InSpective

Tier 4 · ODS & Parties

Party & data integrity

The Operational Data Store (ODS) is where party data quietly decays: the same organisation recorded four times under four spellings, connections pointing at records that no longer exist, external references that cannot be tied back to the practice-management system, and organisations that have sat on both sides of the firm's work. Each of those defects surfaces later as a mis-addressed letter, an unreliable report, or a conflict check that misses a party. This audit finds them first, conservatively, and puts every proposed fix in front of a human.

T4Intelligence

1What we examine

roadmapDuplicate organisations, guardrail first

Organisation records are grouped by company reference number, so case, punctuation, and suffix variants (Pty / Proprietary / Limited) of the same entity land in one review group. The guardrail is deliberate: identically named organisations with different reference numbers are rejected as distinct legal entities, and nothing is ever merged automatically.

roadmapConnection integrity

Orphaned edges: connections still pointing at party records that no longer exist. Every orphaned edge is a matter whose party picture is quietly incomplete, which means reports and conflict searches built on the graph are incomplete too.

roadmapPractice-management reconciliation

ODS external references reconciled against the practice-management system (PMS). We look for synthetic references (placeholder values that were never replaced) and non-unique references (one PMS number shared by several ODS records), both of which break the round trip between your matter system and the system your finance team trusts.

roadmapInactive parties on live matters

Parties flagged inactive yet still connected to open matters. An inactive flag on a live participant means correspondence, reporting, and screening can all silently skip a party who is still in the matter.

roadmapActed-both-sides screening

Organisations appearing in acting roles (client, insurer, defendant) on some matters and opposing roles (plaintiff, third party) on others, ranked by the smaller side so the list starts with the candidates most worth a conflict reviewer's time. These are screening candidates for your conflicts process, never conflict determinations.

2How we examine it

Entity-resolution and graph lenses run over ODS parties and connections on an on-premise stack (client and third-party personal data, so this tier carries the full authority gate and a controlled environment only). Duplicate grouping keys on the company reference number rather than name similarity, which is what makes the no-auto-merge guardrail enforceable; both-sides candidates are derived from the connection graph's role assignments. Every group, rejection, and screening candidate is listed record by record in the report.

Duplicate groups key on the company reference number, not name similarity: identical names with different references are rejected as distinct entities, and nothing is ever auto-merged. Both-sides hits are screening candidates for your authoritative conflicts process, never determinations. Record, group, and rejection counts are observed over the corpus; the productised lens is on our roadmap, generalised from analysis proven on one live engagement.

3Example finding

Illustrative example from the synthetic demonstration corpus (underlying lens on our roadmap, not yet in the productised pipeline)

Across 30,351 organisation records, 333 duplicate groups carried 664 redundant records. The guardrail refused 380 look-alike name groups that shared a spelling but not a company reference number, and the connection graph surfaced 1,305 organisations that have acted on both sides.

  • 664 redundant records across 333 duplicate groups, in a register of 30,351 organisations
  • 380 same-name, different-reference groups deliberately left unmerged: distinct legal entities the guardrail protected
  • 1,305 acted-both-sides organisations ranked by their smaller side, ready for the conflicts team
  • Every proposed merge is listed record by record; a human confirms each one
30,351organisation records in the register
664redundant records across 333 duplicate groups
380look-alike groups the guardrail refused to merge
1,305acted-both-sides screening candidates
Duplicate groups found333Rejected look-alikes380
Grouped for merge review. One company reference number (74480741) recorded as 4 separate ODS records touching 6 matters:
  • Stonecrest Contracting Pty Ltd
  • STONECREST CONTRACTING
  • stonecrest contracting
  • Stonecrest Contracting Pty. Ltd.

One legal entity, one proposed merge, listed record by record for a human to confirm.

Refused by the guardrail. Two records that read as the same organisation:
Recorded nameCompany reference
Bluegum Agribusiness Pty. Ltd.51483824
Bluegum Agribusiness Pty Limited33942778

The names match to the eye; the company reference numbers do not. Two distinct legal entities, deliberately left unmerged.

See it in the sample report. The entity resolution chapter walks the duplicate groups and rejected look-alikes record by record, and the conflict radar ranks every acted-both-sides candidate with its for and against matter counts.
Base rates above; merge candidates versus guardrail rejections in the chart (green is the guardrail working); one worked merge group and one worked rejection below it.

4The benefit

What you walk away with

A party register your letters, reports, and conflict checks can rely on: duplicates grouped for one-pass review with no dangerous auto-merges, external references that reconcile to your practice-management system, orphaned and inactive-party defects listed for cleanup, and a ranked both-sides screening list your conflicts team can clear.

← All audit domains See it in a full report