Automated Workflows · n8n + Claude

A filing goes in.
A brief comes out.

One webhook call turns a raw 10-Q or earnings-call transcript into a structured FP&A brief — YoY growth, margins, guidance direction, ranked risks, and a scored sentiment read — in under a minute, with no analyst in the loop.

TriggerPOST /webhook
ModelClaude Sonnet 5
Output schemaJSON + Markdown

How it's wired

Deterministic code handles parsing and formatting. A single schema-constrained model call handles the judgment: sentiment, materiality, guidance direction.

01
Webhook Intake
Trigger
02
Normalize Input
Code
03
Is PDF?
Router
04a
Extract Text from PDF
Binary → Text
04b
Map Transcript
JSON → Text
05
Merge Branches
Merge
06
Split MD&A / Risk / Guidance
Regex
07
Regex Metric Pre-Pass
Grounding hints
08
Extract Metrics & Sentiment
Claude Agent
09
Build FP&A Brief
Code
10
Sentiment Negative?
Router
11
Respond to Webhook
Response

Dual intake at node 03 means the same endpoint accepts a bound 10‑Q/10‑K PDF or a transcript JSON body — both converge before section-splitting, so nothing downstream needs to know which one arrived.

Grounding, not guessing

Regex pre-pass before the model

Sentence-level regex hits on revenue, margin, and guidance language ride alongside the raw text into the prompt — a cheap way to anchor the model on the numbers actually present before it has to reason about them.

Nulls are a valid answer

Schema allows "not disclosed"

The structured-output schema treats every hard number as nullable. A company that doesn't report a metric should yield null, not a hallucinated figure — the parser enforces that at the type level, not just the prompt.

Escalation, not just archiving

Sentiment routes the response

A sentiment score below -0.2 branches to a Slack alert before the brief is returned — the workflow doesn't just file the analysis, it flags deteriorating filings for the FP&A team automatically.

Live output — NVDA, Q2 FY2027

Real figures from NVIDIA's Aug 26, 2026 earnings release, run through the workflow above. Generated by Claude, not hand-written.

NVIDIA Corporation
NVDA · Earnings Call Transcript · Q2 FY2027
GENERATED 2026-08-27T01:40:01Z
EXECUTION #33 · webhook/earnings-summarizer
YoY Revenue Growth
0%
Gross Margin
0%
Guidance Change
Initiated ↑

Revenue

Total revenue of $96.2B grew 106% year over year, well above the high end of prior guidance. Data Center revenue was $89.0B, the dominant growth engine, fueled by Blackwell-generation system ramps and large forward orders from hyperscale and sovereign AI customers.

Guidance

Q3 FY2027 revenue guidance initiated at $108.0B (±2%), implying continued strong sequential growth from Blackwell Ultra and early Rubin platform revenue. Gross margin guided ~74.0% (±50bps) — a 100bps step-down from Q2. Outlook explicitly excludes any China Data Center compute revenue.

Management sentiment — score 0.92

Highly confident and bullish. Management frames supply constraints as a demand-management challenge rather than a demand problem, and guided well above consensus with an "initiated" Q3 outlook implying further sequential acceleration.

Top risk factors

  • Advanced packaging / HBM supply constraints could limit near-term demand fulfillment.
  • China export controls — Data Center compute revenue entirely excluded from guidance.
  • Customer concentration: a handful of hyperscalers drive most Data Center revenue.
  • Custom silicon (ASICs) competing with merchant GPUs at hyperscale accounts.
  • Component and power availability as global buildouts scale.

Red flags

  • China revenue excluded from guidance with no quantified impact — opaque downside.
  • HBM/packaging supply could cap Q3 upside even with robust demand.
  • Q3 gross margin guide is a 100bps sequential step-down from Q2's 75.0%.

Key takeaways

  • Revenue surged 106% YoY to $96.2B; Data Center alone contributed $89.0B.
  • Gross margin held at 75.0% despite a full architecture transition to Blackwell.
  • Q3 guidance of $108.0B implies continued strong sequential growth ahead.
  • Management treats supply constraints as demand-management, not weak demand.
schema: manual JSON Schema · nullable numeric fields · parser: outputParserStructured v1.3 · model: claude-sonnet-4-6