Generic AI reports were not enough.
The reports needed to be meaningfully different. A numerical balance review, a client-facing statement overview, and a financial KPI analysis should not sound like the same answer with different titles.
Saldo needed AI-assisted reports that could distinguish balance checks, plain-language financial statement overviews, and KPI-focused financial analysis.
Reports and supporting files enter the workflow.
RTJ guidance and supplied notes shape analysis.
Balance checks, overview, and KPI analysis are separated.
Unsupported claims and weak recommendations are caught.
Financial judgement stays human-controlled.
The work focused on turning uploaded financial documents and accounting guidance into separate report types, with clearer prompts, RTJ rule context, better source handling, and outputs that accountants could review before use.
The reports needed to be meaningfully different. A numerical balance review, a client-facing statement overview, and a financial KPI analysis should not sound like the same answer with different titles.
The workflow used direct prompting with a focused knowledge base instead of overcomplicating the system, separated the prompt design by report type, and used client examples to test weak recommendations and unsupported claims.
AI recommendations, anomaly explanations, equity comments, generated reports, and rule-sensitive statements require professional review before client use, especially where accounting rules and judgement are involved.
The result is a clearer direction for finance AI: differentiated outputs, better review expectations, and a workflow that supports accountants instead of replacing judgement.
Design the report types, evaluation checks, and human review path before turning AI loose on sensitive documents.
Scope an AI workflow