Inflation. Weather. Employment. Energy prices. Who lives in the trade area, and how that changed. The context your forecasts are missing is published, current, and free — and most teams never get to it. Not because it wouldn’t help, but because working out which source fits, before you commit to building anything, takes an afternoon nobody has.
External Data — two minutes, using the demo data that ships with every account.
The data is good. The search is what stops people.
The Federal Reserve, the Census Bureau, the Bureau of Labor Statistics, NOAA, the Energy Information Administration, SEC EDGAR and the World Bank publish an enormous amount of genuinely useful information on open APIs, for free, and have for years. Demand models get better with weather. Cost models get better with input prices. Market sizing gets better with real demographics. None of that is controversial.
What stops it is the step before the work: which of the thousands of published series is about your problem, and does your data even have a column it could be joined on? That question is genuinely hard to answer quickly, it has to be answered before any value can exist, and it competes with whatever is due Friday. So it gets deferred, and the value sits there unclaimed.
What it does
In the Leverage tab, External data takes a use case you’ve already framed and shows you which public sources fit the shape of that work, and how to connect them. It reads the objective, the assets, their topics, and the columns they declare. It does not read your data.
One thing worth being direct about, because it’s unusual: we don’t ingest public data and blend it into yours. We never fetch your rows, never join anything, never store a third-party credential, and never send any part of your work to an outside API. You get the match and a connection recipe you run in your own environment, with your own key. The analysis stays yours.
Two questions, answered free
Before any AI runs and before a credit is spent, you get a readiness check on the two things that determine whether a source can actually help:
Topic match
Is there a public source about this subject? Scored from the use case objective and the topics on its assets.
Connection key
Is there a shared column to match it on? Usually a date, often a place.
They’re scored separately because they imply different next moves. Weather is an obvious fit for a demand forecast — but if no asset declares a date or a region, the answer isn’t “no relevant source.” It’s “good source, no key yet,” and the fix is fifteen minutes of cataloging rather than a dropped idea. Often the thing standing between you and a useful source is smaller than it looks.
And it’s honest about the ones that don’t fit
Point it at something like Quarterly executive KPI delivery and it declines:
We would not spend your credit on this one.
This use case is about a process rather than a measured subject, so no public dataset fits it — renaming or re-tagging assets will not change that. Pick a use case that measures something (demand, cost, headcount, market) and the scan will earn its credit.
That’s a feature, not a warning about your odds. A tool that finds a match for everything is a tool whose matches mean nothing. Because it will say when a use case isn’t a fit, you can act on it when it says one is — which is the whole reason to look.
Fit is a claim we can defend
Every suggestion carries two labels. Match strength — Strong match, Worth testing, or Speculative — describes how well the source fits the shape of your work. The second never changes: Effect: not yet tested.
We could have put a predicted lift there. But nobody can know whether a weather series improves your forecast until you run it and measure, and an invented number would be worth less than nothing the first time it was wrong. Fit we can assess from your metadata and stand behind. What it does for your model is yours to find out — and now you can find out in an afternoon that starts from a working script instead of a blank page.
What you get when something fits
A starter script for your platform: the API call, a landing table, the suggested join, and the attribution the source requires. It runs in your warehouse with your own key. We never execute it and never see the result — and where the join column was inferred from field names rather than confirmed, the script says so in a comment.
Accept a source and it becomes a cataloged asset with an enriches relationship to the work it feeds. It appears in lineage and change-impact like any other dependency, so in six months the answer to “why does this model have a CPI column?” is in the graph rather than in somebody’s head. It doesn’t count against your plan’s asset limit — charging you for taking our own recommendation would be a strange way to encourage it.
What it costs to look
The readiness check, the source list, previews, recipes, accepting and dismissing are all free. Only the scan itself costs one AI credit. It works on every plan including the free Explorer trial, and needs no cloud provider configured.
The context is already out there
Load the demo estate and watch it evaluate six use cases in about a minute, then point it at work of your own. No cloud provider needed, no credit card, and no data leaves your environment.
See the plans →