Real-Time Credit Feature Recovery Authority — Digital Lending

The mandate

The decision-data executive was suspended after a source replay showed that some affordability features had served values older than their published freshness limits during a high-volume campaign. Portfolio impact remains under review, but online decisions cannot be recreated reliably from offline training views and correction events were applied differently across channels. The risk committee has capped originations until one leader controls feature evidence end to end.

The interim must start within two weeks for nine months, occupying the executive seat throughout forensic review, remediation and two model-release cycles. A permanent search opens after the affected population is bounded in month two. Four weeks of joint decision-making are reserved for the successor, and the fixed term cannot extend merely because feature migration or recruitment falls behind.

Handover requires every material credit feature to carry event time, source freshness, transformation version, point-in-time join and online-serving evidence; three monthly reconciliations must find no unexplained decision variance; and two unseen replay tests must reproduce approval inputs. The successor will sign a live feature release, accept residual source limitations and own a tested rollback and customer-correction playbook.

The interim may disable features, force a fallback, suspend releases, direct the sanctioned ฿620 million remediation and appoint temporary data-control leaders. Changes to credit policy, customer redress, model-risk acceptance, permanent appointments or spend beyond the envelope require CRO or committee approval. The seat may prevent a decision input from being used but cannot approve lending policy or individual applications.

Rebuilding the origination journey, renegotiating bureau contracts and developing new credit models are outside this assignment. The authority can require reliable interfaces and data-provider evidence but does not own marketing eligibility, collections strategy or historic cleansing unrelated to decisions within the review population. Remediation must remain centred on feature integrity and reproducible use.

Why this seat is open

The stale-feature finding created an accountability gap precisely where model, streaming and credit-data teams overlap. Suspension of the incumbent prevents credible self-review, while the origination cap demands decisions faster than a permanent search. A temporary executive must establish point-in-time truth, support fair customer treatment and leave a qualified platform to permanent leadership.

What you will own

• Bound the affected decision population by tracing feature source, event time, materialisation, serving, fallback, model version and channel outcome.
• Establish critical-feature contracts for freshness, completeness, point-in-time correctness, online-offline consistency, correction and permissible degradation.
• Decide which features may return, require conservative fallback, remain in shadow mode or must be removed from production decisions.
• Direct replays and counterfactual comparisons that separate stale inputs, model behaviour, policy overlays and downstream manual intervention.
• Implement release evidence linking training definitions, production transformations, serving latency, monitoring, owner approval and rapid rollback.
• Reconcile decision and customer impact with Risk, including populations requiring review without pre-empting accountable redress judgement.
• Transfer authority through an unseen replay, live release council and signed successor acceptance of feature, source and operational debt.

Candidate qualifications

• Held executive data or ML-platform authority for high-volume regulated credit, payment, insurance or fraud decisions made in real time.
• Reconstructed point-in-time features after a decision incident and distinguished event-time errors from model, policy and channel effects.
• Governed feature stores or equivalent serving platforms with tested online-offline consistency, fallback and correction behaviour.
• Worked directly with model risk, compliance and customer-remediation teams without claiming authority over their distinct judgements.
• Stopped or constrained originations because decision-data evidence was inadequate and managed the resulting commercial pressure.
• Handed a recovered decision platform to permanent leadership through observed qualification rather than repository and document transfer alone.

Non-negotiables

• Available in Bangkok within two weeks for exclusive service and prepared to lead incident decisions outside normal hours.
• Has personally governed features used in consequential regulated decisions; batch analytics or generic lakehouse leadership is insufficient.
• No undisclosed financial relationship with bureau, alternative-data, feature-platform, cloud or model-tooling suppliers in the perimeter.
• Will preserve point-in-time evidence and support transparent population review even if the eventual customer exposure exceeds initial estimates.

1. 49 words maximum. State your Bangkok availability and one stale or point-in-time feature incident for which you held authority.
2. 49 words maximum. How did you separate online-serving error from model or policy effects when reconstructing affected decisions?
3. 49 words maximum. Which unseen replay would convince you that a repaired credit feature can return to production?