A local bank asked Marketways to examine payment completion without weakening fraud and financial-crime controls. Discovery showed that customer friction, alert quality, support capacity and model reliance were interacting across separate teams. We structured the engagement around the complete payment journey and the decisions made when a transaction or alert did not follow the expected path.
The engagement objective
Through the initial discovery, Marketways defined the objective: connect payment completion, fraud, operations and customer impact.
How Marketways translated the problem
We began with a practical question: Where do legitimate payments fail and suspicious payments pass? The first analysis used payment events, authentication, declines, disputes, devices and merchant context.
That evidence could not be read in isolation. Stronger friction can reduce fraud while also excluding legitimate customers or shifting fraud elsewhere. Rules remain necessary, but isolated thresholds can miss patterns and overwhelm investigators.
A more aggressive alert threshold could appear safer while creating false positives and unnecessary customer friction. We treated detection, investigation capacity and the cost of each kind of error as one decision.
We did not judge each component by its isolated KPI. We examined how people, assets, decisions and constraints affected one another, then used the evidence to test whether an apparent improvement would strengthen the complete system or merely move cost, pressure or risk elsewhere.
How the engagement developed
The initial work on payment failure and fraud journey exposed dependencies with adaptive aml risk intelligence, branch, contact-centre and digital capacity, model inventory and reliance review. Treating them as separate recommendations would have left the operating trade-offs unresolved.
- Payment failure and fraud journey: Improve customer completion and control effectiveness together.
- Adaptive AML risk intelligence: Reduce low-value alerts while surfacing contextual risk.
- Branch, contact-centre and digital capacity: Reduce avoidable waiting and preserve support for complex needs.
- Model inventory and reliance review: Prioritise validation and controls according to real business consequence.
Evidence we examined
- Payment events, authentication, declines, disputes, devices and merchant context.
- Transactions, customers, entities, geography, alerts and investigation outcomes.
- Visits, calls, digital journeys, handling time, abandonment and staffing.
- Model purpose, inputs, validation, overrides, monitoring and incidents.
Industry conditions we accounted for
- Stronger friction can reduce fraud while also excluding legitimate customers or shifting fraud elsewhere.
- Rules remain necessary, but isolated thresholds can miss patterns and overwhelm investigators.
- Digital adoption does not remove demand; it changes contact reasons and escalation patterns.
- A technically accurate model may still fail through poor data, workflow use or changing conditions.
How our engagement contributed to business impact
We connected every method to a decision and a business measure. The organisation could assess the engagement through operating results as well as model performance.
- Payment failure and fraud journey
- Method: Process, Workflow & Systems, Machine Learning & Predictive Analytics.
- Evidence: Payment events, authentication, declines, disputes, devices and merchant context.
- Decision supported: Improve customer completion and control effectiveness together.
- Impact measure: Fraud loss, false decline, completion and review time.
- Adaptive AML risk intelligence
- Method: Machine Learning & Predictive Analytics, Forecasting, Risk & Optimisation.
- Evidence: Transactions, customers, entities, geography, alerts and investigation outcomes.
- Decision supported: Reduce low-value alerts while surfacing contextual risk.
- Impact measure: Useful alert yield, missed risk, investigation time and explainability.
- Branch, contact-centre and digital capacity
- Method: Forecasting, Risk & Optimisation, Process, Workflow & Systems.
- Evidence: Visits, calls, digital journeys, handling time, abandonment and staffing.
- Decision supported: Reduce avoidable waiting and preserve support for complex needs.
- Impact measure: Completion, waiting, first-contact resolution and cost to serve.
- Model inventory and reliance review
- Method: Data Foundations & Business Intelligence, Machine Learning & Predictive Analytics.
- Evidence: Model purpose, inputs, validation, overrides, monitoring and incidents.
- Decision supported: Prioritise validation and controls according to real business consequence.
- Impact measure: Coverage, unresolved findings, drift response and controlled use.
Implementation
The engagement was structured as multi-quarter transformation. We connected the analysis to the decisions, operating constraints and measures that the organisation would continue to use.
How success was assessed
The overall assessment considered safer growth with fewer avoidable failures and clearer control. The supporting measures included:
- Fraud loss, false decline, completion and review time.
- Useful alert yield, missed risk, investigation time and explainability.
- Completion, waiting, first-contact resolution and cost to serve.
- Coverage, unresolved findings, drift response and controlled use.
Services and methods used
Services: Process & Workflow Analysis & Redesign, Risk Detection, AI & Model Risk, Workforce Planning & Capacity, Business Systems Design & Architecture.
Methods: Process, Workflow & Systems, Machine Learning & Predictive Analytics, Forecasting, Risk & Optimisation, Data Foundations & Business Intelligence.