Graph

Procurement data traceability: an edge without a source never enters the graph

Procurement data traceability is decided when the data is written, not when someone challenges it. In the graph built by Graph, every relationship carries the document, the page, the line, the date, the agent that wrote it and its confidence level. Whatever has no source never gets in.

Book a demo Request a diagnostic

The problem

A figure with no source cannot be defended

A dashboard showing consolidated spend without showing where it comes from is not enforceable. In front of a supplier, in front of an auditor, or simply in front of a finance director who has doubts, the question is always the same: where does this figure come from? If the answer takes a week of manual rebuilding, the figure is useless.

That is why traceability is not a feature bolted on afterwards but a condition of writing. The graph applies an admission control: whatever does not conform to the delivered procurement ontology is rejected, not written. Three rules describe it.

An entity the ontology does not declare is not created. The graph does not invent a node type to house data that fits nowhere.
A relationship absent from the blueprint is not written, even when both ends exist. Linking an item to a payment without going through the invoice means nothing: the link is refused.
A relationship with no provenance is not admitted. A fact no document carries cannot be relied on later, so it does not get in.

Provenance, edge by edge

Every relationship written into the graph opens onto what produced it. The principle is the same as the clause / invoice line demonstration of the Compliance agent, applied this time to the structure itself: on the left, what the graph asserts; on the right, the document line that asserts it.

What the edge says

Relationship: “priced by”
Source node: Item: technical solvent, 25-litre drum
Target node: Schedule of unit prices v3 of the framework agreement
Value carried: Unit price per litre, period from 1 April to 31 December

What grounds it

Document: Schedule of unit prices v3, appendix 2 of the framework agreement
Page and line: Page 4, line 27
Date of the fact: Effective date read on the amendment, not inferred
Agent: Written by Compliance, during the check
Confidence: Confidence level carried by the relationship, available on request

No piece of information in the graph escapes that rule: we can write nothing into it that we cannot trace back to the document line it came from. It is a construction constraint, and it is what turns verification into one click rather than one week. Every extracted value stays linked to the exact place in the document where it was read.

“How do we know the graph is telling the truth?”

You do not take its word for it, you open it. Verification needs no export, no ticket to the data team and no access to a third-party system: it starts from the figure on screen and goes down to the document.

A figure opens onto the graph relationships that produced it; each relationship opens onto its document line. The double link is always available, not only in a dispute.
Information with no source never entered: missing evidence is not an edge case to handle, it is an impossible case.
Whatever is uncertain carries its uncertainty. An ambiguous match is flagged as such rather than presented as an established fact.
The trace is dated and attributed: which agent wrote the relationship, from which document, on which date. The history of an edge reads like the history of a ledger entry.

Nothing is lost. Everything can be checked, everything can be proven. Memory stops being individual. A renegotiation starts from what was actually invoiced, not from the theoretical rate.

Procurement intelligence · Graph

All your purchasing knowledge, connected and queryable.

Turns the documents the agents have already read (contracts, amendments, schedules of unit prices, rate cards, purchase orders, goods receipts, invoices, credit notes) into a typed graph built on a procurement ontology that ships with the product: entities joined by explicit relationships, each one tied to the document, the page and the line it came from. An edge without a source never enters the graph.

Memory stops being individual. A renegotiation starts from what was actually invoiced, not from the theoretical rate.

Discover Graph What the AI does, and never does: the Zylio AI mandate

What Graph detects

The same supplier under several legal entities
The same item under several references depending on the supplier
Units of measure that do not compare
An internal classification that has drifted
Contracts nobody can tell still cover the spend in progress
Credit notes promised and never applied to the invoice they came from

Nothing is lost. Everything can be checked, everything can be proven.

The contract clause and the invoice line, highlighted side by side.
Every extracted value stays linked to the exact place in the document where it was read.
The same case produces the same decision, today as in six months: the rules are applied deterministically.
No discrepancy is set aside in silence. Anything that matches no rule is raised, with its reason.
The agent records what it did, in the order it did it: who, what, how much, when.

Frequently asked questions

What is the admission control of the graph?

The rule that decides what is allowed to be written. An entity, a relationship or a value that does not conform to the procurement ontology is rejected rather than recorded, and a relationship without provenance is never admitted. The graph therefore holds only links that were deliberately modelled and sourced.

What happens if a document is unreadable or incomplete?

Nothing is guessed. The readable part enters the graph with its source; what is missing stays missing and is flagged as missing information, with the document concerned. An estimated value would be indefensible in front of a supplier, so it is not written.

How do I check a Zylio conclusion?

Every discrepancy opens onto its evidence: the contract clause and the invoice line, highlighted side by side, with the calculation shown.

Is my data used to train models?

No, never. It is not shared between clients and stays hosted in France.

Measurable impact in every environment

More than 5 million procurement documents analysed

Between 1 and 7% of margin recovered

on the scope analysed

From 15 to 45% of time given back to teams, per FTE

depending on the scope and on data maturity

Zylio fits into your existing ecosystem.

The ERP runs the process. Zylio handles the exception and recovers the value that escapes it: invoices without a purchase order, line-by-line price discrepancies, duplicates and overbilling, off-contract spend.

All integrations

Your data under high security.

Zylio meets the most demanding standards, and nothing is committed without your approval.

Certifications: SOC 2 Type II · ISO 27001
Hosting: Hosted in France
Encryption: End-to-end AES-256 encryption
Access: Enterprise SSO · multi-factor authentication · Zero Trust approach

Security and compliance

See what this looks like on your own data

Twenty minutes, on a spend category of your choosing. We show you what the agents detect, with the evidence behind it.

No commitment, on your own data
Result in 3 weeks
20 minutes, no sales pitch
Your data stays hosted in France
No change of tool or process