Relationship Intelligence for Financial Institutions

A Relationship
Intelligence Layer
Above Core Banking

Explicit, traceable analysis across connected member, account, loan, and household relationships.

Evaluate on Your Data How It Works
How Connected Analysis Works

One default. Four connected exposures.

A borrower misses payments on an auto loan. In a traditional system, that is one delinquent loan. In a relationship graph, it is the start of a traceable exposure path.

Traversal: 3 hops from source record
1
Member A: Auto Loan delinquent
Source record: the loan is linked to Member A in the graph. Status, balance, and payment history are stored on the loan record, traceable to the core export.
2
Spouse B discovered via spouse relationship
Hop 1: traverse the spouse relationship. Spouse B has their own accounts and loans. Household exposure now includes both members.
3
Joint Checking Account via account ownership
Hop 2: both members hold a joint checking account. The account links to both owners in the graph. Balance is shared exposure.
4
Home Equity Loan via collateral relationship
Hop 3: the joint account secures a home equity loan as collateral. Total connected exposure across 3 hops is now visible in one view. Every finding links back to source records.

This traversal uses real relationship types from the NexiNexta graph model: loan ownership, spouse, account ownership, and collateral. Each hop produces a traceable path that can be reviewed and verified against the source export.

Why a Relationship Graph

One governed model of your members’ relationships

Members, accounts, loans, businesses, and the people connected to them, modeled once and reused by every analysis on the platform.

Defined Once
A shared relationship vocabulary
Members, accounts, loans, cards, businesses, and related parties are connected through explicit, named relationship types under a governed, FIBO-informed vocabulary. Defined once, reviewed once, and queried from any starting point.
Proven, Not Guessed
Every relationship earns its place
Every relationship is proven from a source record before it enters the graph. Unproven relationships are rejected. What the graph holds, the institution can defend.
Households, Automatically
Connections surface themselves
Signals including joint accounts, co-borrowers, beneficiaries, trustees, guarantors, and shared collateral connect members into households and connected-obligor groups. No manual stitching, and every connection traces to the record that produced it.
Why the Graph Matters for AI
The graph is what keeps the AI honest
  • •Every question the assistant asks is validated against the graph model before it runs. The AI cannot query a relationship that does not exist.
  • •Answers are computed by traversing proven relationships. The AI narrates the result. It does not generate it.
  • •If a relationship is not in the graph, it does not exist for the assistant. There is nothing to hallucinate against.
  • •Every answer carries its path: which members, which relationships, which source records. Reviewers check the path, not the prose.

NexiNexta does not replace warehouse reporting. It provides explicit reusable relationship structure for connected analysis that complements what the warehouse already does.

Why Governed Meaning Matters

When “the model inferred this” is not an answer

For regulated financial analysis, the value of a semantic layer is only as durable as its provenance story. Statistical inference produces confidence scores. Governed mappings produce lineage paths that examiners can follow.

Dimension Auto-Generated Semantic Layer NexiNexta: FIBO-Informed Graph
How meaning is established Statistically inferred from usage or authority signals Hand-grounded FIBO class and property mappings reviewed by domain experts
Provenance “The model inferred this” Every vertex and edge traces to a source row and foreign key from the export
Reproducibility Varies as training data or usage patterns shift Same inputs produce the same graph and the same answer, every time
Auditor answer to “why?” A confidence score or authority ranking A canonical mapping and a complete lineage path to the source export
Best fit “What is our definition of ARR?” “Is this member’s household exposure $X, and prove it.”
In compliance An unexamined vendor risk The foundation of the product
NCUA 2026 Supervisory Priorities

NCUA examiners will evaluate credit risk, including allowance for credit losses, and whether BSA/AML programs are tailored to each institution’s risk profile. Any methodology driving those estimates must be documentable on demand.

NCUA Letter to Credit Unions, Jan. 14, 2026
NCUA Guidance on AI

NCUA supports the use of AI when implemented in a safe, sound, and compliant manner. Existing regulations are technology neutral and apply to AI use. Credit unions remain fully responsible for due diligence over third-party AI tools, including understanding how those tools reach their outputs.

NCUA, Artificial Intelligence Regulatory & Compliance Resources, ncua.gov
GAO-25-107197, May 2025

NCUA lacks statutory authority to examine third-party technology service providers. Institutions using external AI tools for compliance output bear the vendor risk that examiners cannot independently assess.

GAO, Artificial Intelligence: Use & Oversight in Financial Services

Platform Capabilities

What the platform computes

Deterministic modules, each producing outputs traceable to source records. The AI assistant surfaces these through natural-language queries. It narrates results. It does not compute them.

CECL Engine
Allowance for Credit Loss
Loan-level ACL estimates using readable, auditable calculation logic. Every input and output traces to the source loan export, reviewable by examiners without additional translation.
NBA Engine
Next-Best-Action
Rules-based household opportunity prioritization, ranked by eligibility, propensity, and expected value. The propensity score comes from a versioned scorecard that can be recomputed by hand, uses no runtime model call, and excludes protected attributes such as age, marital status, gender, race, and income by design.
Risk Analytics
Contagion & Exposure
Connected-party exposure simulation across member households. Traces cascading default paths through co-borrower and collateral chains with source-record linking.
Household Analytics
Multi-Hop Traversal
Spouse, joint-holder, co-borrower, and signer relationships modeled once and reused across every analysis. Every relationship hop traces to source records.
Analysis Assistant
Role Lenses
Relationship Manager, Credit Officer, Collections, Compliance/BSA, and Executive lenses each route natural-language questions to the appropriate deterministic engine. The AI narrates; the engine computes.
Audit Trail
WORM-Logged Operations
Hash-chained, tamper-evident audit log on AI operations, CRM actions, and rules-based decisions. No autonomous writes. Every CRM action requires explicit review and confirmation before execution.
Architecture

Where NexiNexta Fits

A relationship intelligence layer that sits above your core systems. It reads exported data. It does not write back to core.

Data Sources (Read-Only Exports)
Core Banking
Members, accounts, loans
Data Warehouse
Reporting, batch analytics
Loan Origination & Cards
Applications, co-borrowers, collateral
NexiNexta
Financial Relationship Graph  ·  Rules Engine  ·  Analysis Assistant
Outputs
Risk & Exposure
Household and connected exposure
Growth Signals
Rules-based opportunity prioritization
Reviewable Reports
Traceable to source records
Analysis Assistant

What is deterministic. What is AI-assisted.

The platform separates deterministic logic from AI-assisted investigation. Both produce reviewable outputs.

Deterministic (Rules-Based)
  • ✔All relationship traversal and graph queries
  • ✔Exposure calculations and risk scoring
  • ✔Trigger rules and opportunity prioritization
  • ✔Propensity scoring and expected-value ranking from a versioned scorecard, with no runtime model call
  • ✔Layered data quality validation before every graph build
  • ✔Source-record linking and audit trail
AI-Assisted (Governed)
  • •Natural-language query translation to graph traversal
  • •Queries validated against the graph schema before execution
  • •Results come from the modeled graph, not generated content
  • •Structured member PII, including Social Security numbers, account numbers, contact details, and dates of birth, scrubbed in-process before data reaches the language model
  • •All AI operations logged in a hash-chained, tamper-evident audit trail
  • •No autonomous writes. CRM actions require explicit confirmation before execution
  • •All outputs reviewable and traceable to source records

The Analysis Assistant does not make credit, compliance, or underwriting decisions. It supports natural-language investigation over governed, modeled data, with structured PII scrubbing and injection controls applied in-process before every language model call. Final judgment remains with the institution.

Security & Governance

Built for regulated financial institutions

Each institution receives a single-tenant deployment on Microsoft Azure with no shared infrastructure across clients.

Isolation & Access
  • ✔Single-tenant Azure deployment per institution
  • ✔No cross-institution data sharing
  • ✔Role-based access control
  • ✔VPN-gated administrative access
Data Protection
  • ✔Encryption at rest and in transit (TLS)
  • ✔Zero public network access on all data stores
  • ✔Private endpoints for database connectivity
  • ✔Managed Identity authentication (no stored access keys)
Audit & Compliance
  • ✔All analysis outputs traceable to source records
  • ✔Audit logging on AI operations, CRM actions, and rules-based decisions
  • ✔Controlled export and review workflows
Infrastructure
  • ✔Infrastructure defined as code (reproducible per client)
  • ✔Health monitoring with automated alerting
  • ✔Private telemetry ingestion (no public monitoring endpoints)

Full security architecture documentation and infrastructure details are provided during evaluation under NDA.

How It Works

From systems of record to reviewable results

A defined path from your systems of record to reviewable relationship analysis, with an explicit remediation loop. No real-time integration required.

1
Curate
Canonical data is curated from your systems of record: core banking, loan origination, card processing, and digital banking activity. Read-only extracts delivered over secure transfer. No live integration, and nothing writes back to source systems.
2
Map
Source fields are mapped to the governed relationship vocabulary. Known core systems start from pre-built mappings. Anything ambiguous is flagged and proposed for human review, never guessed. Confirmed decisions become reusable mappings, so each review happens once.
3
Validate
Data quality gates check primary key uniqueness, foreign key integrity, enum validation, entity type completeness, and referential completeness. The pipeline fails closed: data that does not pass does not move forward.
4
Remediate
Every issue lands in a resolution report with its cause and its fix: a missing code table, an unmapped field, a broken key reference. Your team corrects or confirms, the pipeline re-runs, and the loop repeats until the data is clean. Nothing is silently patched.
5
Build the Graph
Clean, validated data rolls onto the financial relationship graph. Every entity and relationship is proven from a source record before it enters, and unproven relationships are rejected. Health checks run after the load to certify that what landed matches what was sent.
6
Analyze + Review
Rules-based analysis evaluates exposure, triggers, and opportunity signals against the graph. Results are reviewed in a dashboard with source-record linking, the Analysis Assistant answers natural-language questions over the modeled graph, and outputs export to downstream workflows.
What Changes for Your Team

Specific outcomes from relationship structure

Find connected exposure that siloed systems miss
Household and co-borrower exposure is computed from explicit graph relationships, not manual SQL joins. A single delinquent loan surfaces all connected positions across the household.
Reduce manual relationship stitching
Spouse, joint-holder, co-borrower, and signer relationships are modeled once and reused across every analysis. Analysts do not rebuild these connections per report.
Surface household opportunities missed in silos
Next-best-action rules evaluate the full household picture: what products both spouses hold, what gaps exist, and whether the household can support the recommendation.
Improve review consistency with traceable outputs
Every finding links back to source records. Review workflows show the relationship path, the rule that fired, and the data that triggered it. Consistent across analysts and reviewers.
What NexiNexta Is Designed To Do
  • ✔Structure connected financial relationships for governed analysis
  • ✔Support review workflows with traceable, source-linked outputs
  • ✔Surface exposure and opportunity signals across connected records
  • ✔Complement warehouse reporting with reusable relationship structure
What NexiNexta Does Not Replace
  • –Core banking systems
  • –Institution underwriting authority
  • –Compliance judgment or final decisions
  • –Existing data warehouse reporting
Get Started

Structured Pilot on Your Data

The pilot uses your exported data to build a relationship model, validate outputs against source records, and produce reviewable results your team can inspect.

Pilot Inputs

Curated extracts from your systems of record, mapped to the governed relationship model.

Typical data:
  • ▸Members, accounts, loans
  • ▸Co-borrowers, spouses, joint holders
  • ▸Cards, products, insurance
No API integration required for the structured pilot.
What You Get Back
  • ✔Household and connected exposure analysis verifiable against source records
  • ✔Relationship opportunity list with reviewable rules-based prioritization
  • ✔Dashboard access for review during evaluation
  • ✔Data quality report from the validation pipeline
Scoped
Defined timeline
Configurable
Schema-adaptive mapping
Single-Tenant
Isolated Azure environment
1 Sponsor
CRO, CLO, or CIO

Evaluate NexiNexta on your data

See household exposure, connected exposure paths, and relationship-based opportunity signals built from your institution's exported data.

Contact Us

Designed for evaluation by CIO, CRO, CLO, and analytics leaders.