Generative Narrative Reporting

Problem class

Writing financial narrative is the most time-intensive human task in finance. Analysts spend days each cycle drafting MD&A and variance commentary that follow predictable structures. Generative AI drafts the first 70-95% in minutes.

Mechanism

Structured financial data feeds an LLM via retrieval-augmented prompts. The model drafts narrative following a templated structure (variance type → cause → magnitude → outlook). Analysts review, edit, and add the strategic context the model can't infer. Iterative feedback improves draft quality over cycles.

Required inputs

• Structured financial data (variances, KPIs, segment results)
• Prior-period narrative corpus for tone and structure
• Disclosure templates and regulatory style guides
• Strategic context inputs from leadership

Produced outputs

• Draft variance commentary at 70-95% completeness
• MD&A and earnings call narrative drafts
• Board reporting summaries
• Multilingual narrative variants for global reporting

Industries where this is standard

• Public companies with quarterly disclosure cycles
• Large investment banks producing analyst reports
• Multinational groups with multi-language reporting
• Asset managers producing fund commentary
• Insurance carriers with regulatory narrative reporting

Counterexamples

• Highly judgment-driven reports (M&A fairness opinions, going-concern disclosures, fair value commentary) where the model's confident drafting creates false-precision risk.
• Privacy-sensitive contexts without on-premise or private LLM hosting — sending material non-public information to public model APIs is a disclosure incident waiting to happen.

Representative implementations

• Goldman Sachs — internal AI tools draft 95% of an S-1 IPO prospectus in minutes (work that previously required a 6-person team for 2 weeks); CEO David Solomon: "the last 5% now matters because the rest is a commodity."
• Workiva (Forrester TEI, Sept 2025) — 3,565 hours/year saved in reporting tasks, $868K annual savings, 208% ROI over 3 years; Bancolombia cut Form 20-F data collection 79 → 34 days (57% reduction).
• Large global banking group — Yseop NLG for credit analysis narratives; 16,000 reports/year across 200+ analysts, ~30 minutes saved per report (~8,000 hours annually).

Common tooling categories

LLM (private or API) + RAG over financial data + templated prompt library + human review workflow + version-control draft management.