Daloopa

Daloopa: Source-Linked Fundamental Data

• Every number hyperlinks to its source document, with a stated average accuracy above 99 percent across millions of data points.
• 6,000+ global tickers with fourteen years of history and, by its own claim, four to ten times more data points per company than other providers.
• Six delivery surfaces: data sheets, an Excel add-in, the Scout agent, an API, an MCP server and cloud integrations.
• Built for the 13-week research cycle — 70 percent less time initiating coverage, two hours saved per ticker at earnings.
• A 47 million dollar Series C led by Brighton Park Capital, announced on the homepage.

Why Does Every Number Need A Source Link?

Daloopa's central claim is unparalleled transparency, with each number hyperlinked to its original source and an average accuracy above 99 percent across millions of data points.

Source linking sounds like a convenience feature and is closer to a governance one.

Consider what happens when a number in a model is wrong. Someone spots an implausible margin, and the question becomes where it came from — a filing, a transcript, a press release, or an analyst's own adjustment three quarters ago. Without a link, answering that means an hour of archaeology, and the model is untrusted until it is finished.

With a link it is a click. That difference changes how quickly a mistake is found and, more importantly, how confidently a number can be defended when someone senior challenges it.

The 99 percent figure needs the same reading any accuracy claim does. Averaged across millions of data points, one percent is still a large absolute number of errors, and the point of source linking is precisely that you do not have to take the average on faith.

A number you cannot trace is a number you cannot defend. Source linking turns an hour of archaeology into a click.

Why Auditability Matters More With Agents

Daloopa positions itself explicitly as the data layer under AI agents, and publishes a benchmark it calls FinRetrieval comparing how leading agent systems perform on financial retrieval against an auditable database versus the public web.

That framing is the most forward-looking thing on the site, and the reasoning holds up.

A human analyst reading a number they doubt will check it. An agent will not — it will use the number, build on it, and present a conclusion with the same confidence whether the input was a filing or a badly scraped summary page.

So the failure mode changes. With a human, bad data produces a caught error; with an agent, it produces a confident wrong answer several steps downstream from anywhere you would think to look.

Grounding agents in an auditable dataset is the structural answer to that, and it is the same argument that makes evaluation tooling like Promptfoo necessary rather than optional for anyone shipping agents into a regulated workflow.

Six Thousand Tickers And Fourteen Years

The coverage claim is specific: more than 6,000 public companies globally, fourteen years of history, and four to ten times more data points per company than other providers.

The depth figure is the interesting one, because breadth in financial data is close to a commodity.

Every vendor has revenue and EPS for large caps. What separates them is the company-specific operating metrics — the subscriber counts, the segment splits, the same-store sales, the bookings — that live in a footnote or a transcript rather than a standard accounting line.

Those are the numbers an equity model actually turns on, and they are the ones a standardised database usually omits because they do not map onto a common schema.

Fourteen years matters for a related reason: it spans a full cycle plus a pandemic, which is the minimum for asking how a business behaves when conditions turn. The question to ask a vendor is how restatements are handled across that history, since a comparison against figures a company has since revised is worse than no comparison at all. Once the data is clean, analysing it is a separate tool's job — Quadratic being the spreadsheet-shaped answer.

• Test coverage on the specific KPIs your models turn on, not on revenue.
• Ask how restatements are handled across the fourteen-year history.
• Click through source links on a company you know well.
• Pilot at earnings season, where the time saving is concentrated.
• Check the MCP and API surfaces if agents are part of your roadmap.

Scout, The API And The MCP Server

Six delivery surfaces sit on one data layer: downloadable data sheets, an Excel add-in for updating models in a click, Scout as an AI Excel agent, an API for programmatic ingestion, an MCP server for connecting any LLM, and cloud integrations.

That spread is a deliberate answer to a fact about this industry: analysts work in Excel and will continue to.

A data vendor that requires leaving the spreadsheet is asking an analyst to change the habit their entire workflow is built on, and that request is usually declined. Meeting them inside Excel is the only route that gets adopted.

Scout is the more ambitious surface — an agent that builds and maintains financial models with the data layer built in, described as delivering source-linked, hallucination-free outputs. Read that claim as scoped rather than absolute: grounding the retrieval removes one large class of error, not every class.

The MCP server is the piece worth noting strategically. It lets a firm's own agents, in whatever client they run, reach audited financial data — which is a considerably better position than being one more destination an analyst has to visit.

Reading The Customer And Time-Saving Claims

Two sets of numbers sit near the top of the page: that 185 or more of the largest hedge funds, mutual funds and bulge bracket banks rely on the platform, alongside five named AI companies; and that it cuts 70 percent of model-building time when initiating coverage while saving roughly two hours per ticker during earnings season.

The customer counts are unqualified as to what the relationship is, which is normal and worth noticing anyway. A firm running a pilot on one desk and a firm with a platform-wide licence both count as one.

The time-saving figures are more useful because they name a task. Initiating coverage is a genuinely enormous job — reconstructing years of financials into a model's own structure — and 70 percent of it is a specific, testable claim.

Two hours per ticker at earnings is the figure that compounds. An analyst covering twenty names during a reporting season is the target buyer, and forty hours across three weeks is the difference between reading the results and merely entering them.

Both are worth testing on your own coverage rather than accepting, because the saving depends entirely on how closely the vendor's structure matches the models you already run. None of it touches technical analysis, where a charting platform like LuxAlgo answers an entirely different question.

Fitting Daloopa Into The Research Cycle

Daloopa frames its value around a 13-week investment research cycle, and the sensible evaluation follows the same shape rather than sampling features.

This sequence tests the claims where they are actually load-bearing.

1. Pick one company you already model deeply and compare data point by data point.
2. Click through the source links on anything that differs.
3. Time an initiation on a genuinely new name against your own baseline.
4. Run the add-in through one full earnings update before judging.
5. Check the company-specific KPIs, not the standard accounting lines.
6. Test the API or MCP surface if agents are on your roadmap.

Daloopa: Coverage And Backers