This is an illustrative walkthrough, not a description of a specific client. Every firm's tools and procedures differ; treat this as a pattern to adapt, not a blueprint.
Ask anyone who works in an accounting or bookkeeping firm where the time goes during busy season, and document intake will come up quickly. Clients send documents by email, through a portal, as phone photos, and occasionally on paper. Files arrive with names like "scan0042.pdf". Some are missing. Some arrive twice.
Very little of this work needs an accountant's judgment — but it still takes an accountant's (or an administrator's) time. That makes it a good candidate for automation.
The process today
In a typical small firm, intake looks something like this:
- A document arrives by email or portal upload.
- Someone opens it to work out what it is and which client it belongs to.
- They rename it according to the firm's naming convention.
- They file it in the right client folder for the right year.
- They check it off against the client's list of expected documents.
- Periodically, someone reviews each client's list and emails reminders for anything still missing.
Every step is simple. Together, repeated across many clients, they add up to hours of careful, repetitive work — and small errors (a misfiled document, a missed reminder) that cause delays later.
The automated version, step by step
Here's how the same process might look with automation and AI working together.
Step 1 — Collect. Documents from the intake inbox and portal flow into a single processing queue automatically. Nobody needs to forward or download anything.
Step 2 — Identify. An AI step reads each document and suggests what it is (for example, a wage statement, a bank statement, or a receipt), which client it belongs to, and which tax year it relates to. This is where AI earns its place: documents don't arrive in a consistent format, and reading them is exactly the kind of variable task rules struggle with.
Step 3 — Check confidence. If the AI is confident and the details match a known client, the document moves on. If it's unsure — an unfamiliar format, a name that matches two clients, a blurry photo — it goes to a person's review queue instead. This is the most important step in the whole design.
Step 4 — Name and file. Rules take over again: rename the file using the firm's convention and save it to the correct client folder. No AI is needed here; a simple rule is more predictable.
Step 5 — Update the checklist. The client's list of expected documents is updated automatically, so anyone can see what has arrived and what is still outstanding.
Step 6 — Remind. On a schedule the firm chooses, clients with outstanding items receive a polite reminder listing exactly what's still needed. Staff can review reminders before they go out, or let routine ones send automatically — the firm decides.
Where AI helps, and where rules are enough
A useful principle: use AI for reading and interpreting; use rules for everything predictable.
In this workflow, AI is only needed for step 2 — working out what an unstructured document is. Collecting files, renaming them, filing them, updating a checklist, and sending scheduled reminders are all well-handled by conventional automation.
Keeping AI to the steps that need it makes the whole system cheaper to run, easier to test, and easier to explain.
Where people stay in the loop
Automation changes what people spend time on, not whether they're involved. In this design, people:
- Review every document the AI isn't confident about
- Spot-check a sample of automatically filed documents, especially in the first weeks
- Decide which reminders send automatically and which need approval
- Own every professional judgment that follows: review, preparation, advice, and sign-off
The firm also decides which systems the automation can access and keeps a record of what it did with each document.
A note on client information
Client financial documents are sensitive. Before building anything like this, a firm should confirm that each tool involved is appropriate for that information — including where data is stored, who can access it, and whether the provider uses it for training. If those answers aren't satisfactory, change the tool or narrow the design.
What to measure yourself
We won't promise a particular result; every firm is different. But it's worth measuring your own baseline before you change anything, so you can judge the improvement honestly:
- How long does intake take per client, roughly, today?
- How many documents are misfiled or need to be re-requested?
- How many reminder emails does the team send by hand each week?
- How often is a return delayed waiting for a missing document?
Measure the same things a few weeks after the change. If the numbers haven't moved, adjust the design — or stop.
Getting started
Document intake is often a good first project precisely because it's well-defined and easy to check. If you'd like to explore it for your firm, see our Workflow Automation service, or read more about how we work with accounting firms.
