Data Intelligence
Platform

Prime Capital Bank has embarked on a transformational journey to consolidate its fragmented data landscape into a single, governed, cloud-native data intelligence platform. Operating across 200 branches with 500,000 active retail, commercial, and corporate customers, and managing a portfolio exceeding R2.3 trillion in assets under management, the bank required a platform capable of meeting the dual demands of real-time operational intelligence and rigorous regulatory compliance.

The legacy environment comprised more than a dozen isolated system-of-record databases — Temenos T24 core banking, Salesforce CRM, a proprietary card management system, Kondor+ treasury, Oracle Financials general ledger, and multiple third-party data feeds from Reuters, the South African Reserve Bank (SARB), and TransUnion credit bureau. Data resided in siloed on-premises servers, batch-transferred nightly through brittle ETL pipelines, with analytical workloads competing for OLTP resources and risk models running days behind the market.

Key Business Outcomes

The Data Intelligence Platform has delivered measurable outcomes across four strategic dimensions: operational efficiency, risk management, regulatory compliance, and customer intelligence.

The two charts below break down the platform’s most critical headline numbers: how the bank’s four primary loan product categories (representing R312.6 billion of the R380 billion total loan book) are distributed, and how the bank’s capital ratios compare to SARB minimum requirements.

Platform Components

The platform is built on Azure Databricks with a Medallion (Bronze → Silver → Gold) architecture hosted on Azure Data Lake Storage Gen2 (ADLS Gen2). Data transformation is managed through dbt Core running inside Databricks notebooks, with data quality enforced by Great Expectations. The governance layer is powered by Unity Catalog, providing column-level security, data lineage, and PII masking at the platform level. Machine learning workloads leverage MLflow and the Databricks Feature Store, with 12 production models serving real-time predictions via Azure API Management.

Business intelligence is delivered through Power BI Premium with 8 purpose-built dashboards serving executive leadership, risk management, compliance, branch operations, and treasury. Regulatory reporting — including SARB BA700, IFRS 9 ECL provisioning, Basel III capital adequacy, and FICA/AML returns — is fully automated within the platform, reducing the monthly close cycle from 5 business days to less than 6 hours.

Strategic Importance

In an increasingly competitive South African banking landscape — with challenger banks, digital wallets, and open banking pressures — Prime Capital Bank's data platform provides a durable competitive advantage. The ability to compute real-time credit scores at point-of-application, detect fraudulent card transactions within 50 milliseconds, and generate SARB-ready regulatory returns at the push of a button positions the bank at the forefront of data-driven banking in sub-Saharan Africa.

The platform is designed for scale: the Databricks cluster configuration supports linear horizontal scaling to accommodate projected 5-year data growth of 300%, while the Unity Catalog governance model ensures that every new data asset inherits the same security, lineage, and quality standards established at platform launch.

The Prime Capital Bank Data Intelligence Platform is designed around the Medallion Architecture pattern — a multi-layered data organisation framework that progressively refines raw source data through Bronze, Silver, and Gold quality tiers. This architecture, hosted entirely on Microsoft Azure, provides a clear separation of concerns between ingestion, transformation, and consumption, enabling independent scaling and governance at each layer.

Figure 2.1 — Prime Capital Bank Azure Data Platform Architecture. Source systems on the left feed ADLS Gen2 storage zones (centre-left), processed by Databricks notebooks (centre), governed by Unity Catalog, and served to consumption layers (right).

Azure Services Deployed

Data Flow Architecture

Source data enters the platform through two primary ingestion mechanisms. Batch ingestion uses Azure Data Factory (ADF) pipelines and Databricks Auto Loader to pull structured data from source systems on defined schedules — typically hourly for core banking and card transactions, daily for HR and GL, and monthly for external credit bureau feeds. Streaming ingestion routes card transaction events and ATM activity through Azure Event Hub, processed by Databricks Structured Streaming notebooks into the Bronze layer within seconds.

All data at rest is encrypted using AES-256 via Azure Storage Service Encryption. All data in transit is protected by TLS 1.3. Azure Private Endpoints ensure that ADLS Gen2 and Databricks communicate exclusively over the bank's private VNET, with no public internet exposure. Network Security Groups (NSGs) enforce allow-list-only ingress and egress rules for each subnet.

Network & Identity Design

The Databricks workspace is deployed in a customer-managed VNET (VNet injection) with separate subnets for the driver and worker node pools. Azure Private Link connects Databricks to ADLS Gen2, Azure SQL, Key Vault, and Event Hub — ensuring all data movement stays within the Azure backbone. All user access is mediated through Azure Active Directory with Privileged Identity Management (PIM) for Just-In-Time elevation of data engineer and DBA roles.

The Bronze layer is the first tier of refinement in the Medallion architecture. Its primary purpose is to faithfully capture and persist data from source systems with the minimum possible transformation — preserving the original schema, data types, and even known data quality issues for full traceability. Bronze is the system of record within the data lake; if a discrepancy arises between the platform and a source system, Bronze is the forensic reference.

Ingestion Mechanisms

Databricks Auto Loader (cloudFiles format) handles the majority of file-based ingestion from the Raw zone of ADLS Gen2. Auto Loader provides incremental processing with exactly-once semantics via a checkpoint mechanism stored in ADLS, making it resilient to cluster failures. Schema inference and evolution are managed automatically — when a source system adds a new column, Auto Loader detects the schema change and uses column merging to extend the Bronze Delta table without data loss or pipeline failure.

For core banking and card system feeds arriving as structured database exports, Azure Data Factory orchestrates the movement from on-premises SQL Server instances to the Raw zone via a Self-Hosted Integration Runtime (SHIR) installed in the bank's data centre. For real-time card transactions, Databricks Structured Streaming consumes from Azure Event Hub using the Kafka consumer API.

Bronze Table Inventory

The chart below visualises the relative scale of each Bronze source table, making the volume disparity immediately clear and highlighting which feeds drive the platform’s throughput requirements.

Schema & Partitioning Strategy

Bronze tables are partitioned by _ingestion_date (date of load) rather than source business date, ensuring partition pruning works effectively even when source systems deliver late-arriving data. High-volume streaming tables (card transactions, EFT) are additionally partitioned by hour to enable efficient time-range queries during fraud investigation workflows.

Delta Lake's OPTIMIZE and ZORDER commands are scheduled nightly on high-query Bronze tables to compact small files produced by streaming micro-batch writes. The VACUUM command retains 30 days of Delta history, supporting point-in-time recovery and audit queries within the regulatory retention window.

The Silver layer transforms raw Bronze data into a clean, conformed, business-meaningful dataset. Silver applies the bank's master data management (MDM) rules, data quality frameworks, and business logic — producing a set of tables that are source-agnostic and conforming to the bank's canonical data model. Silver is the primary input for the Gold analytics layer and for ad-hoc data science exploration.

Transformation Categories

Silver transformations are organised into four categories, each implemented as a discrete set of dbt models executed in dependency order by Databricks Workflows:

• Cleansing — null handling, format standardisation (dates to ISO 8601, amounts to ZAR base currency), whitespace trimming, duplicate elimination via row_number() OVER (PARTITION BY ... ORDER BY _ingestion_timestamp DESC) deduplication logic
• Conforming — applying the bank's reference data (branch codes, product codes, risk codes) to map source-system identifiers to canonical values stored in the silver.reference schema
• Enriching — joining customer demographic data with credit bureau scores, appending ML-generated risk scores, and resolving customer identity across source systems using the Enterprise Customer Master
• PII Masking — South African ID numbers are tokenised using HMAC-SHA256 with a Key Vault-managed key; full names are masked using format-preserving initialisation; account numbers are partially masked for display contexts

Data Quality Framework

Silver quality gates are enforced using Great Expectations suites embedded in each transformation notebook. Quality checks include:

SCD Type 2 Implementation

Customer, account, and employee dimensions are implemented as Slowly Changing Dimension Type 2 (SCD2) entities at the Silver layer, using the MERGE INTO command on Delta Lake tables. When a source record changes (e.g., a customer changes their residential address or credit segment), a new version row is inserted with an updated scd2_effective_date, the previous row's scd2_expiry_date is set to the change date minus one day, and the is_current_flag is updated accordingly. This approach preserves full change history for regulatory audit and enables point-in-time portfolio analysis — critical for IFRS 9 stage migration reporting.

Silver Table Summary

The Silver layer comprises 42 tables across 6 domain schemas: silver.customers, silver.accounts, silver.transactions, silver.loans, silver.treasury, and silver.reference. Total compressed storage is approximately 22TB across all partitions, with a 60-day active query window supported by Delta Lake's time-travel capability and ADLS Cool tier lifecycle management for data older than 90 days.

The Gold layer is the analytics-serving tier of the platform — purpose-built for high-performance querying by Power BI, Azure SQL, data scientists, and regulatory reporting engines. Gold implements a classic Kimball star schema with denormalised dimension tables and centralised fact tables, partitioned and Z-ordered for sub-second query performance at petabyte scale.

Figure 5.1 — Gold Layer Star Schema ERD. Three central fact tables surrounded by eleven dimension tables. All foreign key relationships shown. SCD Type 2 dimensions maintain full history.

Fact Tables

fact_transaction

The highest-grain fact table in the schema, storing 2.1 billion individual financial transactions spanning all channels: teller, ATM, online banking, mobile, POS, and EFT. Each row represents one debit or credit event against an account. Key measures include transaction_amount_zar, running_balance_zar, fraud_score, and aml_alert_flag. Partitioned by transaction_date and Z-ordered by customer_sk, date_sk to optimise customer-level time-series queries used in the Customer 360 dashboard.

fact_loan_portfolio

Monthly snapshot of the bank's R380 billion loan book — retail mortgages, vehicle finance, personal loans, credit cards, and commercial credit facilities. Each row represents one loan account at month-end, with IFRS 9 Stage classification (1/2/3), PD score, LGD estimate, and ECL provision in ZAR. Partitioned by snapshot_month for regulatory reporting. Total row count: approximately 85 million rows spanning 36 months of history.

fact_account_balance

Daily snapshot of every active account's opening balance, closing balance, interest accrued, and days-past-due status — 730 million rows covering 36 months of daily snapshots across 1.8 million accounts. Used primarily by the Treasury dashboard for liquidity analysis and the Branch Performance dashboard for deposit book tracking. Partitioned by balance_date and Z-ordered by branch_sk, product_sk.

Dimension Tables

Partitioning & Performance

All Gold tables use liquid clustering (Databricks Runtime 13+) on the highest-cardinality filter columns, replacing static partitioning for dimension tables. Fact tables retain date-based partitioning for partition pruning in regulatory reporting queries, supplemented by Z-order on the most common join keys. The Photon vectorised query engine typically delivers 3–8x query speedup over standard Spark SQL for analytical aggregations on Gold tables.

dbt Core (data build tool) is the transformation engine for the Silver and Gold layers. Running natively inside Databricks notebooks via the dbt-databricks adapter, dbt provides SQL-first transformations with built-in dependency management, automated documentation, and a rich testing framework — replacing hundreds of ad-hoc Spark notebooks with a structured, version-controlled, testable codebase.

Project Structure

Testing Strategy

dbt tests are categorised into four types, with 1,847 total tests across the project:

• Schema tests (not_null, unique, accepted_values, relationships) — 1,240 tests covering all primary keys, foreign keys, and enumerated columns
• Custom SQL tests — 312 business rule tests (e.g., "ECL provision must be non-negative", "Stage 3 PD must exceed 50%", "transaction amounts must balance to GL journal totals")
• Freshness tests — 180 source freshness checks ensuring Silver inputs are within SLA before Gold build commences
• Singular tests — 115 full-table assertion queries for cross-table consistency (e.g., "every fact_loan_portfolio record must have a matching dim_customer record")

Lineage & Documentation

dbt generates a fully interactive lineage DAG documenting every model, its upstream sources, downstream consumers, column descriptions, and test results. This lineage is surfaced in Unity Catalog, allowing data consumers to trace any Gold metric back to its Bronze source record — a requirement for IFRS 9 audit trails and SARB regulatory inspections. All 54 models and 847 columns are documented with business-friendly descriptions maintained in schema.yml files, auto-published to the dbt documentation site hosted on Azure Static Web Apps.

The ML capability of the Data Intelligence Platform represents the bank's most significant analytical investment. Twelve models are deployed to production via MLflow Model Registry on Databricks, served through Azure API Management as REST endpoints for real-time scoring and as batch jobs for portfolio-level calculations. All models are governed by the bank's Model Risk Management (MRM) Policy, requiring independent validation, champion-challenger testing, and quarterly performance reviews.

IFRS 9 ECL Model

The Expected Credit Loss (ECL) model is the regulatory centrepiece of the ML suite. It combines the PD, LGD, and Exposure at Default (EAD) models to compute the three-stage ECL provision required under IFRS 9 — Financial Instruments. Stage 1 accounts (performing) receive a 12-month ECL; Stage 2 accounts (significant increase in credit risk) receive a lifetime ECL; Stage 3 accounts (credit-impaired) receive a lifetime ECL with individually assessed components for material exposures exceeding R5 million.

The ECL pipeline runs monthly — triggered 2 business days before month-end close — producing provisioning entries for the Oracle Financials GL, regulatory disclosures for the SARB BA210 return, and IFRS 9 note disclosures for the annual financial statements. Total ECL provision as at the most recent calculation: R4.8 billion against a gross loan exposure of R380 billion (1.26% provision coverage ratio).

Fraud Detection — Real-Time Architecture

The fraud model operates in a sub-50ms latency inference pipeline. Card transaction events arrive at Azure Event Hub, are consumed by a Databricks Structured Streaming job, enriched with the customer's Feature Store features (30-day spend velocity, merchant category history, device fingerprint), and scored by the MLflow-served GBM model. Scores above 0.85 trigger an immediate card hold and SWIFT message to the acquiring bank. Scores between 0.65 and 0.85 are flagged for the Fraud Operations team in their real-time dashboard. In 2024, the model prevented R45.2 million in fraudulent transactions with a false positive rate of 0.8% — below the industry benchmark of 1.5%.

Feature Store

The Databricks Feature Store contains 340 pre-computed features organised into 12 feature tables, serving all 12 production models. Features are computed on scheduled Databricks jobs (5-minute cadence for fraud features, hourly for credit features, daily for AML behavioural features) and versioned with point-in-time correctness — preventing training-serving skew by ensuring models trained on historical features see identical feature values during batch scoring of historical periods.

Prime Capital Bank operates under one of the most demanding regulatory environments in Africa. As a Systemically Important Financial Institution (SIFI) designated by the South African Reserve Bank (SARB), the bank is subject to the Prudential Authority's full suite of regulatory requirements. The Data Intelligence Platform automates the generation, validation, and submission of all prudential regulatory returns, reducing the compliance team's data preparation burden by 78% and eliminating the manual errors that previously triggered SARB Directive Letters.

SARB BA700 — Risk Data Aggregation

The BA700 return is the SARB's primary prudential data collection, covering credit risk exposures, capital adequacy, liquidity, and large exposures. The platform generates the BA700 XML submission from a dedicated dbt mart (gold.regulatory.rpt_sarb_ba700), populated from the credit risk Gold tables. The submission is validated against the SARB's published XML schema definition (XSD) before upload to the SARB's BI reporting portal. The platform achieves T+2 submission (two business days after month-end) against the regulatory deadline of T+10.

Basel III Capital Adequacy

The platform computes Liquidity Coverage Ratio (LCR) and Net Stable Funding Ratio (NSFR) on a daily basis using the Gold layer's account balance and treasury position data. Risk-weighted asset (RWA) calculations for credit risk use the Standardised Approach weights stored in dim_risk_category.regulatory_weight, with Internal Ratings-Based (IRB) approach RWA calculated from the PD model outputs for the qualifying retail and corporate portfolios. The bank maintains a Common Equity Tier 1 (CET1) ratio of 14.2% against the SARB's effective minimum of 7.0% (4.5% CET1 floor plus the 2.5% capital conservation buffer), providing a 7.2 percentage-point headroom — consistent with Figure 8.1 below.

POPIA — Data Privacy

The Protection of Personal Information Act (POPIA) compliance is implemented at the platform layer through Unity Catalog tags and column masking policies. All columns containing personal information (SA ID numbers, full names, contact details, biometric data) are tagged with the pii_classification Unity Catalog tag at three levels: PII_HIGH (ID numbers, biometrics), PII_MEDIUM (names, contact details), and PII_LOW (demographic aggregates). Column masking functions applied via Unity Catalog policies ensure that users without the pii_reader privilege see tokenised or masked values in query results — enforced at the Delta Lake level, not application level.

FICA & AML/CFT

The Financial Intelligence Centre Act (FICA) compliance layer includes automated Know Your Customer (KYC) completeness checks, enhanced due diligence (EDD) flagging for high-risk customers, and automated Suspicious Transaction Report (STR) generation. The AML scoring model identifies suspicious transaction patterns — structuring, layering, round-trip transactions, and dormant account activation — and generates STR drafts pre-populated with the FIC's required fields, routed to the Compliance team for review and submission to the Financial Intelligence Centre within the 15-day statutory window.

All data pipeline execution is orchestrated by Databricks Workflows — the native job orchestration engine integrated with the Databricks workspace. The platform runs 47 production pipelines across four scheduling tiers: streaming (continuous), near-real-time (5-minute), hourly, and daily. Databricks Workflows provides dependency-aware DAG scheduling, retry logic, cluster auto-provisioning, and native integration with Azure Monitor for alerting.

Pipeline Hierarchy

SLA Monitoring & Alerting

Pipeline SLA breaches trigger a three-tier alerting cascade: (1) Databricks Workflows email notification to the on-call data engineer; (2) Azure Monitor alert to the PagerDuty on-call rotation for Severity 2 (Gold layer delay) or Severity 1 (streaming fraud pipeline failure); (3) automated Slack message to the #data-platform-ops channel with a direct link to the failed task's log output. The platform maintains a 99.8% pipeline SLA achievement rate over the trailing 12 months, with the 0.2% breaches attributable to upstream source system maintenance windows.

Disaster Recovery

The platform implements an RPO (Recovery Point Objective) of 1 hour and an RTO (Recovery Time Objective) of 4 hours for the Gold layer. ADLS Gen2 geo-redundant storage (GRS) replicates all data to the Azure South Africa West secondary region asynchronously. Databricks cluster configurations and workspace settings are stored in a Terraform repository enabling full workspace re-deployment within 45 minutes. Delta Lake transaction logs enable point-in-time recovery to any hour within the 30-day VACUUM retention window. Monthly DR drills validate the full recovery procedure against the documented RTO/RPO targets.

The consumption tier of the platform delivers insights through eight purpose-built Power BI Premium dashboards, collectively serving 500 active users across the bank. Dashboards are designed against the bank's master data model in the Gold layer, using a mix of DirectQuery (for real-time operational views) and Import mode (for complex analytical views requiring DAX measures). All dashboards are secured through Azure Active Directory row-level security (RLS), ensuring each user sees only the data appropriate to their role and geographic scope.

Dashboard Inventory

1. Executive Dashboard

Board and C-suite facing overview of the bank's key performance indicators: total assets, net interest income, cost-to-income ratio, customer growth, and capital adequacy. Single-page design with drill-through to business unit detail. Updated daily from Gold layer. Audience: CEO, CFO, CRO, board members.

2. Credit Risk Dashboard

Comprehensive view of the loan portfolio: exposure by product and geography, IFRS 9 stage migration waterfall, PD distribution heatmap, ECL movement analysis, and top-20 large exposures. Includes a Monte Carlo simulation-based ECL sensitivity analysis. Audience: Chief Risk Officer, Credit Committee, Provisioning team.

3. Fraud Intelligence Dashboard

Real-time fraud monitoring with a 5-minute data refresh: fraud score distribution, confirmed fraud by channel and merchant category, fraud rate trend, savings vs. prior year, and a geographic heat map of fraud incidents by postal code. Audience: Fraud Operations, Card Operations, Head of Financial Crime.

4. AML & Financial Crime Dashboard

Anti-money laundering case management view: suspicious transaction volumes, STR pipeline, customer risk rating distribution, high-risk jurisdiction exposure, and PEP (Politically Exposed Person) screening status. Audience: MLRO, Compliance team, FICA officers.

5. Branch Performance Dashboard

Branch-level P&L view for regional managers: deposit book growth, loan origination, fee income, customer satisfaction scores, and staff productivity metrics. Geography-filtered by the user's RLS profile. Audience: Regional Managers, Branch Managers, Retail Banking leadership.

6. Treasury & Liquidity Dashboard

Intraday liquidity position, LCR and NSFR ratios vs. limits, FX exposure by currency, bond portfolio mark-to-market, and funding concentration risk metrics. Refreshed every 30 minutes from treasury source data. Audience: Head of Treasury, ALM Committee, SARB Prudential team.

7. Customer 360 Dashboard

Customer-level analytics accessible to relationship managers: product holding summary, transaction recency/frequency/monetary (RFM) scoring, churn propensity score, cross-sell propensity index, and lifetime value estimate. PII-masked for non-privileged users. Audience: Relationship Managers, Digital Banking, Marketing.

8. Regulatory Reporting Dashboard

Compliance-facing view of all regulatory return status: BA700 submission timeline, IFRS 9 provisioning summary, Basel III capital ratios vs. limits, POPIA compliance score, and FIC STR submission log. Provides a single pane of glass for the Chief Compliance Officer. Audience: CCO, Compliance team, Internal Audit.

Data governance is not a layer on top of the platform — it is embedded within it. Unity Catalog provides the centralised governance control plane, while the Medallion architecture's clear layer boundaries enforce the governance model structurally. Every data asset on the platform — tables, views, ML models, dashboards — is catalogued, classified, owned, and access-controlled through a unified framework aligned with the bank's Enterprise Data Management Policy.

Unity Catalog Structure

Access Control Model

Data Lineage

Unity Catalog automatically captures table-level and column-level lineage for all Spark SQL and dbt transformations executed within the Databricks workspace. Lineage is queryable via the Unity Catalog API and visualised in the Databricks UI — showing, for example, that the gold.facts.fact_loan_portfolio.ecl_provision_zar column derives from the PD model feature, which in turn derives from the credit bureau bronze table. This end-to-end lineage satisfies BCBS 239 (Risk Data Aggregation and Risk Reporting) principles and provides the audit trail required for SARB regulatory inspections.

Data Retention & Lifecycle

ADLS Gen2 lifecycle management policies automate tier transitions: data older than 90 days in the Bronze and Silver layers moves from Hot to Cool tier (60% cost saving); data older than 2 years moves to Archive tier (90% cost saving). Gold layer data remains on Hot tier for 5 years, aligning with the bank's Records Retention Schedule and POPIA's storage limitation principle. Delta Lake table history is retained for 30 days for operational recovery, with regulatory snapshots committed to Archive storage for 7-year statutory retention (Companies Act 71 of 2008).

South Africa's payment technology landscape has undergone a structural transformation since 2012. The emergence of app-based QR payments, SME-focused card acquiring, and developer-first online gateways has created a fragmented but dynamic ecosystem sitting alongside the traditional card rails (Visa, Mastercard) and bank-owned channels (EFT, RTC). Prime Capital Bank has positioned itself as the settlement backbone for this ecosystem — providing acquiring bank infrastructure, settlement accounts, and risk oversight for five key fintech partners whose combined annual processing volume exceeds R27.8 billion.

Partner Comparison

Partner Profiles

Yoco — South Africa's SME Payment Champion

Founded in Cape Town in 2013, Yoco has grown to become South Africa's largest independent payment provider with over 250,000 active SME merchants. The product line spans four card machine tiers — Yoco Go (R599), Yoco Khumo (R699), Yoco Neo (R1,499), and Yoco Counter + Neo Touch (R2,999) — serving the full spectrum from market traders to restaurant chains. Following its USD 83 million Series C raise in 2021 (led by Dragoneer, with existing investors Partech and Quona), Yoco has expanded into business banking services including working capital loans, business insurance, and a business management app. The 2.95% MDR is among the most competitive in the POS category and has been the primary driver of merchant adoption versus traditional bank card machines at 3.5–5.0%.

SnapScan — Standard Bank's QR Payment Arm

SnapScan pioneered QR-code-based payments in South Africa after its founding in 2012, and was acquired by Standard Bank in 2014. The platform requires no card machine — merchants display a static QR code and customers pay via the SnapScan app. This zero-hardware model dramatically lowered the barrier to entry for informal and pop-up merchants: food trucks, market vendors, street food traders, and event operators. The 2.75% MDR and R340 average basket reflect the hospitality and food-and-beverage concentration of its merchant base. As a Standard Bank subsidiary, SnapScan is a direct competitor of PCB in the SME merchant acquiring space, making PCB's settlement bank relationship strategically important for margin defence.

PayFast — South Africa's E-Commerce Gateway

Founded in Cape Town in 2007, PayFast is the dominant online payment gateway for South African e-commerce with integrations across Shopify, WooCommerce, Magento, Wix, and PrestaShop. Its 20+ supported payment methods — including credit/debit cards, instant EFT, Mobicred, Zapper, and SiD Secure EFT — made it the default choice for SA merchants going online. Acquired by DPO Group in 2021 (which was itself subsequently acquired by Network International in 2022), PayFast now forms part of Africa's largest payment processing group. The 3.50% + R2.00 fee structure is the highest in the comparison set, justified by the complexity of multi-method checkout, fraud screening, and the broad integration ecosystem. The R1,251 average basket reflects the higher-value e-commerce purchase context.

Peach Payments — The Developer-First Gateway

Peach Payments was founded in Cape Town in 2012 with a deliberate focus on API-first architecture and developer experience. Its REST API, hosted payment page, and tokenisation library are widely regarded as the most technically sophisticated in the SA market. Merchants include Superbalist, OneDayOnly, and major hospitality chains. The 2.90% + R1.50 fee structure is the most competitive among online gateways, reflecting a deliberate strategy to win mid-to-large merchants on economics. Peach Payments supports 3D Secure 2.0, recurring billing, and subscription management — capabilities driving rapid growth in the B2B SaaS and subscription commerce verticals. Settlement on T+1 versus competitors' T+2 provides a working capital advantage to merchants.

PayGate — The Enterprise Incumbent

Established in Johannesburg in 2000, PayGate is one of the oldest SA online payment gateways and serves enterprise-tier merchants including large national retailers, travel agencies, and government e-services platforms. Also a DPO Group subsidiary (Network International), PayGate occupies the high-basket end of the market with a R1,891 average transaction value — reflecting the enterprise procurement, B2B invoice, and travel booking use cases. While its technology stack predates the API-first era, PayGate's strength lies in its deep enterprise integrations, recurring billing infrastructure, and 25+ years of institutional trust. PCB's settlement relationship with PayGate covers the largest average transaction values in the fintech portfolio.

Annual Settlement Volume by Partner

2024 estimated annual gross settlement volumes (R billions). Source: PCB Acquiring System — FACT_FINTECH_SETTLEMENT aggregated.

Monthly Settlement Trend — 2024

Combined monthly gross volume across all five partners (R millions). Black Friday (November) and Festive Season (December) peaks clearly visible.

SA Payment Ecosystem — Market Context

The total South African card and digital payment market processes approximately R1.8 trillion per year across all channels. The five fintech partners integrated with PCB represent roughly 1.5% of this market by value but a disproportionately high share of SME merchant transaction count — the fastest-growing segment by number of merchants. Key macro drivers:

• Card-not-present (CNP) fraud driving migration from EFT to card-on-file and 3DS2
• SARB's Rapid Payments Programme (RPP) / PayShap launch (2023) increasing real-time competition
• Tax incentives for POS adoption under the SARS VAT compliance drive
• Informal economy formalisation — ±6 million previously cash-only merchants addressable
• Smartphone penetration at 84% (GSMA 2024) enabling QR and tap-to-pay growth

Data Model: Cards & Fintech Domain

Two new Gold layer tables underpin the fintech analytics capability:

Chargeback Ratio by Partner

Average annual chargeback ratio (%) by partner. POS/QR channels show lower rates than online gateways due to card-present authentication and biometric verification. Industry benchmark: <0.20%.

Integration Architecture

The fintech settlement data flows into PCB's data platform through a dedicated acquiring integration pipeline:

Glossary of Terms

Full Table Catalog

A complete table-level reference is maintained in the separate Data Dictionary document (docs/data_dictionary/data_dictionary.md), which provides column-level documentation, business rules, and PII classification for all 247 catalogued tables across the Bronze, Silver, and Gold layers.

Contact Information

— End of Document —

Prime Capital Bank Limited  ·  Registration No. 1968/000123/06  ·  Authorised Financial Services Provider (FSP 12345)  ·  Registered Credit Provider (NCRCP 9876)
Prudential Authority Licence No. PA12345  ·  © 2025 Prime Capital Bank Limited. All rights reserved.