A financing request becomes an investigation case.
Financing investigation intelligence
Investigate the financing cases that deserve attention.
ANORIUM connects transaction evidence, entities, relationships, behaviour and history to help lenders prioritize which invoice and receivables-financing cases deserve deeper human investigation.
- Entity
- Transaction
- Document
- Evidence
- History
- Behaviour
ANORIUM investigates. The lender decides.
Unknown is not safe.
Example investigationSynthetic dataCASE #AN-1042
Evidence submitted
Tax invoice₹17.70L₹17,70,000
Purchase order₹11.80L₹11,80,000
4 other fields corroborated
ANORIUM findingREVIEW REQUIRED
Invoice amount is not corroborated by the purchase order
Invoice / purchase-order amount inconsistency. The available evidence does not corroborate the transaction amount.
Evidenceev_8c41f2a9ev_5b70de13
What would resolve this
- Verify the invoice amount against an authoritative source
- Obtain the amended purchase order, if one exists
- Confirm payment realization
Built for financial institutions
Tenant isolation, access control and auditability by design
Evidence-backed investigations
Every finding resolves to the claim that established it
Analyst first, always
The engine investigates. An authorized human decides
Fail-closed by design
A check that could not run lowers coverage — it never passes
Built for
- Banks
- NBFCs
- Invoice financiers
- Supply-chain finance
- Fintech lenders
V1 scope: invoice, receivables and working-capital financing
Documents can pass.
The transaction can still be suspicious.
A lender may already verify every document in the file, and every one of them may check out. That tells you the paperwork is consistent. It does not tell you the transaction behind it is real.
Already verified, and passing
- KYC
- GST / e-invoice
- Invoice
- Purchase order
- Credit information
- Existing rules
- LOS / LMS workflows
The risk that survives all of it does not live inside any single record. It lives between them.
One financing case₹50,00,000
Illustrative- BorrowerRequests financing against the invoice
- SupplierIssued the invoice
- BuyerOwes the receivable
- FinancingThe facility being drawn on
- HistoryWhat these parties did before
Each individual record may look valid.
The relationship between them may still deserve investigation — and no document-level check is looking at the relationship.
The investigation layer
Existing systems verify the pieces.
ANORIUM investigates whether the pieces make sense together.
Every system below answers one question about one record, and answers it well. None of them is asked whether the entities, the relationships, the timing and the history behind the case are consistent with each other. That question is the one ANORIUM is built to ask.
ANORIUMInvestigation layer — evidence, entities, relationships, behaviour, history
Your existing stack — unchanged
- KYCIs the party who they claim to be?
- GSTWas the invoice reported?
- OCR / document verificationDoes the document read correctly?
- Credit bureauHow has the borrower repaid before?
- BRMSDoes the case break a rule?
- LOS / LMSWhere is the case in the process?
ANORIUM does not replace your existing lending or fraud systems. It reads what they already produce and investigates what none of them is asked.
Eleven steps, and only one of them is a decision.
A case moves through the same sequence every time, and the sequence is the same for every case. Nothing here is generated prose and nothing here is sampled — the same inputs produce the same findings.
01 — 04Assemble the picture05 — 08Interrogate it09 — 11Act on it
01Case 02Ingest Collect available financing evidence.
03Resolve Identify the entities involved.
04Connect Build the entity and transaction relationships.
05Detect Surface suspicious patterns.
06Correlate Connect independent signals.
07Evidence Show what supports the finding.
08Prioritize Identify which cases deserve attention.
09Investigate Guide the analyst.
10Decide The lender makes the final decision.
11Report Create the investigation record.
ANORIUM investigates. The lender decides.
Nine surfaces, nine questions.
Every screen in ANORIUM exists to answer one question an analyst actually asks out loud. Nothing in the workspace is named after the machinery behind it.
What happened?
One financing request, with the documents, parties and amounts that arrived with it.
Why is this case receiving attention?
The investigation stated in six sections, assembled from findings and evidence — never written prose.
What looks unusual?
The patterns that fired on this case, each one a reason to look rather than a conclusion.
Who is connected?
The entities behind the case and why they are linked — resolved on GSTIN, PAN, CIN and DIN.
What supports the finding?
The append-only ledger every finding rests on, with the source and time of each claim.
What happened and when?
The case placed in order — entity age, dormancy, and how old the relationship was when financing was sought.
What should the analyst do next?
The specific checks that would resolve the case, and the evidence each one would produce.
What did the lender decide?
The analyst's outcome, recorded with who made it and what they were looking at when they did.
What can be shared with the committee?
A deterministic record of the findings, the evidence, the coverage and the checks that could not run.
Ten patterns worth a second look.
Each of these is a reason to investigate — never a conclusion, and never a score. A finding tells an analyst where to point their attention and what evidence would settle it.
Potential duplicate financing
Has this receivable already been financed, here or elsewhere in the book?
Potential invoice cloning
Does this invoice reproduce another one with the identifiers changed?
Potential related-party exposure
Are the supplier and the buyer connected through shared directors or control?
Potential circular trading
Do goods and invoices return to where they started?
Potential round-tripping
Does value leave and come back through a chain of intermediate parties?
Network topology concentration
Does the network form a hub, a closed loop, or an unexplained intermediary?
Behavioural anomaly
Has this counterparty started behaving unlike its own history?
Temporal anomaly
Do the dates hold — entity age, dormancy, and relationship age at financing?
Transaction velocity anomaly
Is the invoice-to-presentation rhythm out of step with the same party's baseline?
Shell-entity indicators
Does the counterparty show the markers of an entity with no real operations?
A finding is not a finding of fraud. ANORIUM reports what is unusual and what supports that view. Whether it is fraud, an error, or a perfectly ordinary arrangement is a judgment the lender makes.
Every important finding should be explainable.
A finding an analyst cannot take apart is a finding they cannot defend six months later to someone who was not there. So every one is assembled from links that can each be pointed at.
- FactSomething the evidence statesInvoice = ₹50L
- RelationshipHow the parties are connectedDirector X is associated with Supplier B
- SignalWhat a detector observedTransaction activity materially exceeds baseline
- CorrelationWhat the signals say togetherMultiple independent signals converge
- RecommendationWhat would settle itVerify receivable uniqueness
Unknown ≠ safe.
If important evidence is unavailable, ANORIUM does not pretend the case is clean. Missing data is reported as missing, on the case and in the report, and the case can come back as:
Indeterminate
A verdict of its own, in its own colour — not amber, which would be a warning about the counterparty, and not green, which would be a lie. Unknown is not safe.
The case, stated as
an argument.
An analyst opening a case should not have to reconstruct why it is in front of them. The thesis is the first thing on the screen, and it is assembled from findings and ledger claims — never composed as prose.
No language model contributes a factual statement to it. The same case produces the same thesis every time it is generated, which is what makes it something an analyst can put in front of a committee.
- What we knowThe facts the evidence establishes, each attached to the claim it came from.
- What changedHow this case departs from what these parties did before.
- Why this is unusualThe argument for attention — stated, not scored.
- What supports itThe specific evidence behind each statement, by identifier.
- What remains unknownThe checks that could not run, and what they would have covered.
- What should be investigatedThe next actions, and the evidence each one would produce.
Prioritize the portfolio, not just the case.
A financing book is screened in one submission. What comes back is not a score per row — it is an order of attention, so a finite analyst team spends its week on the cases where the week is worth spending.
Your financing portfolio
Historical or in-flight cases, submitted as one file
ANORIUM
Every case investigated the same way, in the same order
- Clear—No material signal, on the evidence available
- Review—Something is unusual; an analyst should look
- Escalate—Converging signals; look first
- Indeterminate—Not enough evidence to say either way
Counts are shown as placeholders. ANORIUM has no published portfolio results, and a number here that came from nowhere would be the least defensible thing on this page.
The purpose is not to reduce the number of cases a lender looks at. It is to change which cases get the hour.
Start with your historical portfolio.
Give us 24 months of historical financing data. ANORIUM analyzes the portfolio retrospectively, so you can compare its findings against what your existing process concluded — on cases whose outcomes you already know.
Your existing process
What your team, rules and systems concluded at the time.
- Cases your process flagged
- Cases your process cleared
- The decision that was actually taken
ANORIUM
What the same book looks like when the cases are investigated as a set.
- Cases ANORIUM would raise
- Cases ANORIUM would clear
- Cases ANORIUM cannot call either way
The comparison reports both directions. What your process caught and ANORIUM missed is shown as prominently as what ANORIUM added — and rows whose outcome could not be read are counted separately rather than scored as agreement.
What the audit measures
These are the measurements we use to validate the product. ANORIUM has not published results for any of them, because no lender’s book has been run.
- Incremental findingsWhat did ANORIUM surface that your process did not?To be measured
- Analyst agreementOn review, how often do your analysts agree the finding was worth raising?To be measured
- False-positive burdenHow much work does a finding cost before it can be closed?To be measured
- Investigation timeHow long does a case take with the evidence already assembled?To be measured
Run a Controlled Dark Audit
Ten questions, four minutes. No data is exchanged through the form.
Six columns and
24 months of history.
A controlled historical dataset is enough to run the first audit. Four of these columns are required; the other two make the temporal and portfolio checks meaningful.
Start with a controlled historical dataset. No production integration is required for the first validation.
- No production integration
- No API access to your systems
- No borrower documents in the first pass
- No PII beyond the identifiers above
- No change to your existing workflow
- invoice_numberThe lender's own reference for the invoiceRequired
- supplier_gstinThe party that issued the invoiceRequired
- buyer_gstinThe party that owes the receivableRequired
- invoice_amountThe financed valueRequired
- invoice_dateNeeded for every temporal checkOptional
- financing_statusRequested, financed, declined, repaid — read only if suppliedOptional
Plus 24 months of history, so behaviour can be measured against a baseline rather than against a threshold we invented.
Security before scale.
ANORIUM is designed for evidence traceability, tenant isolation, controlled access, auditability, and fail-closed handling of missing security dependencies. Every control named here exists in the product today.
There are no certifications to claim, and none are implied. The security page states what is implemented and, more usefully, what is not — including the work a deployment into a regulated environment would still require.
View security approachAuthentication
Sign-in is by invitation. Sessions are HttpOnly cookies; refresh tokens rotate, and a reused token revokes its whole family.
Role-based access
Every route and every destructive action is gated on a named permission, enforced by the backend rather than the interface.
Organization isolation
Cases, documents and evidence are scoped to one organization. A cross-organization identifier resolves to nothing.
Controlled document access
Documents are content-hashed and integrity-verified on retrieval, and reached through an authorized endpoint — never a public path.
Evidence-linked findings
A finding cannot exist without the evidence records it was computed from, so nothing on screen is unattributable.
Case activity history
Uploads, findings, changes and approvals are recorded per case with the acting user and a request correlation id.
The eight that come up first.
Short answers, and the limits stated.
Does ANORIUM replace our existing lending or fraud systems?
No. Your KYC, GST checks, document verification, credit bureau, rules engine and LOS/LMS stay exactly where they are. ANORIUM reads what they already produce and investigates the question none of them is asked: whether the entities, relationships, timing and history behind the case are consistent with each other.
Does ANORIUM make the final credit decision?
No. ANORIUM investigates; the lender decides. It produces findings, the evidence behind them and the questions still open. An authorized person resolves the case, and the platform can separate the person who investigates from the person who approves.
Do we need to give you production data?
Not for the first validation. A Dark Audit runs on a controlled historical dataset — six columns and 24 months of closed cases, exported from a system you already run. No integration, no API access, and no borrower documents in the first pass. Production data comes only after your own security review.
Is there a score behind this?
No. There is no score anywhere in the product. Findings are produced by deterministic detectors — the same case produces the same findings every time — and each one is attached to the evidence it was computed from. A language model is used only to explain findings that already exist, must cite evidence for every claim, and cannot create, alter, remove, approve or rank a case. If the model is unavailable the findings are unaffected.
What happens when the evidence is missing?
The case is reported as indeterminate rather than clear. Missing evidence is recorded as missing on the case, in the thesis and in the report, and the checks that could not run are named. A system that treats absent data as a pass is wrong in the one direction that costs money.
How is customer data handled?
Cases, documents and evidence are scoped to a single organization; a cross-organization identifier resolves to nothing. Access is role-based and enforced by the backend. Sessions are HttpOnly cookies. Documents are stored with integrity verification and served through authorized, short-lived links. Case activity is recorded. The full list of implemented controls is here.
Can ANORIUM integrate with our systems?
The platform has connectors for the source types an investigation needs — documents, transaction records, and external evidence sources your institution is entitled to query. Which of those are enabled is a deployment decision, scoped during evaluation. We do not claim a live integration with any provider we have not connected for you.
Is ANORIUM production-ready?
ANORIUM is in early commercial deployment. The platform is being hardened for financial-institution use: the controls listed on the security page are implemented, and the ones that are not — an independent penetration test, and any external certification — are named there too. We would rather you evaluate that honestly than discover it later.
ANORIUM is built around a simple question: what if the financing case most worth investigating is the one where every document already passed?
That means investigating the entities, relationships, behaviour and timing behind a financing case — not just the paperwork in front of it — with every finding attached to the evidence it came from, checks that produce the same answer every time they run, and a report an analyst can defend months later to someone who was not there.
The ask
Give us a controlled historical portfolio.
We are looking for one or two lenders to run a controlled historical Dark Audit — 24 months of closed financing cases, investigated retrospectively, compared against the decisions you already took.
ANORIUM investigates. The lender decides.
Already have access? Open ANORIUM