Clear books. An unbroken trail.
An accounting practice is not short of software. It is short of the middle: documents arriving in pieces, keying by hand, and three months later nobody can trace a ledger line back to the invoice behind it. AI Global Hub closes that gap.
- Documents in1
- Read and classify2
- Reviewer3
- Reconciliation4
- Close5
- Close packet6
The stage bar, as it appears in the product
Where a practice loses its margin
Not on the difficult technical work. It goes on repetitive volume that cannot be billed at a higher rate.
Documents from every direction
Email, photos through chat, scans, paper. The first job each month is collection.
Keying eats the hours
Qualified people typing numbers off images. The lowest-billing work takes the most time.
Ledger lines lose their origin
Three months on, “where did this line come from” costs half a morning.
Filing season arrives all at once
Volume triples, the team does not. Review quality is the first thing sacrificed.
From documents in to books closed
Pick a step to see what the software does, where data comes from, and what has to be true before an entry is posted.
0106Stage 1 of 6
Documents in
- What it does
- Takes documents by upload or from a connected inbox, files them under the right client, and lets you ask the client for whatever is still missing.
- What stops it
- A corrupt or unreadable file goes into a manual queue rather than being skipped quietly.
- What it leaves behind
- The original is kept, with its source and time of arrival.
Where data comes from
- Which client Sender or the assigned folder
- Document type Classified from the content
- State New, in progress, or done
A reference table, not buttons.
Start a pilot at this stageRead and classify
- What it does
- Reads the text off the scan, pulls the fields out, and classifies it. Whatever the model is unsure about is flagged for a person.
- What stops it
- Reading runs under the workspace spend cap. At the cap the system says plainly that it is degraded or blocked rather than quietly returning worse results.
- What it leaves behind
- Each field keeps a link to the document it was read from.
Where data comes from
- Supplier, number, date Text read from the document
- Amounts and tax The table and totals area
- Confidence Model score, per field
A reference table, not buttons.
Start a pilot at this stageReviewer Needs a reviewer
- What it does
- Pushes the uncertain items to the top of the queue so a reviewer spends time where human eyes are needed, instead of rereading everything.
- What stops it
- A hard stop. A field below the threshold does not post on its own, it waits for confirmation.
- What it leaves behind
- Who corrected which field, from what to what, and when.
Where data comes from
- Review items The confidence threshold you set
- Reviewer Assigned per client
- Review notes Written by the reviewer
A reference table, not buttons.
Start a pilot at this stageReconciliation
- What it does
- Pairs bank transactions with documents, batches the confident matches for one approval, and separates out what needs a decision.
- What stops it
- An unmatched item stays in the open list. It is not swept into a suspense account to tidy the screen.
- What it leaves behind
- Each match links back to the document and the statement line behind it.
Where data comes from
- Transactions The imported bank statement
- Matching document Reviewed documents
- Matching rules Rules you set per client
A reference table, not buttons.
Start a pilot at this stageClose Needs a reviewer
- What it does
- Runs the pre-close checks, lists what is still open, then closes the period once that list is clear.
- What stops it
- A hard stop. Open items block the close.
- What it leaves behind
- The closed period is frozen with who closed it and when.
Where data comes from
- Open items The review and reconciliation queues
- Check results Run against the period's data
- Who closed it Workspace role
A reference table, not buttons.
Start a pilot at this stageClose packet
- What it does
- Builds the packet for the closed period, with the reports and the links back to source documents, ready to send to the client or hand to an auditor.
- What stops it
- A packet can only be built from a closed period.
- What it leaves behind
- From a line on the report back to the exact document.
Where data comes from
- Period reports The reviewed ledger
- Evidence links Every step above
A reference table, not buttons.
Start a pilot at this stageFrom a ledger line back to the invoice
This is what separates a text reading tool from a system an auditor can work with. Each field keeps its link to the document it was read from, so “where did this number come from” is one click away, even years later.
Select an entry to highlight it on the document
the region this value was read from
Illustration, sample data
What we guarantee
Each client stays in their own house
Each client sits inside its own boundary, checked in the data layer, not only in the interface.
Unsure goes to a person
A field below the threshold does not post before someone confirms it.
Document reading has a ceiling
Capped per workspace, with the degraded state shown rather than hidden.
Every change leaves a mark
Adjustments create new records, history is not overwritten.
Four things practices ask
What happens when it reads something wrong?+
That is why confidence is recorded per field rather than per document. Anything below the threshold goes to the review queue and does not post. The goal is not perfect reading, it is knowing exactly where human eyes are needed.
We serve many clients. Can their data mix?+
Each client sits in its own boundary and that boundary is checked in the data layer, not only in the interface. A query inside one engagement cannot see another client's records.
Can we control the document reading spend?+
It is capped per workspace. At the cap the paid call is refused before it runs and the state is shown on screen. It does not quietly return worse results and leave you guessing.
Does this replace our accountants?+
No, and it was not built to. It takes the mechanical keying and matching, the lowest-billing work, so qualified people spend their hours on review and advice.
Another industry, same way of working
All three share the same principles: data knows its source, anything going out needs a person to approve it, and every run can be replayed. If your team works across more than one, it all lives in one place.
Start with one client, one period
Pick the client with the heaviest document volume. Run one period alongside your current process, then compare the real hours.