Product Overview

Trained on your data. Built for action.

Liminal Clarity AI enables organisations to proactively monitor internal and external behaviours using a custom-trained neural engine. Built to detect fraud, corruption, and other high-risk activities, the platform analyses your enterprise data to identify behavioural anomalies and flag suspicious patterns as they emerge.

How It Works

Step 1

DATA
INGESTION

We securely connect to a variety of data sources,

such as:

Email data
Chat integration
HR Employee data
Supplier & Customer records
Transaction data
Legal Registries (i.e. Windeed, DHA, CIPC)

Step 2

NEURAL ENGINE PROCESSING

Our system applies advanced AI techniques, including:

Natural Language Processing (NLP)
Pattern recognition
Anomaly detection
Entity correlation
Behavioural risk modelling
Relationship mapping

Step 3

MONITORING
OUTPUT

The dashboard presents findings, tools and outputs such as:

Networking
Keyword recognition
Entity recognition
Anomaly detection
Network diagramming

Dashboards that surface what matters

Our monitoring dashboard transforms complex data into clear visual intelligence. Alerts, audit trails, timelines, and actor maps help you zero in on risk—without sifting through noise.

Dashboard Screenshot 2_edited_edited.jpg

Use Cases

Liminal Clarity AI is designed to adapt—because no two organisations face risk in the same way. Below are examples of how our neural engine has been deployed to surface invisible behaviour patterns, flag emerging threats, and empower decisive action.

From procurement fraud to ghost employees, these use cases show the engine in action—trained on real data, in real organisations.

Internal Employee Fraud/Theft

Problem: Employees submit false or inflated claims for travel or accommodation.

Data Ingested: Receipts, credit card data, GPS logs, travel bookings, calendar entries

Engine's Task: Look for duplicate submissions, inflated amounts, or mismatched dates/times

Dashboard flags the suspicious claims with LLM-generated rationale
Problem: Fake employees are created pay HR or payroll staff to siphon salaries

Data Ingested: HR records, device logins, badge access, tax records

Engine's Task: Identify employees with zero system activity, no email or login activity, but regular salary disbursements

Dashboard flags ghost employee profile, salary total to date, and system activity summary
Problem: Physical inventory is stolen and masked by falsified logs

Data Ingested: inventory logs, CCTV timestamps, access control, and stock movement vs sales records

Engine's Task: Detect patterns of removal without corresponding sales/shipment activity

Dashboard flags item, warehouse, responsible party, and visual timeline of abnormal stock movements
Problem: Procurement staff accept bribes or kickbacks in return for awarding contracts at inflated prices

Data Ingested: Procurement logs, internal emails/chats and vendor payment trends

Engine's Task: Detect unusual price hikes or repeated awards to one vendor despite negative performance

Dashboard displays supplier/contracts heatmap flagging pricing and behavioural anomalies
Problem: Employees falsify work hours or log fake overtime

Data Ingested: Timesheets, access control logs, login records, task data

Engine's Task: Identify entries where reported hours exceed actual system activity

Dashboard lists user, period in question, discrepancy hours, and AI narrative on mismatch
Problem: Employees alter data to meet KPIs or performance metrics that trigger bonuses

Data Ingested: KPI logs, edit trails, performance dashboards

Engine's Task: Compare historical trends in KPIs vs current surges, especially near bonus cycle; check edit trail

Dashboard shows KPI name, change vs baseline, involved employee, and LLM suspicion score
Problem: Employees approve fake or inflated invoices from vendors in exchange for personal gain

Data Ingested: Vendor onboarding/setups, communication, payment patterns and invoice approvals

Engine's Task: Identify collusion patterns (e.g., familiar tone in emails or hidden links between staff and vendor).

Dashboard generates network graph showing relationship links and message excerpts flagged by LLM
Problem: Employee misuse of company vehicles, credit cards, or equipment for personal benefit.

Data Ingested: GPS logs, fuel card spend, vehicle/device schedules

Engine's Task: Compare scheduled vs actual use of company assets, highlight anomalies like high fuel spend on weekends or non-approved locations

Dashboard provides map view and flagged trips or spend entries, with explanations like “Fuel card used 100km from site.”
Problem: Losses hidden or profits inflated in financial statements

Data Ingested: Financials, audit logs, system edit trails

Engine's Task: Summarise all changes to financials with material impact; flag ones without proper notes or sign-offs

Dashboard provides comparison panel with before/after reports compared; and analyses discrepencies/rationality
Problem: Staff direct contracts to companies in which they or relatives have interests, bypassing due diligence

Data Ingested: Employee declarations, supplier/business registries, comms data

Engine's Task: Scan declarations, emails, supplier addresses for hidden connections or overlaps

Dashboard charts “related party risk” showing potential linked parties with confidence ratings

External Supplier or Customer Fraud/Theft

Problem: Vendors submit multiple or fraudulent invoices for the same goods/services

Data Ingested: Invoice records, purchase orders, payment logs

Engine's Task: Use semantic similarity to flag “near duplicates” with small changes

Dashboard displays side-by-side invoice comparisons with LLM flags
Problem: Vendors invoice for more than they deliver

Data Ingested: Receiving logs, weight data, scanner entries, invoices

Engine's Task: Highlight mismatches with recurring patterns by the same vendor

Dashboard provides visualisation of delivery timelines visualised and mismatch heatmaps
Problem: Customers return damaged, used, or incorrect goods for full refunds

Data Ingested: Return logs, photos, original sale records

Engine's Task: Identify discrepancies in SKUs or repeated return patterns per customer

Dashboard flags customer profile with return fraud risk score and description
Problem: Multiple vendors collude to fix prices and eliminate competitive pricing

Data Ingested: Tender responses, pricing sheets, proposal documents

Engine's Task: Monitor bids received for similar tenders; look for repeated phrases, patterns, and abnormal price consistency across bidders

Dashboard displays network graph of collusion clusters and price variance histogram.
Problem: Customers or suppliers file false claims under warranty policies or insurance arrangements.

Data Ingested: Warranty logs, usage data, complaint records

Engine's Task: Check claim volume against average rates, and compare to usage data or complaint logs, spot claim justification inconsistencies

Dashboard creates risk queue sorted by likelihood of fraud and flags excerpts
Problem: External actors impersonate vendors to divert payments

Data Ingested: Domain records, unusual communication requests, bank detail change requests

Engine's Task: Detect suspicious language shifts in vendor emails (e.g., urgent tone, unusual greetings) and flag metadata mismatches

Dashboard shows Alert box with sender metadata and narrative like “Tone change detected in recent emails.”
Problem: Fake supplier entities bill for non-existent goods or services

Data Ingested: Supplier profiles, registry data, websites

Engine's Task: Detect red flags like no web presence, shared physical address, or new bank accounts

Dashboard flags shell risk indicators across vendor portfolio with checklist
Problem: Vendors intentionally inflate prices or add obscure charges not outlined in contracts

Data Ingested: Contracts, invoice line items, historic pricing data

Engine's Task: Detect fee descriptions that are vague or duplicated across vendors, analyse line-item price across time, identify off-contract charges.

Dashboard provides vendor billing tracker highlighting contract breaches with insights (i.e. “Extra charge not found in master agreement.”)
Problem: Suppliers deliver substandard or counterfeit goods billed as premium products

Data Ingested: Inspection reports, sensor data, delivery records

Engine's Task: Identify mismatches in certification, quantity, or weight logs

Dashboard flags and provides explanation of inspection outcome vs difference from SLA
Problem: Customers use stolen or fabricated identities to secure credit or payment terms, then default

Data Ingested: Customer data from credit bureaus and ID
registries, IP addresses

Engine's Task: validate customer data, compare application narrative to known fraud patterns

Dashboard flags mismatched metadata by presenting a fraud risk scorecard + reasoning e.g. “IP address in Nigeria, applicant in Cape Town.”

Adaptable Intelligence, Aligned to your Business

Liminal Clarity AI’s neural engine is designed to be trained and refined using your organisation’s unique data. This ensures the system aligns with your specific processes, risk landscape, and monitoring priorities.

Each deployment is tailored to ingest the data sources most relevant to your environment, enabling a focused, high-precision approach to behavioural risk detection.

While current implementations are centred on fraud, theft, and corruption, the platform is fully configurable. With the right data inputs, it can be extended to support additional risk domains as your needs evolve.

Compliance without Complexity

Our platform helps satisfy Section 34 of South Africa's PRECCA (the Prevention and Combating of Corrupt Activities Act), which makes mandatory the reporting of fraud and corruption over R100,000.

By automating the early detection of suspect behaviours, your business becomes not just compliant—but protected.

Designed with Security and Scale in Mind

Whether deployed on-premise or in secure cloud environments, the platform is built to handle high-volume, sensitive data without compromising performance or confidentiality. Each solution is tailored to your infrastructure, access controls, and compliance parameters.