Agentic SEC EDGAR

The question. When a technology company spends a larger share of revenue on R&D, does it earn better financial results? Leaders ask this when they set budgets or benchmark against peers.

The approach. I pulled 17 years of audited annual filings (10-K) from the SEC’s free XBRL API for 15 companies: mega-caps, enterprise software, and a contrast group of legacy hardware and infrastructure firms. I built a company-by-year panel of R&D intensity (R&D ÷ revenue), CapEx intensity, and net margin. Then I tested the relationship across companies, within each company over time, and with forward lags. 

Research pipeline in Python, plus a weekly monitor in n8n. The comparison and alert step is still in progress.

What the data allowed. Public data has gaps, and they shape what I can claim. Amazon and Xerox report no standalone R&D line, so they are excluded from R&D comparisons. Palantir and Snowflake have only six years of filings, so I treat their results as indicative. 

Fiscal years of annual data retrieved per company and metric. Hatched cells mean the company doesn’t report that item as a standalone figure. 

Key findings

• Pooling companies is misleading. Across all companies, higher R&D intensity appears to go with lower net margin. That comes from comparing different business models, not from R&D hurting results.

• Within a company, the link is close to zero. Companies that spend more on R&D in a given year don’t show better or worse margins afterward.

• Intensity is mostly a company trait. It varies widely, from under 3% of revenue at HP to over 40% at Snowflake, and it holds fairly steady year to year.

What this means for decision-makers

• Don’t benchmark R&D spend against a peer average without checking that the peers are comparable.

• R&D intensity alone doesn’t predict margin. Any payoff is more likely to show up in revenue growth or over longer horizons.

• Treat results from companies with short filing histories as hypotheses to monitor, not findings.

Limitations. Fifteen non-randomly chosen companies is a small sample. Reported R&D differs by accounting treatment. Net margin is a noisy outcome, and revenue growth and returns on capital are natural next tests.

Next. I’m extending this into a self-hosted n8n workflow that refreshes the data weekly and flags meaningful changes.