The Document-Chasing Agent | DigiKat

Every professional services firm has a staff member nobody hired. It's the one who writes "just following up on the below" twenty times a day. In accounting, law, conveyancing, broking and advice firms, that work is spread across everyone, and it quietly eats the best hours of your best people. It's also the single best first job for an AI agent.

I'll explain why document chasing is where we usually start, what the agent needs to know to do it well, the chase sequence we use, and the lines it never crosses.

Why chasing is the perfect first job

When we run the Context step with a firm (check the data, pick one job, agree what done means, measure today's cost), document chasing nearly always wins. Here's why:

• It's rules-based. A file needs a known list of documents. Either they're in or they're not.
• It's measurable. You can count requests sent, documents received and days a file waited.
• It's low risk. A reminder email is about as safe as client contact gets, especially when a person approves it first.
• It's painful. Nobody went to university to chase bank statements.

We wrote about this in our post on AI agents for accounting firms, but it applies just as well to a law firm waiting on signed authorities or a broker waiting on a claims history.

What the agent needs to know

An agent can't chase what it can't see. This is where most "AI automation" falls over, and it's the reason most AI agents fail because of the data, not the model. If the document list lives in a staff member's head, the agent has nothing to follow.

So before we switch anything on, we set up HubSpot so every file carries its own checklist:

• The client record: contact and company, with the right people associated (the client, their partner, their bookkeeper, the other director).
• The service or matter: what you're doing for them, for example an annual return, a property purchase or a policy renewal. HubSpot's Services object or a custom object works well here.
• The job record: a ticket or pipeline record for this period's work, with a property for each required document and its status: requested, received, accepted.
• A secure way in: an upload form or client portal that attaches files straight to the right record, so the agent can see what arrived.

With that in place, the agent can answer the only question that matters for any file: what's still missing, and who do we need it from?

The chase sequence

Chasing well is mostly about being specific and persistent without being annoying. The sequence we start firms on looks like this (the timings are an example; we tune them per firm):

1. Day 0, the request: one email listing exactly what's needed, why, and a single upload link. No attachments to fill in, no "see below".
2. Day 3, the friendly nudge: names only what's still missing. If two of five items came in, the client sees three, not five.
3. Day 7, a different channel: a short SMS or a second email from the client's usual contact at the firm, still drafted by the agent.
4. Day 10, hand to a person: the agent stops chasing and creates a task for the file owner, with the history attached. Some clients need a phone call, and that's a human job.
5. The moment it's in: the agent stops, thanks the client, updates the record and tells the team the file is ready.

That last step matters more than people think. The most irritating thing a firm can do is chase someone for a document they sent yesterday. Because the agent works from the record, not a calendar reminder, it doesn't.

If this sounds familiar, it's the same pattern we use for quote follow-up agents in the trades: know the state of the record, follow up on what's outstanding, stop the moment it changes.

Tone, permission and the draft-first rule

Your clients already have a relationship with your firm, and the agent has to sound like you. So every agent we build starts by drafting. A team member approves each message before it goes out, and edits it if it's off. That's the Coordination step: people approve, agents act.

An agent only earns the right to send on its own once 95% of its drafts have been approved unchanged for two weeks. Until then, it's a very fast junior who writes everything for you to check. Most firms find that after a fortnight or so of approving the same reminder over and over, they're happy to let it go.

What the document-chasing agent won't do

Clear boundaries make an agent safer and easier to trust. Ours are written down before the build starts:

• It doesn't judge whether a document is right. It can check that something arrived and that it's the expected file type. Whether a statement covers the right period, or a contract is complete, is checked by a person.
• It never asks for sensitive identifiers by email. Tax file numbers, bank details and ID documents go through the secure upload, never a reply.
• It doesn't chase money. Overdue invoices are a different conversation, with a person deciding the approach.
• It doesn't give advice. If a client replies "do I even need this?" or "what does this mean for me?", the agent routes it to the right professional with the context attached.

Measuring whether it's worth it

McKinsey's State of AI report found that while 88% of organisations use AI, only 39% report any EBIT impact. The gap is usually measurement: nobody wrote down what the old way cost.

So every agent gets a monthly scorecard. For a document-chasing agent, that means requests sent, documents received, average days from request to complete file, hours of chasing it took off the team, what it cost to run, and the net return. As an illustration only: if a firm's team spent eight hours a week chasing before, and two after, that's six hours back for real work, every week, measured rather than guessed.

Want to see what this looks like with your own files? Book an AI Readiness Audit and we'll measure what chasing costs you today and map the checklist, or jump to Build my solution if you're ready to get started.