Context
A pre-call account summary for reviewers at a business in regulated finance, where the facts about an account are spread across several kinds of record.
The problem
A reviewer has about 30 seconds to size up an account before a call, with the facts spread across agreements, invoices, payment history, and filings. For the summary to be useful, the reviewer has to be able to trust every number on it, so each amount and date needs to be backed by the records.
Constraints
- Designed to be read in about 30 seconds, before the call starts.
- Every amount and date had to be verifiable against the system of record.
- Where data was missing, the summary had to say so rather than fill the gap.
Approach
The assistant isn’t handed facts to repeat. It reads data only through tools that look up the system of record, so each figure it uses comes from a record a reviewer can check. The design rule is strict: an amount or date is stated only if it appears exactly in that data, and “no data” is preferred to a guess. It also looks for contradictions, such as an account marked current that still carries a past-due balance, and flags them instead of smoothing them over.
Key engineering decisions
Facts come from record lookups
The AI reads the system of record through lookup tools instead of being given data in its instructions, so each figure has an origin a reviewer can trace.
Exact values, or “no data”
An amount or date that does not appear exactly in the source is not stated, and a gap is reported as a gap.
Surface contradictions
Records that disagree, like a current account with a past-due balance, are flagged for the reviewer to check first rather than silently reconciled.
Result
A quick, source-backed summary a reviewer can check at a glance, with a clear “check this first” note wherever the data disagrees with itself.