It is late September, and for a lot of accounting firms that means one thing: the October 15 extension deadline is about two weeks away, and a stack of returns is sitting in a folder marked "waiting on client." The preparer is ready. The software is open. What is missing is a 1099, a brokerage statement, a K-1, or a signed engagement letter that the client said they would send "this weekend" three weekends ago.
As Gideon Wafula, AI Automation Engineer, I have been looking at where accounting and bookkeeping practices lose time and revenue, and the pattern is consistent: the bottleneck is rarely the accounting. It is the chase. This post is a teardown of that leak and a build for closing it. It applies to tax season, but also to monthly bookkeeping closes, where the same wait happens twelve times a year.
I am not asking you to take my word for it. Liscio, a client-communication platform for accounting firms, summarizes a Wolters Kluwer survey of nearly 2,000 firms in which "late and unprepared clients" ranks as the profession's number one operational challenge. The same piece notes that the US accounting workforce shrank by 17 percent between 2020 and 2024, which makes hours spent on reminder emails more expensive, not less. And in January 2025 Thomson Reuters acquired SafeSend for a reported 600 million dollars, largely on the strength of document workflow automation. Those are the vendors' and researchers' figures, cited here as reported, not numbers I measured. The point is that the chase is a recognized, priced problem.
It is also a problem a small firm can address without buying an enterprise platform. If you are weighing that choice, my piece on build versus buy for small business AI agents covers how I think about it.
Many firms send a general "please send your tax documents" email. The client does not know what counts, forgets half of it, and replies with a photo of one form. Nobody knows what is still missing until a preparer opens the file. A specific checklist per client, based on last year's return, prevents that.
Follow-ups happen when a staff member has a quiet afternoon. In busy season, that afternoon does not come. The clients who are slowest are usually the ones who need the most reminding, and they get the least.
Documents trickle in over email, a portal, a text photo, and a drive link. Completeness is checked only when the preparer starts, which is the most expensive moment to discover a gap. By then, an easy request has become a rushed one.
Ask most owners how many clients are more than seven days late on documents and they will estimate. There is no board that says who is waiting, for what, and for how long. Without that, the pattern is never fixed year to year. It is the same shape of problem I described in automating invoice chasing: the money or the work is stuck, and the process depends on memory.
Before touching n8n, pull your own numbers from last season. How many clients delivered their last document more than two weeks after your first request? How many extensions were filed mainly because of missing documents rather than complex returns? How many staff hours went to reminder emails and calls? Do not use anyone else's benchmark here. Your own late-client count multiplied by your realistic hourly rate is the number that justifies or kills this project.
When a client's engagement is confirmed, generate a document checklist from a template plus last year's file: employers, brokerages, mortgage, dependents, business income. Store it in a sheet or CRM row with one line per expected document and a status of requested, received, or not applicable.
Send the checklist with a secure upload link, not a request to reply with attachments. Use plain language and a due date that sits a few days before your real internal deadline. Keep sensitive documents off plain email and SMS.
A schedule node checks each client daily. If items are still open at day 5, day 10 and day 14, send a short message listing only what is missing. Cap it at three automated nudges, then route the client to a person for a phone call. Persistent silence is a signal a human should act on.
When a file lands in the upload folder, a webhook updates that line on the checklist. Matching a file to a checklist item can be a simple filename or folder rule to start with. If you add an AI classification step, treat its match as a suggestion that staff confirm, not a fact.
A return or close does not move to "ready for prep" until every required line is received or explicitly waived by a person. This is the step that removes the surprise on the preparer's desk. The workflow tracks arrival, it does not interpret the numbers.
Every morning, post a short digest: clients more than seven days late, which documents, how many nudges have gone out, and who needs a call. This is the board most firms do not have. It also turns next season's planning into a matter of reading last season's data.
Client tax data is sensitive, so this build stays narrow. Use secure links for uploads. Give the workflow access to the checklist and status fields, not the contents of returns. Do not paste document contents into model prompts unless your privacy setup and local rules allow it. Respect email and SMS consent, honor opt-outs, and keep the tone of automated reminders polite and factual. The automation never gives tax advice and never decides what a client owes; a licensed professional reviews everything. For a wider look at limiting what an automation can touch, see my post on AI agent permissions for small business.
A DIY n8n build for a small practice usually runs about 30 to 120 USD per month depending on volume, channels and hosting, on top of whatever practice-management tools you already pay for. It will not fix clients who ignore every message, and it will not replace a good portal if you are already at a size where you need one. What it does is make the wait visible and shorter, and give staff their hours back for the work they are qualified to do.
If you want a practice-specific version built and tested with your own tools, you can see what I offer on my AI automation services page.
Gideon Wafula builds custom AI automation systems, n8n, WhatsApp, Voice AI, and more.
See Services →