A week of typing, per quarter
Opening statements one at a time and copying NAV, called and distributed into a workbook is the single largest recurring block of work in most back offices, and none of it is judgement.
Statements
Capital account statements, read once and turned into data
Every quarter arrives as a folder of PDFs in a dozen layouts. Re-keying them is a week of work that produces no insight. This reads them into your portfolio and keeps the page each number came from.
In use at a small number of family offices and investment managers
What the quarter costs you today
No administrator formats theirs the same way
One calls it Ending Capital Account Balance, another Net Asset Value, a third splits it across two tables. Templates break the moment an administrator changes their layout, which they do without telling anyone.
Nobody can prove where a number came from
Once a figure is in the spreadsheet it has no history. At audit, or the first time two sources disagree, someone goes back through the PDFs to reconstruct what was read and when.
What you get instead
Statements in, records out
Forward a statement or connect the mailbox it arrives in. Opening balance, contributions, distributions, fees, allocated gain and ending NAV are read into the position they belong to, whatever the layout.
Every figure keeps its page
Each extracted value links back to the document and the page it was read from. A number in your portfolio is never an assertion. It is a citation you can open.
Uncertainty waits for a person
Extraction is confidence-gated. Fields the model is sure about post straight through; anything below the threshold is queued for review with the source shown beside it. The system would rather ask than be quietly wrong.
Your drive keeps a copy
Documents can be mirrored into your own Google Drive or SharePoint in the folder structure you already use, so the paper trail survives independently of us.
Where the data sits
Hosted in the EU
Documents and extracted data stay on EU infrastructure, under a DPA that names every subprocessor. Enterprise deployments can run in your own environment.
Documents are not training data
Your statements are processed to produce your records and nothing else. They are never used to train a model, ours or anyone's.
A full audit trail
Who read what, when, and what changed afterwards. Every correction to an extracted figure is recorded next to the original, so the history is legible rather than overwritten.
Questions we get about extraction
How accurate is it?
The honest answer is that accuracy alone is the wrong measure, which is why the pipeline is confidence-gated rather than tuned for a headline number. High-confidence fields post automatically, everything else is held for a person. You never get a silently wrong figure; occasionally you get a field to confirm.
Does it work on scanned statements?
Yes. Scans go through OCR first and then the same extraction and confidence gate. Scan quality shows up as more fields queued for review rather than as wrong numbers.
What about K-1s and other tax documents?
K-1s and similar documents are read and stored against the position, and the figures your accountant needs come out with them. They are not a substitute for filing advice. They are the numbers, in one place, with the paper attached.
Can we bulk-load a backlog?
Yes. Historical statements can be loaded in bulk to build the position history back to inception, which is what makes an IRR from your own data worth anything. Most firms start here.
What if an administrator restates a prior quarter?
The restatement is recorded as a new document against the same period rather than overwriting the old figures. Both versions stay visible, which is the only way to explain a moved number six months later.
Get the quarter back
Send us a redacted statement from your least cooperative administrator. We will tell you plainly what comes out of it.