Compliance requires dual controls. AI can generate drafts at scale, but human approval remains essential. Here's how maker-checker works when AI is the maker and human is the checker — and why the audit trail is more complete than manual processes.
In financial services, maker-checker is the baseline control. One person creates the entry, another person reviews and approves it. This prevents errors, fraud, and unilateral decisions. It's not going away — and it shouldn't.
The question isn't whether maker-checker still matters. The question is how it works when AI generates the drafts. If AI is the maker and human is the checker, does the control hold? Or does automation create new failure modes that the traditional workflow doesn't address?
Maker-checker workflows were deployed for a SET-listed company and a Singapore family office. The pattern is consistent: AI accelerates the maker step, the checker step stays human, and the audit trail is more complete than manual processes. Here's the breakdown.
The test cases span two deployments: a SET-listed company in Thailand processing quarterly reports and a Singapore family office managing multi-asset portfolios. Both required maker-checker workflows for compliance. The baseline was measured by comparing cycle times and error rates before and after introducing AI-generated drafts into the maker step.
Key metrics tracked: field extraction accuracy on standard financial documents, cycle time reduction from maker submission to checker approval, and audit trail completeness (percentage of actions logged with timestamps).
The problem with traditional maker-checker workflows is speed. The maker submits a draft, the checker reviews it later, often on a different system or via email. The cycle creates delays, especially when the checker is unavailable. For time-sensitive operations like quarterly reporting or fund data entry, these delays are costly.
The control itself is sound — dual approval prevents errors and fraud. The bottleneck is the workflow, not the principle. When AI generates the drafts, the maker step is instant. The checker step stays human. The cycle time drops from days to hours.
This isn't a theoretical improvement. For a finance team processing 20-30 documents per month, the difference between "days of back-and-forth" and "hours of review" is the difference between a compliance bottleneck and a compliance feature.
AI doesn't replace maker-checker — it accelerates it. The workflow has three parts:
AI as the maker. The system processes documents, extracts data, and generates draft entries. This happens instantly, 24/7. No waiting for someone to sit down and type figures. Document processing pipelines achieve 90%+ field accuracy on standard financial documents — so the checker's job is focused review, not re-keying.
Human as the checker. The finance team reviews the AI-generated drafts. Because the AI achieves 90%+ field accuracy, the review is focused on exceptions and edge cases, not re-typing everything. What took 20 minutes of data entry now takes 2 minutes of review.
Full audit trail. Every action is logged: what the AI extracted, what the human changed, when approval happened. The audit trail is more complete than manual processes, not less. Traditional maker-checker relies on paper trails or email chains. AI-generated drafts log every field, every change, and every approval timestamp.
Most systems add maker-checker as an afterthought — a separate approval step that sits outside the main workflow. The difference between a system that looks good in a demo and one that works in production is whether the control is embedded in the process, not documented separately.
For MAS compliance, this means the control is embedded in the process, not documented separately. For operations, it means approvals happen in minutes, not days. The system enforces the workflow — drafts cannot be published without checker approval.
This is the pattern for compliance in automation: the control is a feature of the system, not a constraint on it. Speed and compliance are not trade-offs — they're designed together.
The maker-checker workflow assumes the checker has the capacity to review AI-generated drafts. If the finance team is already at capacity, shifting from data entry to review doesn't free up time — it shifts the bottleneck. The time savings are incremental, not absolute.
The 90%+ accuracy figure applies to standard financial documents with clear tables and structured layouts. Documents with complex nested tables, multi-column formats, or mixed languages will require more prompt engineering and potentially lower accuracy. The human review step is not optional — it's the safety net that makes the system reliable.
Finally, the system assumes the checker has the judgment to catch errors. If the checker is rubber-stamping approvals without actually reviewing, the control is broken regardless of whether the maker is human or AI. The workflow enforces approval, but it can't enforce attention.
The broader implication is that maker-checker workflows are more viable with AI than without it. The AI handles the repetitive work — data extraction, draft generation, compliance checking. The human handles the judgment calls — exception handling, strategic context, and final approval.
For Singapore businesses, the compliance question is straightforward. MAS guidelines require dual controls for financial data entry. AI-generated drafts with human approval satisfy the requirement while reducing cycle time from days to hours. The audit trail is more complete than manual processes, which means compliance is easier to demonstrate, not harder.
The pattern applies across document processing, quarterly reporting, and financial data management: automate the repetition, preserve the judgment. The AI handles the work that used to define your capacity. Your people handle the thinking that actually matters.