Knowledge Hub
Plain-English definitions of the terms that matter most in AI-powered financial operations — from AML and FCA compliance to transaction categorisation, explainable AI, and ML orchestration.
The set of laws, regulations, and procedures designed to prevent criminals from disguising illegally obtained funds as legitimate income. In AI-powered platforms, AML refers to automated transaction monitoring, suspicious activity detection, and watchlist screening.
The process of classifying bank transaction data into income and expenditure categories to assess a borrower's ability to repay a loan. Accuracy is critical under FCA Consumer Duty rules — Trustinera AI delivers 97%+ accuracy on UK open banking data.
A structured collection of records documenting every AI decision, reconciliation outcome, model version, and risk score for a given period. Required by FCA examiners during supervisory reviews. Trustinera AI generates audit packs in under 4 hours.
An immutable, time-stamped record of every action, decision, or data change in a financial system. A complete audit trail allows regulators and auditors to trace any output back to its raw input data and the exact model version that produced it.
The UK payment scheme for processing direct debits and direct credits. BACS payments typically take 3 business days to process and are commonly used for payroll, supplier payments, and bill collection. Trustinera AI natively categorises BACS transaction references.
A live or near-live data connection between a bank account and an accounting or intelligence platform. Open banking enables automated bank feeds. Trustinera AI ingests bank feeds from Barclays, HSBC, Lloyds, Starling, Monzo, and 40+ other providers.
The ability to deploy and run your own custom-trained machine learning model on a third-party platform. Trustinera AI's TrustFlow module supports BYOM with ONNX and TorchScript model formats, promotion gates, and full versioning.
The UK's high-value, same-day payment scheme, typically used for large business payments and property transactions. CHAPS payments carry high reputational and fraud risk and receive elevated scrutiny in AML screening.
The percentage of transactions that the AI model assigns to the correct category when compared against a labelled ground truth. Trustinera AI achieves 95–99% accuracy out of the box on UK financial transaction data.
A numeric value (0 to 1) output alongside every Trustinera AI categorisation decision, indicating the model's certainty about its prediction. Items below your configured threshold surface for human review. The confidence score is included in every audit response.
The FCA's overarching principle requiring firms to deliver good outcomes for retail customers. Under Consumer Duty, lenders must be able to demonstrate their affordability assessments are based on accurate, explainable data — making AI categorisation auditability a regulatory necessity.
The Trustinera AI endpoint that accepts human overrides to model decisions. When your team corrects a categorisation, the correction feeds into the continuous training pipeline, improving accuracy over time. All corrections are logged in the audit trail.
The requirement that data be processed and stored within a specific geographic jurisdiction. Trustinera AI processes and stores all customer data in UK data centres by default, with EU residency available on request.
A contract between a data controller (your firm) and a data processor (Trustinera AI) governing how personal data is handled. A DPA is required under UK GDPR for any processor relationship involving personal data.
AI systems that can describe, justify, or explain their outputs in human-understandable terms. Trustinera AI provides a per-decision explanation including the top contributing factors, confidence score, and model version for every categorisation — the standard FCA examiners now expect.
The list of transactions or reconciliation items that fell below the configured confidence threshold or failed automated matching. Trustinera AI surfaces only the genuinely uncertain items to human reviewers — typically 2–5% of total volume.
The UK's financial services regulator, responsible for overseeing around 45,000 firms. FCA-supervised firms must maintain documented AI decision trails, audit-ready records, and demonstrate Consumer Duty compliance. Trustinera AI is purpose-built for FCA-supervised environments.
The UK real-time payment scheme enabling near-instant bank transfers (typically seconds) up to £1 million. Faster Payments data is commonly processed by Trustinera AI for real-time fraud scoring and affordability categorisation.
The process of transforming raw data into inputs that improve model performance. Trustinera AI's TrustFlow module includes a feature store that manages, versions, and serves engineered features across model training and inference runs.
A centralised repository that stores, manages, and serves machine learning features for training and serving. Trustinera AI's feature store maintains full lineage of every feature, enabling complete reproducibility of any model decision.
The proportion of legitimate transactions incorrectly flagged as suspicious or high-risk. High false positive rates waste compliance team time and damage customer experience. Trustinera AI customers typically reduce their false positive rate from 15–20% to under 4%.
The EU data protection regulation (retained in UK law post-Brexit as UK GDPR) governing the collection, storage, and processing of personal data. Trustinera AI processes financial data under your instruction as a data processor, with a DPA available for all paid plans.
The policies, controls, and processes that ensure AI systems operate within defined boundaries, with documented decision-making and human oversight. In Trustinera AI, governance includes policy gates, audit event streams, model promotion controls, and period-close reporting.
The degradation in model performance that occurs when the real-world data distribution changes over time. Trustinera AI's OpsIQ module monitors for model drift, alerts teams when accuracy drops below configured thresholds, and automatically routes affected items to the exception queue.
A complete record of the training data, features, hyperparameters, and code version used to produce a given model. Trustinera AI stores full model lineage for every deployed version, enabling any output to be traced back to its exact training provenance.
The practice of combining ML development with deployment, monitoring, and governance operations — analogous to DevOps for software. Trustinera AI's TrustFlow module provides a full MLOps stack purpose-built for regulated financial environments.
The US Treasury agency that administers economic and trade sanctions. OFAC maintains the Specially Designated Nationals (SDN) list. Trustinera AI's Sentrise module screens transactions against the OFAC SDN list and other international sanctions lists in under 100ms.
The regulatory framework (driven by PSD2 in the EU/UK) requiring banks to provide secure API access to customer account data to FCA-registered third parties. Trustinera AI uses open banking connections to ingest live bank feeds for real-time categorisation.
The process of checking whether a customer or counterparty is a politically exposed person — someone who holds or has held a prominent public position. PEP status increases financial crime risk. Trustinera AI screens against global PEP databases in real time.
The process of finalising all financial data, reconciling accounts, and producing reports at the end of a financial period (typically month-end or year-end). Trustinera AI reduces period close from 5 days to 1 day by automating bank reconciliation and exception handling.
The EU directive (transposed into UK law) establishing the regulatory framework for open banking, strong customer authentication, and third-party access to payment accounts. PSD2 is the foundation for the bank feed data Trustinera AI processes.
The process of matching bank statement transactions against ledger entries to identify discrepancies. Trustinera AI's Reconcile module automates this process using AI matching, surfacing only genuine exceptions for human review.
A statutory report filed with the National Crime Agency (NCA) in the UK when a firm has knowledge or suspicion of money laundering or terrorist financing. Trustinera AI's Sentrise module generates SAR-ready case files from flagged transaction clusters.
A third-party audit framework assessing a service provider's controls around security, availability, processing integrity, confidentiality, and privacy over a period of at least 6 months. Trustinera AI holds SOC 2 Type II certification, independently audited annually.
The process of assigning a descriptive label (e.g. "Utilities", "Rent", "HMRC VAT", "Payroll") to a financial transaction based on its reference, amount, counterparty, and pattern. Trustinera AI's Categorise module achieves 95–99% accuracy on UK transaction data.
Trustinera AI's ML pipeline orchestration module. TrustFlow manages the full ML lifecycle: feature engineering, model training, A/B experiments, artifact versioning, policy-gated promotion, and agent deployment — with complete audit lineage at every step.
The automated check of customers, counterparties, or transactions against lists of sanctioned entities, PEPs, or adverse media. Trustinera AI's Sentrise module screens in under 100ms against OFAC, HM Treasury, UN, EU, and custom watchlists.
No matching terms found.
Free · No gatekeeping
Get the Trustinera editorial briefing on AI governance, explainability, and operational trust. No fake survey numbers, no credit card.
Check your inbox for the briefing link and a recommended next step.
Read the full report online →