One complete request, a place to put the answer, and reminders that stop when the item arrives. The automation is easy; the discipline of asking once is not.
Key takeaways
Build one complete request per engagement, deliver it somewhere the client can act on it, let the system file what arrives against the item it satisfies, and let reminders be driven by receipt status rather than by a calendar. That is the whole design. Almost every firm that describes document chasing as a nightmare is failing at one of those four points, and usually the first: the request goes out incomplete, so the client is asked three more times over the following month and eventually stops opening the emails.
AI has a real but bounded role. It builds the expected list for the engagement type, reads what arrives and works out which request it answers, spots that a document is unreadable or for the wrong period, and drafts the chase in a tone that is neither robotic nor apologetic. What it should not do is decide what you need — that comes from your engagement scope.
The request list is a firm asset, written once per engagement type and maintained. It should name each item in the client’s language rather than yours — ‘your December bank statement for the operating account’, not ‘year-end bank confirmation’ — state the period, say what format is acceptable, and mark whether the item is required or conditional.
Two things improve response rates more than any amount of automation. The first is completeness, which removes the second round-trip. The second is explaining why an unusual item is needed. A client who understands that the loan agreement is needed because the interest split affects the statements will send it; a client who receives the word ‘loan agreement’ on a list of nineteen items often will not.
There is a discipline point here too. Ask for what the engagement requires and no more. The Office of the Privacy Commissioner’s principles for responsible, trustworthy and privacy-protective generative AI press organisations to establish that their use of personal information is necessary and proportionate, to prefer de-identified data where personal information is not required, and to be open about what is collected and why — principles that apply to the firm using a tool, not only to the company that built it. A request list that hoovers up everything because it might be useful is both a privacy problem and the reason your client thinks you are disorganised.
Email is a poor container for a document request because nothing in it tracks state. A portal turns the same list into a set of items with a status. Karbon describes its document management as requesting and collecting documents in the client portal with auto-created job folders and an audit trail on every attachment, and its client portal as offering personalised task lists, secure file sharing and automated reminders that nudge clients when action is needed, with uploads linked back to the work item. On the legal side, Clio’s client portal software is described as a secure place for clients to log in and exchange documents rather than send them by email.
The specific product matters less than the property: a request that knows whether it has been satisfied. Without that, your reminders are guesses, and a reminder for a document the client already sent is the fastest way to lose their attention.
Three rules. Reminders are per outstanding item, not per client — a client who has sent nine of eleven documents should be asked for two, not for the list again. They stop automatically on receipt, which requires the receipt-matching step above. And they escalate through channels rather than through frequency: a second automated reminder is fine, a fourth is noise, and the right escalation after that is a person on a telephone.
One caution on content. A reminder about an engagement you are already performing is service correspondence. If you attach a promotion for another service to it, you have changed what the message is, and Canada’s anti-spam regime becomes relevant — the CRTC’s spam and malware pages carry the consent, identification and unsubscribe requirements, and our sister firm’s note on implied versus express consent under CASL explains the consent distinction in business terms. Keep chasers clean and put the marketing somewhere else.
Illustrative. A practice serving several hundred personal tax clients historically sent a generic checklist in February and then spent March chasing individuals by email, with no reliable way of knowing who had sent what.
The rebuilt process starts in January with a per-client list generated from last year’s return rather than from a generic template — if a client had rental income last year, the rental documents are on their list and the irrelevant items are not. The list goes to the portal with a short covering note. Arrivals are matched to items automatically, with anything ambiguous queued for a person to classify. Reminders run per outstanding item on a fixed cadence and stop on receipt. A weekly exception report shows clients who have sent nothing at all, which is the group that gets a phone call rather than another email.
The change in workload is concentrated in one place: staff stop reading inboxes to work out who is outstanding. That is the task automation actually removes. The client conversations that remain are the ones worth having.
Clients frequently ask why you want documents they have already given someone, or how long they must keep them. The statutory anchor is section 230 of the Income Tax Act, which requires every person carrying on business to keep records and books of account, and subsection (4), which requires them — in the ordinary case — to be retained until six years from the end of the last taxation year to which they relate. Our sister firm’s answer on what records to keep in case of a CRA audit is a good plain-language explanation to forward.
Your firm’s own retention of what clients upload is a separate decision, driven by your professional obligations, your insurer and your privacy commitments. Make it explicitly, write it into the engagement terms, and make sure the portal actually implements it rather than accumulating a decade of client records because nobody chose otherwise.
Can AI read what the client sends and tell us if it is wrong?
It can classify documents, check the period, spot an unreadable scan and flag an obvious mismatch. Whether the content is correct is a professional review, and it stays one.
Should we automate requests for every engagement type?
Start with the highest-volume, most repetitive one. The list is the work, and writing four good lists beats writing forty mediocre ones.
Is this the same as chasing missing receipts in a client’s bookkeeping?
No. Transaction-level chasing inside a client’s ledger is a bookkeeping function and it lives on the bookkeeping automation page. This is about collecting what an engagement needs before the work begins.
If the wider problem is checking the file is complete once the documents arrive, see keeping records and compliance tidy without doing it yourself.
A 30-minute call is enough to tell you whether AI pays for itself here.
