then does something about it.

★ Shark Tank India featured◆ DPIIT-recognised startup◇ Live and self-serve at cortex.mnbresearch.com
(01)The wedge

Not “run your business better”.
Money that is already at risk.

Three specific bleeds, in the order an Indian SME feels them. Each is quantified in rupees from the customer’s own rows, each has a deadline we did not manufacture, and each is verifiable on day one — which is how trust gets established before anyone is asked for money.

Receivables

Earned, invoiced, not collected. The owner finds out when they need the cash.

Section 43B(h)

Pay an MSME supplier late and the deduction is disallowed until you do. A tax event with a clock on it.

Statutory dates

GST, TDS, advance tax, ROC, the audit report. Each one a penalty with a date, and the calendar differs by registration.

The free 60-second Business Health Check is the top of this funnel — the product’s first chapter, not a lead magnet.

(02)The product

10 playbooks, running from the first import.

128 modules and 438 agents is the inventory, and the wrong thing to lead with. The unit that matters is the playbook: a specific condition in the customer’s data, and the action attached to it.

Nobody has paid youThe 45-day MSME clockThe date you were going to missWhen the cash runs outChasing, without you doing itAbout to run out of stockA customer going quietThe Monday planThe price you are leaving on the tableWho is actually worth keeping
Rules decide

Statutory windows, ageing, reorder points — tested arithmetic over real rows. A model asked to compute a tax date will eventually be wrong, and once is enough.

The model reads and writes

Tool calls over the workspace's own rows, an explanation in the owner's language, and the draft they were going to have to write.

The product acts

Draft, approve, send, and record what came back. A recommendation nobody executes is a PDF.

Tuned for 27 Indian industries. The customer keeps their accounting software — rip-and-replace is how SME software deals die, so “keep your Tally” is a sales asset rather than a limitation.

(03)Why now

Four forcing functions. None of them is “AI is exciting”.

A statutory clock that did not exist two years ago

Since FY 2024-25, payment to an MSME-registered supplier beyond the statutory window is disallowed as a deduction until it is actually paid. Supplier-payment hygiene stopped being a cash-flow preference and became a tax event with a date — an inherently computational problem, and a recurring, government-created reason to look.

SME data became machine-readable

GST e-invoicing thresholds have walked down year by year, so a meaningful share of Indian SMEs now produce structured transaction data as a by-product of compliance. Before that, “read their numbers” meant OCR on a photograph of a ledger.

Inference cost collapsed

A daily per-workspace analysis, plus drafting, plus a weekly plan, has to cost single-digit rupees for a ₹799 plan to work. Two years ago the unit economics of this product were negative by construction.

Discovery is moving to assistants

SMEs are increasingly found — or not found — through AI answers rather than ten blue links. The AI Visibility module is a second wedge for D2C and services, where the owner can see the problem on one screen.

(04)The moat

Ranked by how hard each is to copy, not by how good it sounds.

01
Statutory logic as tested code, not prompts

The 45-day MSME window, per-deadline notice periods, IST date handling and a per-business statutory profile live in tested modules — not in a prompt. Ask a model for a tax date and it will eventually be wrong once, and one wrong date destroys the trust the rest of the product runs on. Unglamorous, and genuinely hard to shortcut.

02
The ingestion layer nobody wants to build

Tally, Vyapar and Busy exports plus arbitrary CSV, normalised into one schema with customer identity resolved across spellings. This is where a new entrant loses three months, and it is the difference between a demo and a product an Indian SME can actually use on Monday.

03
Closing the loop, with evidence

Reminders draft in the customer's own name, send from their address, stop the instant an invoice is marked paid, and leave a recovery ledger of what came back. A kill switch, a circuit breaker and a do-not-contact list exist because the loop is real. Generating a reminder is easy; running outbound messaging on someone else's behalf, safely, is an operational moat.

04
A workspace worth more in month twelve

Metric history, decisions and memory accumulate per workspace. A rival starts from zero for every customer they win. The slowest moat to build and the most durable once built.

05
Accountants as the channel

One firm brings dozens of SMEs who already trust it, and credit pooling across client workspaces makes the firm the account. B2B2B beats paid acquisition in a market where trust is local — and a firm is a channel, not a different product, which is precisely what the earlier positioning got wrong.

06
The honest counter-argument

Tally, Zoho or a bank could bolt a warning layer onto data they already hold, and they have the distribution. The defence is not that they cannot — it is that warn → draft → send → prove is a different discipline from record-keeping, and record-keeping companies are structurally poor at acting on a customer's behalf. That is a real risk and belongs in the room, not in a footnote.

(05)Competitive frame

Everyone in this list is real, and none of them does this.

Tally · Zoho Books · Vyapar
Record what happened, correctly
Do not tell you what is about to happen, or act
A BI dashboard
Shows a chart when you open it
Notices nothing while you are not looking
ChatGPT and general assistants
Answer questions well, about text
Do not know your rows, your dates or your customers
A consultant or your CA
Judgement, monthly or yearly
Not daily, and not at ₹799
MNB Cortex
Watches the rows daily, names the specific risk, drafts and sends the action, and proves what came back. Deliberately does not replace the accounting system.
(06)Business model

Usage-priced credits from a ₹149 pack with no subscription, plans from ₹799 to ₹39,999 a month with a monthly allowance, and credit pooling so a firm can buy once and spend across client workspaces. Payments are live via Cashfree. The free health check needs no card.

Margin is enforced in code: the AI cost of each mode is modelled per credit so a plan cannot be priced below what it costs to serve. That check is how video generation was found to be loss-making at ₹270 a clip.

(07)Expansion
  1. Warn — the free check and the first import. Trust, in one screen.
  2. Act — collections, the weekly plan, the compliance calendar.
  3. Become the record of decisions — memory, the action board, the recovery ledger. Switching cost.
  4. Underwrite — a business whose receivables, payables and statutory position are verified daily is a business a lender can price. Stated as a thesis with a prerequisite — thousands of workspaces with continuous data — not as a roadmap item with a date.
(08)What has to become true

The product is deep. The company is early. Both are worth saying out loud.

There is no revenue chart on this page, because we will not publish a number the product cannot reproduce on demand. This repository runs a suite whose entire job is stopping published claims from drifting from the code; the same standard applies here. These are the metrics that decide whether this works, and where each one honestly stands.

Paying workspaces, and month-on-month growth
The only proof the wedge converts
Measured in cortex_payments; too early to publish
Import → first genuine warning
The core promise, timed
Instrumented (funnel_events)
Week-4 retention of the weekly plan
Whether the loop becomes a habit
Instrumented (weekly_plan_sends)
Exposure surfaced per workspace, in rupees
Value delivered in the customer's own units
Computable today; not yet aggregated
Recovered after a reminder, as a share
The claim the collections module exists to make
Recovery ledger shipped; needs volume
Gross margin after AI cost
Whether the pricing survives scale
Modelled per mode in test-margins; needs real usage
Firms live, and SMEs per firm
Whether the channel compounds
Console shipped; channel unproven
(09)Team

Built by MNB Research — the team behind AbroBot.

MNB Research is an Indian business growth and consultancy firm, featured on Shark Tank India and DPIIT-recognised. Its first AI product, AbroBot, is a study-abroad platform built on the same pattern: a deep vertical, real operational data, and AI doing work rather than answering questions. Cortex applies that to a far larger market.

AbroBot’s own metrics are its own; they are evidence that this team ships and distributes an AI product in India, and they are not presented as Cortex traction.

(10)The ask

Fund the distribution, not the demo.

The product is live, self-serve, and deeper than it needs to be for the wedge it sells. What it has not yet had is a go-to-market: firms signed as a channel, the health check put in front of owners at volume, and the outcome metrics above turned from instrumented into proven.

MNB Research · operated by Abrobot Technologies (Pvt Ltd), Delhi, India · +91 97114 88481