How FFP One works

Here's how researching a company usually goes, what changes when you add AI, and what FFP One does differently.

Say you've found a small company you like

It's a microcap, it looks interesting, but there's no sell-side analyst covering it. So the research is on you.

The traditional way is to assemble it yourself: pull the 10-K and the other filings, look up the share price, search for news, then piece it together in a spreadsheet and a set of notes. It works, but a 10-K can run to a few hundred pages, and that's a long afternoon at least before you've formed a view.

Why not just ask a chatbot?

More recently, you can ask a general-purpose AI assistant, ChatGPT or Claude or one of the many, many others, and it'll pull something together. It will do it quickly and it will sound very confident. But you don't control where it looks. It may cite a few links, but you don't know what comes from a high-quality source, what has been pulled from social media, what is from model training data, and what has been pulled completely from the ether (the dreaded AI ‘hallucination’).

And when it does go to the filings, their sheer size works against it. Several hundred pages at once can swamp what a model can keep in view at one time (its context window), so it skims, drops parts or loses the thread. FFP One uses the same kinds of models; the difference is what they work from.

What FFP One does instead

FFP One doesn't start from scratch. Think of it like mise en place in a restaurant kitchen: the chopping and measuring is done before service, so when an order comes in, the cook goes straight to what's needed. FFP One does the reading ahead of time and keeps the results in the FFP One Data Layer, its research database, so when you ask a question, the AI goes straight to the right figures or the right section of the filing instead of wading through the whole document.

That's where the speed, the low cost and the accuracy come from.

Two paths: the numbers and the words

Investment research runs on both, so FFP One builds both, across over 2,100 companies andover 5,200 funds.

The numbers. Each company's reported financial statements go into standard tables, so revenue, margins and cash flow line up across years and across companies. Standard measures (P/E, return on equity, the Piotroski, Altman Z and Beneish scores) are calculated ahead of time, the same way for every company. Fund holdings are linked to the companies they hold, which is how FFP One can measure a whole fund as if it were one company. Daily share prices and macro data from FRED (interest rates, Treasury yields, inflation) sit alongside.

The words. Each filing is split into its sections (the business, the risk factors, management's discussion) and each section is summarized with its key points, so the AI can work with what management says as readily as with what it reports. Companies are tagged by the themes their own filings discuss, such as AI infrastructure or geopolitical risk. The original text stays searchable word for word.

Both paths are checked. Before anything is summarized, the filing's text is checked against the original SEC document. Every dollar figure and percentage in a summary is checked against the section it came from, and when a Research Report quotes a filing, the quote is matched word for word against the original.

Translated out of data jargon: the repetitive part of research (finding the number, lining it up with last year's, reading what management said about it, then doing the same for the next company) is already done when you arrive.

Built to be relied on

FFP One is an AI-native platform, built single-handedly by Randall Helms, a data leader with more than 15 years of experience working in data in finance, B2B SaaS and consumer media. Every company and fund goes through the same process, and every stage has its own checks, from the filing text to the finished answer, because a research tool is only worth using if you can rely on what it tells you.

What does that get you?

• Speed: asked who Marvell competes with and how their margins compare, a Research Report took the competitor list from Marvell's own 10-K, ranked the peers on their margins and wrote up the comparison in about five minutes. Reports usually finish within 20 minutes, and they run in the background while you get on with something else. Imagine having your own on-tap equities analyst who can produce a high-quality deep dive on your topic of choice in mere minutes!
• Cost: looking up the figures and passages an answer needs is much cheaper than rereading whole filings. A chat answer typically uses 2 to 7 credits, and a full Research Report 100 to 300 on Claude, or 5 to 40 on the China-based models.
• Accuracy: the AI works from figures the database already holds, for the right filing and the right year, not from its memory or a skim of a PDF. A chat answer that uses FFP One's data lists the filings it drew on, with a link to each, and a Research Report ends with a table showing where its figures came from.
• Questions that span many filings: “Which funds have the most exposure to the AI buildout?” means joining thousands of funds' holdings to what each company's own filings say. With that already linked, it's one question in Chat.

Where does the AI come in?

The database is the foundation, and the AI is how you reach it. When you ask something, the AI works out which figures and passages it needs, looks them up and writes the answer. When an answer needs a calculation nobody stored in advance, such as growth over five years, the AI works it out from the figures it looked up.

Then the answer is checked before you see it. Chat compares the numbers in its answer with the data it fetched and sends back an answer that doesn't hold up, and a Research Report gets a second AI pass on its figures and tells you if that pass didn't finish. Every answer is labeled with what it drew on, and links its source filings when it used FFP One's data, so you can check anything you'll rely on.

What's not in yet

Quarterly (10-Q) figures are on the way; today FFP One has annual reports and recent 8-K filings. Fund holdings come from each fund's latest N-PORT filing, so they're usually last quarter's. And it's a research tool, not an adviser: nothing it writes is investment advice.

For the detail on sources, labels and checks, see where the numbers come from. To see it on questions you'd actually ask, read the use cases.