WorkBuddy batch research report summaries: Saving two afternoons during earnings season

Conclusion first: For batch research report summarization, WorkBuddy's ROI is excellent in my tests, provided you clarify rules with it upfront.

Saw news this morning about Kunlun Wanwei's Q2 net profit surging 2382%; comments section boiling. My first reaction: Where exactly in the raw PDF is this data hidden? Manually flipping through semi-annual reports would take at least half an hour. I've been using WorkBuddy for batch earnings summary scenarios for two weeks, took detours, documenting the smoothed-out workflow.

Specific Workflow

My environment is Web version of WorkBuddy on Mac. May not apply to everyone, but core logic should be similar. Using Kunlun Wanwei's interim report as example, path is roughly:

Step 1: Drag earnings PDFs into WorkBuddy document area. Note: Don't drag more than 10 files at once. Tried dumping 20; processing started serializing at file #15, summaries mixed data from other companies. Details below.

Step 2: Input extraction instructions in chat box; be specific. I wrote: "Extract Net Profit Attributable to Parent, Deducted Non-Recurring Net Profit, Revenue, Gross Margin, Debt Ratio from each doc, output as table." If you just write "Summarize this earnings report," it likely gives vague overview, missing indicators.

Step 3: Output as Excel. Key point: Add "Please output as table, including page number and original sentence for indicator location" at end of instruction. First time I omitted this, got plain text summaries with no source for numbers; couldn't find corresponding original text during verification.

Result: Quick check showed data basically accurate. Kunlun Wanwei's Q2 figure of 1.975 billion was grabbed from Consolidated Income Statement on Page 3; page annotation correct. Efficiency boost ~60%. Work originally requiring manual flip through 5-6 PDFs done in under 10 mins.

Pitfall Avoidance

Biggest pitfall: Serialization issue during batch processing. Day before yesterday, dragged in six different companies' earnings together. Result: Investment advice sections from two materials mixed; Company A's rating appeared in Company B's summary. Dangerous, especially for preliminary screening. If unnoticed and put into report, disaster.

Later practice: Process only different quarters of same company, or few companies in same industry at once, but add "Independently annotate data for each document, do not mix" in instructions.

Another pitfall: PDF scanned copy recognition. Clients often send scanned earnings. WorkBuddy can't directly extract structured data from tables in images. Official support says no OCR. My solution: Feed to ChatGPT image recognition first, convert to text draft, then feed to WorkBuddy. Extra step but unavoidable.

Configuration Suggestions

Personal habit: Create folder named "Earnings Season Raw Materials" in WorkBuddy. New subfolder each quarter, dump received reports/PDFs there. Use fixed extraction template in chat box.

Reason: WorkBuddy remembers chat history. Reusing same template stabilizes output format. First run might have messy table columns; second run aligns referencing previous output.

Permissions: Team of four shares one workspace. I gave colleagues member permissions; only I edit folder structure and adjust templates. Configurable in team settings. Recommend at least one person manages templates, otherwise everyone has different tactics, archiving becomes chaotic.

Daily Ops: Weekly light check. Randomly pick one/two tables run last week, verify against original PDFs. Mainly confirm if it used wrong page numbers. Takes <15 mins, prevents error accumulation after continuous weekly runs.

Summary: WorkBuddy helps preliminarily structure info in earnings summaries, saving half the time. But final number verification and judgment must be done manually. At least that's how I use it.