Collateral operations runs on numbers that originated in legal documents. The threshold, the minimum transfer amount, the independent amount, the eligible collateral table and the valuation percentages are negotiated in a credit support annex, executed, filed, and then typed into the collateral system by someone reading the PDF. From that moment the authoritative version and the operational version are two separate things that happen to agree.
They stop agreeing gradually. An amendment adjusts a threshold for one counterparty. A collateral eligibility schedule is replaced by a new annex. A corporate reorganization changes which entity faces which. Each change should flow to operations, and most do, but the ones that do not are invisible until a margin call is disputed and both sides open their documents.
The underlying problem is that the credit support annex is treated as a legal artifact rather than as a source of reference data. Firms that treat it as both, holding the terms as structured fields linked to the executed document, reconcile the two in minutes rather than discovering the gap through a dispute.
The terms that drive a daily margin call
A small number of fields determine almost every collateral outcome. They are the fields most likely to be re-keyed, and therefore the fields most worth holding authoritatively.
- Threshold and independent amount, the unsecured exposure each party will tolerate and any additional amount required regardless of exposure, both frequently counterparty-specific and sometimes rating-linked
- Minimum transfer amount and rounding, the smallest call that will be made and the rounding convention applied, which together explain a meaningful share of small-value disputes
- Eligible collateral and valuation percentages, the instruments each party may deliver and the haircut applied to each, typically a table that is amended more often than the main body
- Valuation and notification timing, the valuation date convention, the notification deadline and the settlement timing, which decide whether a call is timely
- Dispute resolution mechanism, the procedure when the parties disagree on exposure or value, including the undisputed amount that must still transfer
These are reference data with legal provenance, which is an awkward category for most architectures. The discipline needed is the one applied to regulated contract registers, where each required term is mapped to a field and kept current against the executed document. The register model in the bank’s guide to DORA ICT third-party contract terms solves the same shape of problem for a different regime.
When a margin call is challenged, the investigation frequently finds that the two sides are applying different thresholds, different haircuts or different rounding, rather than disagreeing on the valuation of the underlying trades. That is a data lineage failure, and it is resolvable in advance.
Holding collateral terms as governed data
The objective is a single authoritative set of terms per relationship, versioned, linked to the executed document and reconcilable against whatever the operational system is using.
- One record per netting set. Structure the data around the agreement that actually governs the exposure, including the specific legal entities on both sides. Entity-level ambiguity is a common source of applying the right terms to the wrong relationship
- Version every amendment. Each amendment to a threshold, eligibility table or haircut should create a new version with an effective date, so that a historical call can be assessed against the terms in force at the time
- Eligibility tables as structured rows. Hold the collateral schedule as rows with instrument type, currency, maturity band and valuation percentage, not as an image or a block of text, because that table is what operations needs and what changes most
- Automated reconciliation to the operational system. Compare the authoritative fields against the collateral system on a regular cadence and report differences. This is a short job that consistently finds items worth finding
- Provenance on every field. Keep the clause reference and the document version against each value. When a counterparty challenges a term, the answer should be a citation rather than a search
A five-step way to close the gap
Step 1, Scope by exposure, not by count
Start with the counterparties that drive most of the collateral exposure. The documentation population is large, the exposure concentration is not, and the first pass should follow the exposure.
Step 2, Extract the key fields with review
Use document extraction to pull thresholds, minimum transfer amounts, independent amounts and eligibility tables, then review every extracted value for the top tier rather than sampling. These are small numbers with large consequences.
Step 3, Reconcile against operations once
Run a full comparison between the extracted terms and the collateral system values. Record every difference with its cause, because the causes cluster and the clusters tell you what to fix structurally.
Step 4, Fix the amendment pathway
Most differences originate in amendments that reached legal and not operations. Put a single route in place so that any change to a collateral term updates the authoritative record and notifies the operational owner.
Step 5, Re-reconcile on a cadence
Make the comparison a scheduled control rather than a project. Quarterly is sufficient for most portfolios, monthly for the largest relationships, and the exercise shortens considerably once the first pass is complete.
Collateral terms with provenance
Aavenir holds thresholds, minimum transfer amounts and eligibility tables as versioned fields linked to the executed annex.
Why this is harder than it looks
Three characteristics make collateral documentation resistant to the usual contract data approach. The first is that the governing terms are spread across a master agreement, a schedule, a credit support annex and often several amendments and side letters, so the effective position is a computation rather than a lookup. The second is that the same commercial relationship may be documented differently across entities, products and jurisdictions, which means the record has to be keyed to the netting set rather than to the counterparty name. The third is that the data consumers are operational systems running on tight daily cycles, and they will not tolerate a source that is slower or less available than the spreadsheet they replaced. Any solution therefore has to be as fast as the thing it displaces, which usually means publishing the authoritative values into the operational system rather than asking operations to look them up. The contract system holds provenance and version history; the collateral system holds the working copy; the reconciliation between them is the control.
Fields, drift risk and the control that catches it
Not every term carries the same risk of divergence. Prioritizing by drift risk makes the first reconciliation pass considerably shorter.
| Field | Drift risk | Usual cause | Control |
|---|---|---|---|
| Threshold | High | Rating-linked or amended per counterparty | Version by effective date, reconcile quarterly |
| Minimum transfer amount | Medium | Re-keyed once, rarely revisited | Include in standing reconciliation |
| Eligible collateral table | High | Replaced by amendment or new annex | Hold as structured rows, version on change |
| Valuation percentages | High | Adjusted without a full annex replacement | Row-level versioning with provenance |
| Notification timing | Low | Stable but entity-specific | Confirm at onboarding and on novation |
Collateral terms are among the few places where a legal document feeds a daily operational process directly, which makes the quality of the handover between them a measurable thing. Holding the terms as versioned, sourced data and reconciling them on a cadence turns a recurring dispute cause into a routine control. The broader approach to regulated commitments is set out in our financial services obligation management datasheet.
For related guidance, see Vendor contract controls for OCC and FFIEC exam readiness, What agentic AI changes in contract operations and How to build a contract metadata model for AI search.
