Case study
A one-week build raised automated approvals from 44% to 90% in testing
A transaction review could take an operator 30 minutes or more. In one week, Grail built a rules-based review system that cleared straightforward cases automatically and sent uncertain cases to the compliance team.
The problem
Reviewers had to collect the same evidence and apply the same policy rules to every transaction by hand. Even a straightforward approval could take 30 minutes, and the team had no safe way to see how a rule change would affect past decisions before using it.
The system we built
Grail turned the written policy into a rules engine and built a review screen around it. The team can run new rules against the same set of past transactions, compare every changed decision and approve a policy separately before it affects live work.
Bun
TypeScript
React
Microsoft SQL Server
AzureThe result
On the same set of past transactions, automated approvals rose from about 44% to about 90% while the compliance rules stayed in place. Straightforward cases could be cleared in seconds, while uncertain cases still went to the compliance team.
Results at a glance
| Primary result | First version built in one week |
|---|---|
| Operational result | Automated approvals rose from about 44% to about 90% in testing |
| System scope | Every changed decision available for operator review |
| Delivery note | Compliance approval required before any rule reaches live transactions |
How we built it
We wrote each policy rule in code, replayed the same past transactions through the current and proposed versions, compared every changed result and kept the new rules out of live work until compliance approved them.
- Prepare the same past transactions and supporting evidence for every test.
- Run the current rules and the proposed rules over the same records.
- Show every decision that changed, the rule behind it and the evidence used.
- The compliance team reviews the changed cases.
- Keep the proposed rules away from live transactions until the compliance team approves them.
How people stay in control
The software handles repeatable checks and routine work. People still approve decisions that affect customers, money, privacy or compliance.
- A test cannot change the live rules or a live transaction.
- Missing or unreliable evidence stops automation instead of becoming an approval.
- The compliance team approves rule changes and reviews every uncertain case.
What we delivered
The finished scope included:
- A rules engine that follows the written compliance policy.
- A review screen for testing and comparing rule changes.
- Repeatable comparisons over the same past transactions.
- The evidence and reason behind every result.
- Separate test and live versions of every policy.