Alternative data for investors: a buyer's guide

Alternative data is information from outside traditional market data, company filings and broker research that helps investors understand how a company, sector or economy is performing. Used well, it gives earlier or more granular evidence on revenue, demand, costs or risk. Used carelessly, it produces convincing charts that fail in live trading, or creates compliance problems that no signal is worth.

This guide explains the main types of alternative data, the technical questions that decide whether a dataset is usable, the compliance checks to run, and how to structure a trial.

Common types of alternative data

Type	What it can indicate	Typical caveats
Card transaction panels	Consumer spend by merchant or brand	Panel composition, merchant mapping, online versus in-store mix
E-receipts	Item-level baskets and prices	Panels can skew toward certain demographics and retailers
Website traffic	Demand and engagement for digital businesses	Estimated rather than measured; bot traffic; method changes
App usage and downloads	Adoption and engagement for app-based products	Platform coverage; estimates differ between providers
Job postings	Hiring plans, expansion, cost pressure	Duplicates, reposts and aggregator noise
Foot traffic	Visits to stores, venues and sites	Aggregated mobile panels; sampling bias
Satellite and geospatial	Activity at facilities, crop conditions	Cloud cover, revisit rates, interpretation models
Shipping and trade	Import volumes, supplier relationships	Country coverage and legal limits vary
Pricing and product data	Price changes, discounting, assortment	Collection coverage, product matching

You can browse providers in each area, for example card transaction data, website traffic data and job postings data.

Five technical questions that decide usability

Is the history point-in-time?

A backtest is only honest if each data point is what you would have seen on that date. Ask whether history is stored as originally delivered, or restated after later corrections, panel reweighting or changes to the ticker mapping. Restated history makes signals look better than they were. Ask for an as-of timestamp on every record, or for archived snapshots as they were delivered at the time.

How is it mapped to tickers?

Raw data describes merchants, websites, apps or facilities, not securities. The provider has to map these entities to companies and then to tickers, handling brands, subsidiaries, mergers, spin-offs and ticker changes. Ask how the mapping is maintained, whether historical mappings are preserved, and what share of the panel maps to listed companies.

How large and how representative is the panel?

For panel-based data, ask about panel size over time, how members are recruited, how the provider corrects for churn and demographic skew, and whether it publishes a panel-normalized series. A signal that moves because the panel changed is not a signal about the company.

How long is the history?

You need enough history to cover several reporting periods and at least one change in market conditions. Short histories invite overfitting. If history is short, be explicit about how much weight you put on the results.

How fast is it delivered?

Latency matters relative to the event you want to anticipate. Spend data delivered a few days after period end may still be early relative to quarterly results, yet useless for short-horizon strategies. Ask for the typical lag and how consistent it is.

Compliance: MNPI, provenance and privacy

Investment firms generally ask alternative data vendors to complete a due-diligence questionnaire before a trial. Typical topics include:

• Material non-public information: could the data include confidential information obtained in breach of a duty, such as data shared by an employee or under a confidentiality agreement?
• Source rights: does the provider have the right to license the data for investment use, and do its upstream contracts permit it?
• Web data collection: for collected web data, does collection respect site terms, avoid bypassing logins or technical protections, and avoid gathering personal data unnecessarily?
• Privacy: is personal data aggregated or de-identified, what thresholds prevent re-identification, and which privacy laws apply?
• Consent: for panels, did members agree to this use of their data?

Your compliance team will have its own standards. The point is to raise these questions before the research team invests weeks in a dataset. This guide is not legal advice.

How to run a trial that tells you something

1. Write the hypothesis first. For example: "Quarterly spend growth in this panel predicts reported revenue growth for these retailers." A clear hypothesis prevents fishing.
2. Agree the trial scope. Number of tickers, history depth, trial length and what happens to the data at the end. Make sure the trial licence allows research use.
3. Check data quality before modeling. Look for gaps, sudden level shifts, mapping changes and outliers.
4. Test out of sample. Keep a period aside that you do not touch until the end.
5. Account for decay and costs. A signal that works on paper may weaken once it is widely used, or once you include trading costs and the cost of the data itself.
6. Write up results either way. Negative results save the next analyst from repeating the work.

Pricing and contract terms

Alternative datasets are usually sold as annual licences, often tiered by coverage, history, delivery frequency or number of users. Expect to negotiate:

• Trial terms: length, data scope and what happens to trial data afterwards
• Usage rights: research only, use in live investment decisions, or sharing with portfolio companies
• Limited distribution: some providers cap the number of clients, at a price
• Termination: whether you can keep using models and research built on the data after the licence ends
• Methodology change notices: how much warning you get before panel or mapping changes

The total cost includes more than the licence. Budget for data engineering, research time and storage.

Building a sourcing workflow

Teams that use alternative data well treat sourcing as a pipeline: a backlog of candidate datasets, a standard questionnaire, a consistent trial process and a record of what was tested and why it was accepted or rejected. On DataStoa you can search across categories, compare providers and post a data request describing the signal you want, so providers can respond with proposals.

Frequently asked questions

What is the difference between alternative data and traditional data?

Traditional data covers market prices, company filings, analyst estimates and official economic releases. Alternative data comes from other sources, such as transactions, web activity, satellites or job postings, and usually needs more processing before it says anything about a company.

Is alternative data legal to use?

Generally it can be, provided the data was lawfully collected, the provider has the right to license it, and it does not contain material non-public information obtained improperly. Your compliance team should review each dataset.

Do I need a quant team to use alternative data?

Not always. Some providers deliver ready-made indicators or dashboards for fundamental analysts. Raw panel data and large feeds, though, need data engineering and statistical skills to use responsibly.

How long should a trial be?

Long enough to ingest the data, run your tests and review the results, often several weeks. Agree the length and the evaluation criteria before it starts so the trial ends in a decision.