Client document collection becomes expensive when accountants repeatedly inspect inboxes, rename files, work out what is still missing and send reminders. AI can help prepare an organised evidence pack, but the practice should first use the client-request and portal features already available in its software.
This guide proposes a workflow for accounting, bookkeeping and tax practices. It is not legal, tax or professional-standards advice, and it does not claim Binarify client results. Each firm must define the engagement, evidence requirements, privacy obligations and professional review that apply in its jurisdictions. See our AI consultancy for accounting firms service page for the wider implementation approach.
Define the job around an engagement
“Collect the client’s documents” is not a stable specification. The required evidence depends on the service, entity, period, accounting framework, tax jurisdiction and events during the year.
A bounded first outcome could be:
Prepare a review-ready client file for one recurring service, with every requested item marked received, incomplete, superseded, not applicable or awaiting clarification, and each extracted fact linked to its original document.
The request list must come from an approved template and engagement context. AI may interpret a document and propose a status. It should not invent a requirement, decide that evidence is sufficient for a tax position or silently reuse a previous-year answer.
Existing products may already solve much of the chasing. Karbon Client Requests supports task lists, due dates, document uploads and scheduled reminders linked to a work item. TaxDome file requests can create upload links for clients or third parties. Configure those capabilities before building another reminder system.
A controlled document-collection workflow
1. Create the engagement evidence register
Start with the client, legal entity, service, reporting period, jurisdiction, deadline and responsible team. Generate the initial request list from the firm’s approved template, then let the accountant adjust it for the engagement.
Each request needs a stable ID, plain-language description, acceptable evidence types, period, owner, due date and secure submission route. Keep “not applicable” as a reviewed status rather than treating no file as completion.
2. Send requests through the approved channel
Use the practice platform or secure portal when it already supports the workflow. The integration may create tasks, insert engagement-specific wording and schedule reminders within the platform’s controls.
Avoid sending confidential attachments through an unapproved channel. Confirm recipient identity, access expiry, revocation and the behaviour when several client contacts share responsibility.
3. Attach every submission to the correct request
Record the original filename, upload time, submitting person, channel, file hash and request ID. Scan files under the firm’s security process before extraction.
The system can propose a document class such as bank statement, sales report, payroll summary, loan statement, invoice, prior return or identification evidence. Allow unknown, multiple-document and wrong-period states. Similar names are insufficient evidence that a file belongs to the correct entity.
4. Check completeness without deciding the accounting treatment
Use deterministic checks for page counts, dates, statement periods, expected account identifiers and duplicate files. AI can help interpret layouts and compare the document with the request description.
Show exceptions such as:
- missing pages or a truncated export;
- period outside the engagement window;
- document for another entity or account;
- password-protected or unreadable file;
- totals or identifiers that conflict with another source; and
- a narrative answer that does not address the question.
These are preparation findings. An accountant decides whether the evidence is adequate and what additional work is required.
5. Ask only for unresolved evidence
Prepare a concise follow-up that identifies the exact request, what was received, what remains unresolved and how to respond securely. Use approved wording and let staff review sensitive or unusual requests.
Stop reminders when the item is completed, marked not applicable, reassigned or the engagement closes. Group requests where possible so the client does not receive several disconnected messages from different workflows.
6. Prepare the accountant’s review pack
The reviewer should see:
- the engagement and applicable request template;
- every request and current status;
- original submissions and version history;
- proposed document classes and extracted fields;
- completeness checks and conflicts;
- client explanations and previous follow-ups; and
- unresolved questions, deadline and owner.
The accountant confirms completeness, records exceptions and approves the transition to bookkeeping, accounts preparation or tax work. The collection component should not post transactions, create a tax position or mark professional review complete.
Binarify evidence-request register
This proposed design artifact connects the client request to the evidence, reviewer and next action.
| Request | Required period or scope | Evidence received | Automated checks | Accountant disposition | Next action |
|---|---|---|---|---|---|
| Bank statements | Account and date range | Original files and versions | Pages, dates, account suffix, duplicate | Complete, incomplete or clarification | Accept or request exact gap |
| Payroll summary | Entity and pay periods | Export or report | Period coverage and total fields | Suitable for engagement purpose | Accept or refer to payroll owner |
| Sales records | Agreed source and period | Platform export or ledger | Schema, period and reconciliation status | Source and completeness confirmed | Proceed or request replacement |
| Asset purchase | Invoice, payment and business explanation | Files and client response | Identifier, date, amount and attachment link | Treatment requires review | Send to responsible accountant |
| Tax question | Jurisdiction-specific prompt | Client declaration and evidence | Required fields and conflicting answers | Position and evidence assessed | Clarify or complete review |
Store each status change and who made it. A later upload should create a new version rather than erasing the file on which an earlier conclusion was based.
Confidentiality and permitted use
The ACCA’s 2026 discussion of AI adoption risks highlights governance, confidentiality, training, professional judgement and the need to check AI output. Firms should know which tools staff use, what data enters them, whether suppliers train on that data, where it is processed and how it is deleted.
For US tax preparation, Internal Revenue Code section 7216 rules and consent guidance govern specified uses and disclosures of tax-return information by preparers. A software contract or general AI policy does not itself establish that a particular disclosure is permitted.
Map the rule by market and engagement. Record the lawful or contractual basis, client disclosure or consent where required, access restrictions, retention and incident route before using live client documents.
Measure the whole collection cycle
Take a baseline by service and client segment. Track:
- days from first request to review-ready file;
- staff minutes spent requesting, downloading, renaming and filing;
- requests completed correctly on the first submission;
- reminder count and repeat questions per engagement;
- documents attached to the wrong client, entity, period or request;
- completeness errors found after accountant review begins;
- client abandonment or use of an assisted route; and
- security, access and retention exceptions.
Do not optimise reminder volume. The target is a complete, correctly owned evidence set with less effort for the client and practice.
Pilot one recurring service
Choose one service with a stable request template and enough volume to compare, such as monthly bookkeeping, annual accounts preparation or one bounded tax workflow. Start in shadow mode and compare the proposed register with the current accountant-reviewed file.
Agree the supported file types, request states, portal integration, review rules, client escalation, quality sample, outage process and rollback owner before assisted production.
Continue with AI bookkeeping exception and month-end review once the evidence reaches the ledger workflow. Use the accounting-firm AI ROI guide to test whether released capacity becomes additional profitable work or an evidenced cost change. You can also book a 30-minute conversation about one recurring client-document queue.