Start with one client filing register
Use one row per company, document and required date. Include your client code, available PDFs, missing filings and reviewer. Keep different years on separate rows instead of deriving the year from a filename or download date.
Agree a pilot before accepting the full batch
Pilot readable scans, difficult scans and filings with continuation sheets. Agree dates, null values, share classes and client codes. Reconcile each row as source received, extracted, awaiting pages or awaiting review; a total file count alone does not establish completeness.
A batch register: track gaps and assign review
The three rows below are entirely fictional worksheet examples, not real clients, search results or delivery records. Each row represents one company, one document date and one filing.
Scroll sideways to see all columns; focus the table and use arrow keys on a keyboard.
| Client / company reference | Document type | Required return date | Received / expected pages | Status and next step | Review owner |
|---|---|---|---|---|---|
| DEMO-01 / Fictional Company A | NAR1 | 2024-08-01 | 6 / 6 | Extracted; check identifiers, dates and capital | Reviewer A |
| DEMO-02 / Fictional Company B | NAR1 | 2024-09-15 | 2 / to confirm | Excerpt only; request member schedules before acceptance | Reviewer B |
| DEMO-03 / Fictional Company C | NAR1 | 2024-11-20 | 8 / 8 | Date mismatch; return for correction and recheck | Reviewer A |
Check each row before acceptance
First reconcile the client code to the company, requested filing and record date. Then check whether the source is complete or an excerpt. Finally, confirm that extracted fields have page references and unresolved issues. A complete page count does not prove field accuracy.
Worked batch: reconcile requests to files, not just file counts
Entirely fictional: A, B, C and D each require two annual returns, giving eight requests but seven PDFs. A’s two filings share one PDF. One of B’s filings has a duplicate copy. Both of C’s filings and D’s first are present; D’s second is absent. The count alone does not identify the missing request.
Assign a reference to each requested filing, then map the actual file and page range. A has two rows pointing to different ranges in one PDF. B’s duplicate remains a copy, not an extra completed request. D’s missing request stays visible as its own row. Reconcile requests rather than celebrating an extra file in the folder.
Next check continuation pages and record dates. A received file may still be unsuitable: it could be complete but for the wrong year. Separate source-material status from extraction and review status so retrieval, extraction and checking are not represented by one checkbox.
The inventory has seven mapped requests and one missing, out of eight. Mapped does not mean accepted. If R01–R07 later pass source and field checks, record seven accepted, one awaiting retrieval and zero partially accepted: still eight in total. D’s 2024 filing updates R08, not an unrelated row.
Scroll sideways to see all columns; focus the table and use arrow keys on a keyboard.
| Request (fictional) | Company / year | File mapping | Inventory result |
|---|---|---|---|
| R01 | A / 2023 | A-two-years.pdf, first range | Mapped |
| R02 | A / 2024 | A-two-years.pdf, second range | Mapped |
| R03 | B / 2023 | B-2023.pdf | Mapped |
| R04 | B / 2024 | B-2024.pdf plus a duplicate copy | One request; keep the duplicate link |
| R05 | C / 2023 | C-2023.pdf | Mapped |
| R06 | C / 2024 | C-2024.pdf | Mapped |
| R07 | D / 2023 | D-2023.pdf | Mapped |
| R08 | D / 2024 | Not received | Missing; an unrelated new PDF does not close it |
Choose a pilot that exposes differences in the batch
Inspect the actual differences before choosing pilot files: readable text, scans, long filings with continuations or multiple share classes. This is not a prescribed sampling rate, and passing a pilot does not prove the whole batch correct. It helps settle field definitions and visible format issues before scale-up.
Keep identifiers as text, separate record and retrieval dates, agree missing-value statuses and decide how multiple capital rows map. Test an internal client code through the full delivery path so the result joins back to the correct client rather than relying on a name match.
Save the agreed field list and sample version. Treat added director fields, date-format changes or XML restructuring as a change request. Identify which delivered files need revision and which future files use the new version. Otherwise individually readable files may still be impossible to combine.
Route exceptions to the person who can resolve them
Missing originals, continuations, wrong dates and unreadable text need different actions. Attach each issue to its request row and give it an owner and completion condition. “Obtain page 4 and check for further continuation” is actionable; “look at this file again” is not.
Keep unresolved rows open even when other items are accepted. The owner decides whether partial delivery is usable. Reconcile requested filings, received originals, extracted outputs and reviewed items separately. A correction should identify the version it replaces so outdated data is not imported again.
Close with a request-indexed acceptance register showing delivery location, record date, checker, outcome and outstanding items. It supports collection and handover. Field checks do not complete audit procedures or prepare or submit annual returns; the relevant owner arranges those professional tasks.
Frequently asked questions
Does the batch service file annual returns?
This service retrieves and organises filed documents. It does not prepare or submit annual returns.