A number is only useful with its source beside it.
An analyst pulls a quarter’s revenue into a model. A week later a colleague asks whether it was consolidated or standalone, restated or as first reported, and nobody can say without opening the filing again. We build research tools that keep that context fastened to the figure.
Two jobs that are easy to blur.
Delivering data means getting the right figure, from the right document, for the right period, into the place where people already work. It needs careful sourcing, consistent definitions and a trail back to the page the number came from.
Engineers can build, test and check that part closely with your research team.
Forming a view is the analyst’s job: deciding what a figure means, how it compares and whether it matters. Software can speed that up by keeping sources within reach.
It shouldn’t make the judgment for anyone, and good research tools are honest about which part is which.
DataStandard shows how we think about this. Our client owns it; we designed the platform, built it and still run and extend it. Researchers query the filings of Indian listed companies from inside Claude or ChatGPT, and each number comes back labelled with its document, the period it covers and whether it is consolidated or standalone. The analyst stays in a tool they already use and still has enough context to check the answer.
The same principle carries over to other tools a research team might want: an internal library of company documents, a screen across a sector, an assistant that drafts a first-pass note from filings. In each case the hard part is the data model underneath, and agreeing what has to travel with every number.
Indian filings bring their own traps. Companies change their financial year, restate earlier periods after a merger, and report in crores, lakhs or millions depending on the document. A figure that looks comparable across two companies can sit on different bases. The tool has to make those differences visible rather than quietly averaging over them.
Pieces we’d build for a research desk.
- Document ingestion that keeps the page reference for each extracted figure
- Definitions your team agrees once and applies the same way across companies
- Search across filings, presentations and call transcripts by company and period
- AI assistants that reply from your own sources and cite the document behind every reply, as described under internal copilots
- Saved work that colleagues can reopen later with its context intact
- Access rules for licensed data that must not leave the team
Asked by research teams.
Does OLN Labs own DataStandard?
The platform is our client’s. Our team created it for them and looks after its hosting, upkeep and new features.
Could you build something similar on our own data?
We can talk it through. The first questions are which sources you hold, what your licences permit and how your team checks a number today. Our AI development page explains how we take a tool like this from prototype to something people rely on.
Which AI tools would it work inside?
DataStandard works inside Claude and ChatGPT. For a new tool the choice depends on what your team already uses and your policy on data leaving your systems.
Close to this topic.
Show us how your team checks a number.
We’ll start from that habit and work out what a research tool should keep beside every figure.