1% to 58% auto-decisioning. Five people. NZ$1M. Ten months.
- Rebecca Speirs
- 3 days ago
- 2 min read
Updated: 2 days ago
A New Zealand bank was auto-decisioning less than 1% of business lending applications.
That meant almost every application needed a human to review it. Contact centre staff. Credit analysts. Manual touchpoints at every step.
The target for phase one was 60% up to $250,000.
I was brought in to lead scoping, requirements, and business rules definition for the rebuild. The budget was NZ$1M. The team was five people.
The challenge nobody had fully cracked with a previous attempt a few years earlier was understanding the entity structures.
Business lending isn't straightforward. A loan might involve a company, a trust, directors, guarantors, related entities — and the system needed to know exactly whose details it needed, to what degree of detail, to make a compliant decision. Get that wrong and you either miss mandatory data or frustrate customers and waste time collecting information you don't need.
With the stakeholders I agreed a basket of lending scenarios and entities involved, around a dozen in total, from simple to complex and mapped them by hand — stick figures, assets, lines showing relationships and concentric rings to represent degrees of detail required. It showed who needed to be in the 'decision group', the 'application group' and the 'associated entities'. I workshopped them with business stakeholders until we had agreement on every scenario.
That gave the front-end team the logic to capture the right data. It gave the data modelers what they needed to build the decision rules.
The outcome: auto-decisioning hit 58% — effectively the 60% target — at launch.
In the first week alone, $1M NZD of lending was auto-decisioned. A contact centre person could take a call, work through the application, and have a decision for the customer in under 30 minutes. In Auckland. In 2015. That was genuinely fast.
Estimated annualised operational saving: $1.2M NZD, based on removing approximately $165 NZD of manual cost per application across roughly 1,000 applications per month — freeing senior credit analysts to focus on complex, high-value lending instead.
The programme manager noted the bank had 'got a lot of bang for buck':
"Rebecca did a stirling job".
Vaile Mexted, Programme Manager, Retail Models
#Banking #FinancialServices #LendingTransformation #StrategicExecution #DecisionDesign #Transformation
