Alternative data sources: what they are and how to read them
By Maya Koeva · August 20, 2026 · 8 min read · Updated September 30, 2026
Alternative data is any information about a company that does not come from the company itself or from the market's official feeds. Filings, earnings calls, prices, and analyst reports are the traditional diet. Alternative data is everything else that leaves a measurable trace: credit card swipes, satellite photos of parking lots, job postings, app downloads, and what thousands of people are saying online.
A company reports its quarter once every three months, but the quarter itself happens every day, in stores and on websites and in hiring plans. Alternative data sources are attempts to watch it happen instead of waiting for the announcement.
What counts as alternative data
Traditional data is whatever every investor receives on a schedule: quarterly reports, SEC filings, price and volume, official economic releases. Alternative data is the exhaust a business gives off between those announcements. The differences fit in one table:
| Traditional data | Alternative data | |
|---|---|---|
| What it is | Filings, earnings, price and volume | Card swipes, satellites, job posts, chatter |
| Arrives | On a schedule, quarterly | Continuously, as business happens |
| Audited | Yes | No |
| Who sees it | Everyone, at the same moment | Whoever collects it |
| Priced in | Almost instantly | Slowly, if at all |
The last two rows are the appeal and the catch in one: alternative data can be early because few people see it, and it can be wrong in ways nobody checks.
The main alternative data sources
Six types cover most of what gets bought and sold under this label:
| Source | What it watches | The famous case |
|---|---|---|
| Transaction data | Card-panel spending at consumer companies, weeks before the earnings call | Bloomberg's Second Measure: a panel of more than 20 million US consumers, updated within days of the swipe |
| Web and app activity | Traffic, downloads, daily active users, review counts | Snapchat's 2018 redesign: one-star reviews and falling downloads preceded the first reported drop in daily users |
| Satellite and location | Parking lots, ports, oil-tank shadows, foot traffic | Analysts counted cars in Walmart parking lots to estimate quarterly sales ahead of the report |
| Hiring data | Job postings and headcount changes | Apple's car project surfaced in 2015 when reporters traced a hiring spree of automotive and battery engineers |
| Supply chain records | Customs manifests and bills of lading | Panjiva and ImportGenius index public US import records, shipment by shipment |
| Social sentiment | What investors say in public, and who says it | GameStop, January 2021: the trade came together in threads anyone could read |
Two things the table understates. Each source has a blind spot the vendor rarely advertises: the cards in a panel may not shop like the average customer, and a job posting can mean a new product line or a backfill. And the six are not equally reachable: the first five are institutional products, while the sixth sits in public, the most accessible source on the list and the noisiest. We wrote a full explainer on market sentiment.
Alternative data examples
The table is the catalogue. These are the cases people mean when they say alternative data works, each with the catch that came with it.
Card panels and the Chipotle recovery. After the E. coli outbreaks in late 2015, the question was when customers would come back. Card-panel vendors could watch the answer week by week: spending at Chipotle stayed depressed for months, then recovered in a slow curve that the panels showed long before the quarterly comps confirmed it. The catch: a panel of millions of cards is still a sample, and the vendors spend most of their effort correcting for who is in it.
Parking lots and quarterly sales. Satellite firms counted cars outside big-box stores, Walmart among them, and sold the counts as an early read on the quarter. Academic work later confirmed the counts predicted earnings surprises, and something else: the funds that bought the feed traded on it days before the report, so the edge belonged to whoever had it first. The catch is the last question in the checklist below, in one dataset.
App reviews and the Snapchat redesign. Snap's early-2018 redesign was unpopular, and the first evidence was public: one-star reviews piling up, downloads sliding in the store rankings. The reported drop in daily users arrived a quarter later. The catch: review counts also spike for reasons that never touch the business, so the read needs a second source.
Job postings and the Apple car. Reporters and analysts pieced together a hiring spree of automotive and battery engineers in 2015, years before Apple confirmed a car project. Hiring data is the cheapest alternative source on this list and one of the slowest: a posting says what a company intends, not what it will ship, and Apple never shipped the car.
Employee reviews and stock returns. A 2019 study in the Journal of Financial Economics found that changes in a company's Glassdoor ratings predicted its next earnings surprise and its returns over the following months. Free, public, and slow: the signal builds over quarters, not days.
Customs records and the iPhone. Every container that enters the United States leaves a bill of lading, and firms like Panjiva index them. Analysts have used those records to track Apple's shipments from Foxconn ahead of launches. The catch: a manifest shows what moved, not what sold, and air freight, which Apple uses heavily, leaves a thinner trail.
Tanker tracking and oil supply. Ships broadcast their position, and firms like Kpler and Vortexa turn those broadcasts into estimates of how much crude is at sea and where it is headed, days before official inventory numbers. This is alternative data at its most institutional: real-time, expensive, and useless without a model of what the flows mean.
Public conversation and GameStop. In January 2021 the trade came together on a subreddit anyone could read, weeks before the squeeze. It is the standard case for watching social data. It is also the case against reading it raw: the same threads produced a hundred names that went nowhere, and nothing in the volume told you which was which. The Where Quantral fits section below shows what a graded read of the same source looks like, with the Nebius drawdown as the example.
Congress filings and the disclosure lag. Members of the House and Senate report their stock trades under the STOCK Act, within 45 days of the trade, in price bands rather than exact figures. The dataset is free, structured and public, which is rare on this list. The catch is the lag and what it does to the edge: grading every member's filings as calls from the day they became public puts the whole set a little under 44% correct, as the congress stock trades write-up lays out. Read one member at a time and the filings say more: Nancy Pelosi's show most new positions starting as call options, and several "sales" that were charitable gifts.
Insider filings and the two-day window. When a company's officers, directors or 10% owners trade its stock, they have two business days to report it on SEC Form 4. The data is free and fast, and it is noisy: a lot of filings are sales, grants and option exercises made for reasons that have little to do with the company's outlook. Open-market buys are the part worth watching, and they are all the insider trading tracker keeps. Free screeners like OpenInsider list every filing instead; Quantral vs OpenInsider compares the two approaches.
Broker research on Telegram. In South Korea, brokerage research desks post short notes on companies and sectors to their own Telegram channels, outside the usual terminal feeds. Quantral reads that Korean broker research on Telegram, along with Korean investing blogs on Naver, next to finance X, Reddit and Substack, and shows both in English. The catch: each note is one desk's view, and sell-side research has a well-known lean toward buy ratings.
Who buys alternative data
Mostly not individuals. The classic buyers are hedge funds, buying through specialist vendors on contracts that run five or six figures a year. A large fund might subscribe to dozens of datasets and employ a team to clean and test them. The "multi-billion-dollar alternative data industry" you read about is funds buying from vendors, not apps for retail investors.
Not all of it is locked up, though. Some of the most useful sources were never behind a paywall: job postings, app reviews, search trends, and public conversation are all free to anyone willing to read them carefully. The barrier for individuals is the reading, not the access: separating signal from noise takes work that funds pay teams to do.
How to judge any alternative data source
Whatever the source, the same five questions decide whether it deserves your trust:
- What does it see? Every source covers a slice, never the whole. A card panel sees its cards, a sentiment feed sees the accounts it follows. And when a source shows nothing, ask which kind of nothing it is: did it watch and see nothing, or did it not watch?
- Is it early, or just different? A dataset only helps if it moves ahead of the reported numbers. Some alternative data restates what the price already knows, with extra steps.
- Who is in the sample? Panels skew young or urban, app data skews tech-savvy, social data skews loud. Each skew is a way to be confidently wrong.
- Has it been graded? The only fair test of a signal is its record against real outcomes. A source that has never been scored against what happened is a story, not a signal.
- What happens when everyone has it? The more widely a dataset is sold, the faster its edge gets traded away. Yesterday's exotic satellite feed is today's consensus input.
Where Quantral fits
Social sentiment is the one alternative data source individuals can reach, and raw, it fails most of the five questions above: unknown sample, no grading, and a volume of noise that buries the signal. Quantral's approach is to treat it the way a fund would treat any raw dataset. We track a defined set of accounts across finance X, Reddit, and Substack, grade every directional call those accounts make against what the stock did, and weight the conversation by each account's credibility score. The result is one alternative dataset, cleaned and graded, readable at a glance, without the fund-sized contract. The same dataset is reachable from a chat window through Quantral's MCP server.
What that gives you in practice, from our own published write-ups:
| The job | What the graded data showed |
|---|---|
| Holding through fear | Nebius fell 48% and the accounts with records held close to 8-to-1 bullish through the whole drawdown; the stock came back 75% off the low |
| Telling a hot sector's names apart | Headlines treated the memory trade as one story, while graded calls ran 3.5-to-1 bullish on Micron and 2-to-1 bearish on Western Digital; the two stocks went different ways |
| Catching a turn | SpaceX scored 19 for a month, then flipped to 77 while the raw crowd stayed two to one bearish; it cleared its lockup, up 21.8% since the post ran |
| Vetting a voice you follow | The chip leaker jukan05 grades out at 58.6% right across 70 calls, third on our six-month accuracy board, and wrong roughly four times in ten |
The first row of that table, drawn from the raw feed, looks like this:
The bottom line
Alternative data is any measurable trace of a business outside its official reporting: transactions, traffic, satellites, hiring, shipping, and conversation. Most of it is built for and sold to institutions, but the mindset behind it is free to borrow: watch the quarter happen instead of waiting for it, know what your source can and cannot see, and do not trust a signal nobody has graded.
Quantral surfaces signals and context from public sources to support your own research. Nothing here is financial advice or a recommendation to buy or sell.