PageFinderv4.2.11.22Sign in

Blog

How many documents do you actually have? Auditing a company drive in an afternoon

An office manager, an insurer’s questionnaire asking how many client records the firm holds, and the day it took to get a number she could defend. A composite case, and where a tool that reads the drive fits.

November 21, 2026 · 7 min read · PageFinder team

7,686documents · 43,509 pages11kinds, invoices most1992–2026years the documents are about406identical copies52could not be read1,750companies and people named

The people and the firm in this piece are a composite of situations we have seen while building a document tool; the names are invented. The numbers are the kind we see.

Maria runs the office of a fourteen-person accounting practice. In March the firm's insurer sent the annual cyber questionnaire, and question 7 was new: Approximately how many client records do you hold, and what is the oldest?

She did what anyone would do. She opened the shared drive, looked at the folder list, and tried to think of a number. There were 212 client folders. Inside each, anything from a dozen files to several hundred, going back to a migration from the old server in 2021 and, before that, to a scanner that had been in the corner since the firm started. Her first answer was "about 30,000 files, oldest around 2004". She did not believe it, and neither would the insurer.

The afternoon she spent on it

The listing was the easy part. The sync client had a copy of the drive on her machine, and a colleague's script walked it in a few minutes: 41,803 files, 38 GB. That answered "how many files". It answered nothing else.

The oldest? Every file's modified date said 2021, the week of the migration. The real dates were inside the documents, and the documents were not going to open themselves. She sampled: fifty files picked at random, opened one by one, the year written down. Eleven were scans, and for those the year was on a stamp or a letterhead. The oldest in her fifty was a 1998 lease. Extrapolating from fifty files to forty thousand gave an answer with error bars wide enough to drive a van through.

What kinds of records? Tax returns, accounts, payroll, letters, bank statements, engagement letters. Folder names helped some of the time; file names helped less (scan0042.pdf, final v3 (2).docx). The honest answer was "mostly tax and accounts, we think".

How many were duplicates? The script could hash files, and it found 4,100 identical pairs. She had not known that number existed to be found.

By five o'clock she had a listing, a sample, and a form answer she could defend for about two minutes under questioning. It had taken a working day, and it would be out of date by the next client onboarding.

By handinventory · pages · hashessampling for kinds and yearstwo days · five of six answers · redo when it must be currentRead where it isabout an hoursix answers on a dashboard · stays current as files change
Two days by hand against about an hour indexed, and six answers instead of five

What the question was really asking

The insurer's question was a records question dressed as a security question. The firm could not say how many records it held because nobody had ever needed to know. Neither could it say which records were past the retention period, which were scans nobody could search, or which clients' files sat in two places under different names. The questionnaire exposed that; a migration, a data room or a subject-access request would have exposed it just as well, on a worse day.

Maria's partners made two decisions. First, to write the six numbers down once a year — how many, what kinds, which years, duplicates, unreadable, who they are about — so the next form is a comparison and not an estimate. Second, to stop guessing the dates.

Where we think a tool like ours helps

Maria's afternoon was mostly opening files to read three facts off each page: what it is, when it is about, who it is for. That is work software should do, and it is what PageFinder does with a drive like hers. Connected read-only to the OneDrive, with the client folders chosen and the partners' personal folders left out, it reads every file where it is — scans from the page image — and gives each one a kind, the date it is about, the names in it and a line of summary. The six numbers sit on a dashboard as the indexing runs, identical copies are counted by content, and the unreadable files are listed with the reason. The whole catalog exports as a spreadsheet, which is the questionnaire answer and the retention review in one.

It does not decide anything. Which 1998 leases to keep, whether a duplicate is the signed copy, what counts as a client record for the insurer — those remain the partners' calls. What changes is that they are making them from a count rather than a guess, and the count is there the next morning.

If you have a drive like Maria's, start with one folder and see whether the numbers match what you would have guessed: PageFinder for OneDrive.