The Primer playbook

Alternative data research

Find useful datasets, test what they tell you and keep the evidence up to date.

Guide 01

What is alternative data and when is it useful?

Alternative data is information outside a company’s own financial reports. It might be car registrations, website activity, job adverts or industry volumes. It can help you test a research question between results.

Do this

Start with a question about the business, such as whether demand is improving in an important market.
Ask what you could observe outside the company that relates to that question. For example, car registrations may help you understand a car maker’s end-market demand.
Ask Primer to explain what each measure actually captures and what it leaves out.
Choose one promising measure to investigate. You need a clear connection to the business before you need a large dataset.

Primer session message box with company mentions and Add files — Start in New session. Add the company with @ and attach your own work with Add files.

Try this

“Find some data that might help forecast a key KPI for @Company or explain what drives its performance.”

Guide 02

Find datasets for a company or research question

Ask Primer to look for a source you can use repeatedly. A useful dataset needs clear definitions, enough history and a way to obtain the figures.

Do this

Start a company session and describe the measure, geography and time period you need. Give the source website if you already know it.
Ask Primer to find possible datasets and inspect what can actually be accessed.
Compare coverage, update frequency, available history, definitions and access requirements.
Choose a source and ask for a small sample first. Review it before requesting the full history.

Try this

“Find datasets that could answer [question] for @Company. I need [measure] for [geography] from [start date], ideally updated [frequency]. Start with [source website, if known]. For each candidate, give the source link, definitions, history, update frequency and access requirements. Verify what you can retrieve and show a small sample from the best available source. If access is blocked or requires a subscription, explain what I would need to provide rather than claiming the dataset is available.”

Guide 03

Upload, clean and explore a dataset

Cleaning a dataset means making its dates, numbers and labels consistent so you can analyse it. Keep the original data so you can see what changed.

Do this

Start a session and use Add files to attach your data.
Ask Primer to inspect the file before making calculations. Check dates, units, missing values and duplicate rows.
Ask it to keep the original values and record the cleaning steps, such as converting dates or standardising company names.
Ask for a Data Workspace with a clean table and a simple chart. Open the returned link, or find the saved workspace in Library → Data. Check a few rows against the original file.

Library search with the Data filter selected — Use Library → Data to find and reopen your saved Data Workspaces.

Try this

“Inspect my attached dataset from [source], covering [period]. Show me any problems with dates, missing values, duplicates or inconsistent labels before cleaning it. Keep the original data and record every transformation. Create a Data Workspace with a clean table, summary figures and a chart of [measure]. Do not silently fill missing values or change definitions. Tell me what I should check in the original file.”

Guide 04

Compare a dataset with reported company performance

Test whether the outside data helps explain a company’s sales, orders or another reported measure. Compare the same periods and allow for delays between the two.

Do this

Start a company session and attach or identify the Data Workspace. Choose the company measure you want to compare with it.
Ask Primer to align dates, units and geography. Monthly industry figures may need combining into the company’s reporting quarters.
Ask whether the outside measure tends to move before, alongside or after the company’s results.
Review periods when the relationship breaks down. Ask whether the result remains useful in later periods that were not used to develop it.

Try this

“Compare [dataset or Data Workspace] with @Company’s reported [measure] over [period]. Explain the connection before calculating it. Align reporting periods, units and geography, and test any plausible delay between the series. Show the chart, calculations and periods where the relationship breaks down. Account for seasonality, definition changes and acquisitions where relevant. Use the earlier period to develop the comparison and the later period to test it. Do not describe a correlation as proof that one series causes the other.”

Guide 05

Turn a useful dataset into a research routine

Once you have a useful source and a repeatable analysis, a scheduled routine can check for updates. First confirm that Primer can retrieve the new releases; an uploaded file does not update itself.

Do this

Start from a dataset and analysis you have already reviewed. Identify its source, normal release timing and the Data Workspace to update.
Ask Primer to test retrieving the latest release and explain whether it can repeat the process. If it needs a file from you each time, keep that as a manual step.
If the source can be retrieved again, open Routines → New routine → Scheduled. Set a schedule after the source normally publishes and describe the work to repeat.
Have the routine check whether a new release exists, update the agreed workspace and explain what changed. Review its first run before relying on later updates.

New routine form with Scheduled selected — Choose Scheduled to check a dataset at set times. Set the timing after its usual release.

Try this

Use these instructions for a scheduled routine after you have verified that the source can be retrieved again.

“Check [source URL] for a new release of [dataset]. Use the agreed definitions and cleaning steps in [Data Workspace or Note]. Compare the release date with the latest data already saved; if there is no new release, say so and do not duplicate rows. When new data is available, update [Data Workspace], rerun our comparison with @Company’s [measure] and explain what changed. Preserve the source and release dates. If retrieval fails or definitions change, flag the problem rather than reporting stale data as new.”

Set up and manage routines

Explore the other sections