Lesofi
← Writing

How many documents gigsdata processes

Roughly a thousand documents a day, 370,000 a year, and 2.23 million since the platform went live. These are counted figures, not estimates — and the counting rules matter as much as the numbers, so they are set out in full below.

Throughput is the first question anyone asks about a managed extraction platform, and it is the one most easily answered badly. A number quoted without its denominator — per what, counted how, over which window — is not a fact about the system, it is a marketing figure. So here is the whole thing: the volumes, the shape of the demand behind them, and the definitions that decide what counts as one document.

Everything below comes from the gigsdata document intake ledger, which holds one record per document received. It is a complete population back to 13 January 2020, not a sample. The reporting period is the twelve months ending 31 July 2026; calendar-year and lifetime figures run to 12 August 2026.

The cadence

Unique source documents accepted for processing, over the twelve months to 31 July 2026. Each figure is the same population expressed at a different granularity — they are one number, divided four ways, not four independent measurements.

Daily ~1,000 documents / day
Weekly ~7,100 documents / week
Monthly ~30,800 documents / month
Annually ~370,000 documents / year

At an average of 4.3 pages per document, that is roughly 1.6 million pages a year. Since the platform went live in January 2020 it has processed 2,226,055 documents in total — the calendar-year totals below sum to exactly that. The page count is the one soft number on this page; see how these figures are counted.

Monthly volume over two years

Unique source documents per month. The reporting period is shown in gold; the preceding twelve months are in grey for context.

Unique source documents per month, August 2024 to July 2026 A column chart of 24 months. The twelve months to 31 July 2026 sit between 23,328 and 34,446 documents, except February 2026 at 50,328. The preceding year is shown in grey and opens at 60,922 in August 2024. Every value is listed in the table below. 0 20k 40k 60k Aug 24 — 60,922 documents · Prior year Sep 24 — 46,531 documents · Prior year Oct 24 — 30,248 documents · Prior year Nov 24 — 24,865 documents · Prior year Dec 24 — 22,682 documents · Prior year Jan 25 — 29,342 documents · Prior year Feb 25 — 29,355 documents · Prior year Mar 25 — 29,653 documents · Prior year Apr 25 — 33,062 documents · Prior year May 25 — 36,607 documents · Prior year Jun 25 — 28,940 documents · Prior year Jul 25 — 29,645 documents · Prior year Aug 25 — 33,049 documents · Reporting period Sep 25 — 27,999 documents · Reporting period Oct 25 — 32,279 documents · Reporting period Nov 25 — 25,688 documents · Reporting period Dec 25 — 23,328 documents · Reporting period Jan 26 — 32,581 documents · Reporting period Feb 26 — 50,328 documents · Surge month Mar 26 — 31,749 documents · Reporting period Apr 26 — 24,730 documents · Reporting period May 26 — 24,240 documents · Reporting period Jun 26 — 31,427 documents · Reporting period Jul 26 — 34,446 documents · Reporting period Aug 24 Nov 24 Feb 25 May 25 Aug 25 Nov 25 Feb 26 May 26 Jul 26 50,328 34,446
12 months to 31 Jul 2026 Prior year (context) Surge month

Only three of the twenty-four months clear 40,000 documents: August 2024 at 60,922, February 2026 at 50,328, and September 2024 at 46,531. Two of the three sit in the prior year, which is why the reporting period reads flatter than the one before it. Strip February out and the reporting period runs between 23,328 and 34,446 documents a month — a band of about eleven thousand, which is what the ~30,800 monthly average is actually describing.

Read the two years together and the picture is a platform past its growth phase rather than one holding level. The twelve months to July 2026 total 371,844 documents against 401,852 for the year before — down about 7%. What changed is the spread as much as the volume: the reporting period runs in a tighter band, with occasional spikes on top. Both matter more than the average for anyone planning against it.

Two things worth knowing about these numbers

Plan for the peak, not the average

A typical business day runs 1,230 documents. But the busiest 5% of days exceed 2,142, and the single heaviest day in the period handled 4,670 — more than four times the daily average. February 2026 alone carried 50,328 documents, about 60% above a normal month. Volume is bursty; capacity sized to the mean is capacity that fails on the days that count.

There is a larger figure, measuring something else

A single document can be routed through more than one classification stream, so the platform records 536,000 processing events a year against those 370,000 documents. We quote unique documents, because that is what "documents processed" normally means to the person asking. The higher number is real, but it describes total pipeline throughput, and the two are not interchangeable.

The longer view

Unique source documents by calendar year, from the platform's first day of operation.

YearDocumentsNote
2020136,305From 13 January
2021192,316
2022303,759
2023440,184
2024552,648Highest year to date
2025357,724
2026243,119To 12 Aug · tracking to ~395,000

Four years of compounding growth to a peak of 552,648 in 2024, then a step down to 357,724 in 2025. 2026 is tracking to about 395,000 — ahead of last year, still well short of 2024. Work is spread across 38 active processing streams, with the three largest accounting for just over half of all volume.

How these figures are counted

The definitions below are the load-bearing part of this page. Any of these numbers can be made to look different by changing one of them, so they are stated rather than assumed.

  • Source. The platform's document intake ledger, one record per document received. This is a complete population back to 13 January 2020 — not a sample and not an estimate.
  • What counts as a document. One source file accepted for processing. Where a document is routed through several classification streams it is counted once. That is the difference between the 370,000 figure and the 536,000 processing events described above.
  • Pages are extrapolated, and are the one soft number here. Page counts are recorded on about 35% of items — 823,376 pages measured directly. The annual page total is extrapolated from the 4.3-page average and should be treated as indicative, not counted.
  • Windows. The reporting period is the twelve months ending 31 July 2026. Calendar-year and lifetime figures run to 12 August 2026. Daily and weekly rates are derived from that same twelve-month population.
Why the daily figures differ

The ~1,000/day headline divides the annual total across all days. The 1,230 figure in the burstiness note is the typical business day, which is the number worth planning against. Both are correct; they answer different questions, and quoting either one without saying which it is would be the error this page is trying to avoid.

Unique documents by month

The full series behind the chart, for anyone who would rather read the values than the columns.

MonthDocumentsPeriod
Aug 2460,922Prior year
Sep 2446,531Prior year
Oct 2430,248Prior year
Nov 2424,865Prior year
Dec 2422,682Prior year
Jan 2529,342Prior year
Feb 2529,355Prior year
Mar 2529,653Prior year
Apr 2533,062Prior year
May 2536,607Prior year
Jun 2528,940Prior year
Jul 2529,645Prior year
Aug 2533,049Reporting period
Sep 2527,999Reporting period
Oct 2532,279Reporting period
Nov 2525,688Reporting period
Dec 2523,328Reporting period
Jan 2632,581Reporting period
Feb 2650,328Surge month
Mar 2631,749Reporting period
Apr 2624,730Reporting period
May 2624,240Reporting period
Jun 2631,427Reporting period
Jul 2634,446Reporting period

Changelog

First published, from the platform throughput measurement taken on 12 August 2026. Covers the twelve months to 31 July 2026, with calendar-year and lifetime figures to 12 August 2026.

Volume like this, or planning for it?

hello@lesofi.co