Skip to content

Research a company after its quarterly report

You give this workflow a stock ticker. It reads the company’s last three quarterly reports, its prices, its listed competitors and its news, and it makes three files in your Drive: a company report, a peer comparison sheet and an industry report.

  1. It reads the company’s last three 10-Qs on SEC EDGAR, its prices, and the news of the last 30 days.
  2. AI steps take the figures from each 10-Q: revenue, margins, earnings and the segments.
  3. An AI step picks the listed competitors. The workflow reads the same figures and prices for each of them.
  4. AI steps write the company report and the industry report.
  5. The workflow makes the two reports as Google Docs and the peer comparison as a Google Sheet, with one row for each company and the quarter of each row.

The workflow has 39 steps, and 4 of them use an AI model. When the files of a company exist already, the workflow does not make them again.

The ticker of the company. The Drive folder for the files is set in the workflow.

  • Google Docs, Google Sheets and Google Drive, from one Google account that you connect;
  • Finnhub for stock prices, connected with your own Finnhub key;
  • SEC EDGAR for the 10-Qs. It is public and needs no account;
  • web search for the news, which Toolscaled gives every workflow. A web search costs credits.

Two Google Docs and one Google Sheet in your Drive folder.

Each company that you follow files a 10-Q three times a year, and each time the research is the same work. You give the workflow a ticker, and it does the whole research with no chat and no person. The person in the build below starts it by hand when a company files.

One chat in claude.ai with the Toolscaled connector: 17 messages from the person, 99 minutes. This is the largest build of these examples. Many messages are “Continue”: Claude stopped at a clean point when a reply grew long. The person typed these messages.

  1. Use Toolscaled. Each time a company I follow files its quarterly report, I research it. I want that done for me. Today the company is Gorman-Rupp (GRC). Here is what I do now by hand:

    1. A report on the company, as a Google Doc: how the stock and the business did, what its last three 10-Qs say (revenue, margins, earnings, the segments, and what changed), and the news of the last 30 days that matters for the stock.
    2. A Google Sheet that compares the company with its listed competitors, one row per company: price change over 1 year and since the start of this year, revenue of the last quarter, revenue growth against the same quarter a year before, and gross, operating and net margin, from each company’s last 10-Q. Each row says which quarter it is.
    3. An industry report, as a Google Doc: the state of the company’s industry and what drives it now.

    Put all three in my Drive folder “Toolscaled benchmark/stock research”. I already connected Finnhub for stock prices.

    Please do it now for Gorman-Rupp, so I can see it. Then make it something that runs by itself for any company I name.

  2. Yes, please create them.

  3. I’ll start it myself when a company I follow files. No schedule.

  4. Yes, I’d like it to work by itself. Try Badger Meter (BMI).

  5. Continue.

  6. Continue.

  7. Continue.

  8. I’m not sure what that check means. Which would you recommend?

  9. I’d rather keep that check on. Can you make it run by itself without turning it off?

  10. Continue.

  11. Both, please, so I can see the files.

  12. Are the test runs done yet? Can you show me what came out?

  13. Continue.

  14. Are the test runs done now? Can you show me what came out?

  15. Is it done now? Can you show me what came out?

  16. Your suggestion is fine.

  17. Yes, please add those two fixes and run the three companies again. For 7, skip the bad competitor and finish with the rest. For 8, skip it if the files already exist. As I said, I’ll start it myself and name the company each time, so no schedule and no list.

Claude did the research once for Gorman-Rupp, made the three files, and saved the workflow. Then it tested the workflow by itself on three companies in five rounds, and fixed what failed in each round: a failed run that left empty docs, a report that was too long for one model call, a competitor whose figures it could not read, and files that a second run made again.

We ran the workflow on a company that the build chat never saw: Preformed Line Products, at the close of 2026-10-30. The grade counts 63 figures from three 10-Qs, the peer set, 4 news items that matter and 8 judgment points. We compare it with an agent that did the same task from the same request. How we measured gives the method.

Workflow Agent
Time a run 148 s 6.8 min
Cost a run 59 credits, about $0.11 about $0.23
Tool calls 34 steps 50
Figures, of 63 39 to 40, none wrong 30, 3 wrong
Judgment points, of 8 5 4
News items, of 4 3 2
Same answer in 5 runs figures 0.99, peer set 0.20 one run

Neither way is strong on this task with GPT-6.1 Terra. A Claude chat on Opus 5.5 got 8 of 8 judgment points, in 12.3 minutes. The workflow time holds waits, because about 40 runs were open in the workspace at the same time. Its model calls took about 71 s. These 5 runs did not make the files again, because the files of an earlier run existed.

The build chat cost about $41.33 at list price, and your chat plan pays for it. At these numbers the build pays back after about 348 runs.