Analyzing Earnings Calls and 10-Ks with AI
Earnings season is the most time-pressured part of an analyst's calendar. Hundreds of companies report inside a four-week window, each producing a transcript, a press release, and a 10-Q or 10-K. The analysts who win are not the ones who read fastest. They are the ones who triage best — figuring out which two pages of which document actually move the model. AI changes that math.
This lesson shows you how to use AI to extract the signal from filings and transcripts in minutes instead of hours.
What You'll Learn
- How to summarize an earnings call in under two minutes
- How to compare management's tone across multiple quarters
- How to extract guidance and risk factor changes from a 10-K
- A reusable workflow for post-print notes
The Two-Minute Earnings Call Workflow
After the bell, transcripts hit Capital IQ, Bloomberg, or FactSet within 30-60 minutes. Here is the workflow:
- Copy the full transcript into a new Claude conversation (Claude's long context handles the entire call easily — usually 15K-25K words).
- Run this prompt:
You are reviewing this Q[N] [YEAR] earnings call for [TICKER].
I cover this name on the buy-side. Produce four sections:
QUANTITATIVE BEATS AND MISSES
- Reported revenue, EPS, and any segment numbers vs consensus.
Quote the exact figure stated by management.
GUIDANCE CHANGES
- Quote management's exact guidance language for the next quarter
and full year. Note any change in tone vs prior guidance
("strong" vs "in line with expectations" etc.).
ANALYST QUESTIONS THAT MATTERED
- Three questions where management gave a non-answer or hedged.
Include the analyst name and the verbatim question.
RED FLAGS
- Any new language about cost pressure, customer concentration,
inventory, or supply chain. Quote the exact words.
Do not invent numbers. Quote them or skip.
- Verify the quoted numbers against the press release. Two minutes, max.
The result is a structured post-print summary that you would have spent 45 minutes producing by hand.
Comparing Management Tone Across Quarters
A single transcript tells you what management said today. The interesting question is how their tone has shifted. Upload the last four transcripts into a Claude Project or NotebookLM:
Across the four transcripts I uploaded, build a table tracking
management's language on these four topics:
1. Gross margin trajectory
2. Customer demand or pipeline
3. Capital allocation (buybacks, M&A, capex)
4. Competitive environment
For each topic, give me one short quote per quarter (Q1, Q2, Q3, Q4)
and a one-word tone tag (bullish / neutral / cautious / hedged).
Output as a 4x5 markdown table.
This is one of the highest-value uses of AI for analysts. Detecting subtle language shifts — "strong" becoming "stable" becoming "softening" — is the kind of pattern recognition that took senior analysts years to develop. AI surfaces it on demand.
Extracting Guidance From the 10-K
10-Ks are long, repetitive, and 80% legal boilerplate. The 20% that matters is concentrated in MD&A, risk factors, and segment disclosures. A workflow:
Step 1: Upload the 10-K PDF to Claude or ChatGPT.
Step 2: Run this prompt:
This is the 10-K for [COMPANY] fiscal year [YEAR]. Extract:
1. SEGMENT RESULTS
For each operating segment, give me:
- Revenue (current year, prior year, YoY%)
- Operating income or loss (current, prior, YoY%)
- Any qualitative drivers management cites
2. NEW RISK FACTORS
Compared to a generic 10-K, flag the top three risk factors
that look specific to this company (not generic boilerplate
about cybersecurity, key personnel, etc.).
3. CRITICAL ACCOUNTING ESTIMATES
Summarize each in one sentence and flag where small changes
in estimate would materially move reported results.
4. CAPITAL COMMITMENTS
Total contractual obligations table and any large purchase
commitments or operating leases noted.
Quote exact numbers. If something is unclear, say so.
The output is essentially the analyst-relevant 10-K in two pages.
Year-Over-Year 10-K Comparison
The most underused workflow: diff this year's 10-K against last year's. Material changes signal what management cares about.
I am uploading both the FY[YEAR] and FY[YEAR-1] 10-Ks for [COMPANY].
Identify the material changes between them in:
- Risk factors (new, removed, materially reworded)
- Critical accounting estimates
- Segment definitions or reportable segments
- Revenue recognition policy
- Off-balance-sheet arrangements
- Legal proceedings
For each change, quote the new text and the old text, then explain
in one sentence why this matters for an analyst.
A 10-K diff used to take a senior associate a full afternoon. With Claude or NotebookLM, it is a 10-minute task.
Building a Post-Print Note Template
Combine the above into a single reusable Claude Project. Inside the project:
- Save your firm's house style guide as a project knowledge file
- Save a sample of your best prior post-print notes as templates
- Save the company's last four transcripts and last 10-K
Then each quarter you only need to upload the new transcript and run:
A new transcript for [TICKER] is in the project. Draft a post-print
note in the style of the saved examples. Use the prior transcripts
in the project for tone comparison. 350 words. End with two
forward-looking questions for the next quarter.
You went from a four-hour task to a 20-minute review of an AI draft.
A Real Example: Reading Between the Lines
Suppose management on Q1 said "we feel very good about the pipeline." On Q2 they said "the pipeline is healthy." On Q3 they said "we have a solid pipeline of opportunities." On Q4 they said "pipeline development continues."
Each sentence in isolation is neutral. The pattern — escalating hedge words — is exactly what AI is built to detect. Ask Claude:
Are the four quotes I just pasted showing a deterioration in management confidence on pipeline strength? Explain.
The answer will be: yes, and here is why. Your trained intuition would have eventually noticed. AI catches it on the first read.
Key Takeaways
- Drop a transcript into Claude with a structured prompt for a two-minute post-print summary
- Use multi-quarter comparison to detect tone shifts that human readers miss
- 10-K extraction is a five-section prompt covering segments, risks, accounting, and commitments
- Year-over-year 10-K diffs surface what management cares about — a high-value, low-effort workflow
- Build a Claude Project per ticker so context compounds across quarters