The question
Enterprise software firms have poured resources into AI, and executives describe it as a driver of higher margins and shareholder value. This study asks a sharper version: has that value actually shown up — and can the way a company talks about AI in its disclosures predict its stock? Workday, Inc. (NASDAQ: WDAY) serves as a longitudinal single-firm case study.
What I built
I assembled a corpus of Workday-authored documents — SEC filings, investor-relations pages, earnings remarks, and corporate blogs from 2015 to 2026 — via a focused, robots-compliant web crawler, then deduplicated and stored them in a PostgreSQL database. From there I constructed a knowledge graph governed by a versioned "canonical analysis layer" that keeps one primary narrative document per filing event, ensuring temporal comparability across years.
The graph produces an AI Intensity Index (AII) — a quarterly measure of AI-related language density — which I fed into a logistic-regression model to test whether it adds predictive value for next-quarter returns over a price-only baseline.
What I found
- The AII captures four distinct, statistically detectable eras of AI language, including an abrupt structural break in 2019-Q4 (confirmed by a Chow test) and a shift to generative-AI vocabulary in 2023-Q2 — two quarters after ChatGPT's launch, consistent with the SEC filing cycle.
- The predictive result was inconclusive, and I report it honestly: price-only features dominated, and adding AII did not improve accuracy. The headline 0.81 test accuracy of the baseline is fragile (validation AUC of 0, n = 16 test quarters).
- Granger causality offered the more interesting finding: evidence that returns lead AII, not the reverse — consistent with a lagging disclosure narrative where management describes AI more intensely after favorable quarters. This connects to the SEC's 2024 "AI washing" enforcement actions.
Why it matters
The durable contribution isn't the trading signal — it's the knowledge graph itself, designed to answer questions like "what AI capabilities does this firm claim, where, and how often?" and "which risk topics are emerging?" The schema generalizes beyond a single company to any B2B SaaS firm, making it a reusable instrument for analysts, investors, and researchers studying enterprise AI strategy over time.
Full methodology, the KG schema, and reproduction code are in the project repository.