Choose the flow that fits your workflow.
Drop in. Tidy up. Done.
Select Upload
Upload your files
Select files
Choose the files you want to extract data from
Select document type
Tell Harold what type of document it is
Click 'Make messy data tidy'
Review & Export!
Check your data and export with confidence
Set it up once. Let it flow.
Setup a new inbox
Create your inbox
Select document type
Tell Harold what to expect
Name the inbox
Give it a name
Add approved senders
Who can send to this inbox
Send introductory email
We'll help you write it
Use the inbox!
Data flows in, is extracted & you can review
Set it up once, sit back and watch Harold do his thing. ♡
Where Harold earns his keep.
Three layers of power. Each layer builds on the one before it. Set up the bottom layer well and the rest gets easier.
Define which columns Harold extracts and the labels he uses for them.
One template per document shape. Switch the type and Harold offers to swap to that type's defaults; if you've customised, he asks first.
Tell Harold what each supplier calls things. He then knows their dialect forever.
Suppliers label invoice fields differently. Supplier Profiles map each supplier’s labels to your standard column names — once.
The next time Harold sees that supplier, the mapping is silent and automatic.
Show Harold a sample document. Drag the fields to your columns. He remembers forever.
Use this when AI extraction alone isn’t cutting it for a specific supplier or document layout — typically the worst 5% of your suppliers.
Translate values from what suppliers write to what your systems expect.
Pick a source column. Define when X then Y rules. Harold writes the matched value into a target column on every row.
Three match modes
Auto-assign accounting codes by supplier or category. Stop coding invoices by hand.
Build rules that map a value to a GL code and a GL name. Harold writes both into the row, ready to push into Xero, Sage, QuickBooks, or your ERP.
Set a default action for unmatched rows — usually “flag for review” so nothing slips through uncoded.
Compute new columns from existing ones. Fill in fields suppliers leave blank.
Name an output column. Pick two operands — each can be a fieldor a typed-in value. Pick an operator. Harold computes and writes the result on every row.
Six operators
Currency symbols (£, $, GBP) are stripped before maths. Numeric results round to 2 decimals. Computed columns can themselves be referenced by other rules — so you can chain.
Catch bad data on its way through. Compare any two columns or a column to a value.
Give the rule a label (e.g. “Total mismatch”), pick a column, choose an operator, and compare to either another column or a fixed value.
Six comparisons
Three actions when triggered
Severity (Error or Warning) is tracked separately so you can have a red flag that’s a warning, or an amber flag that’s an error. Numbers are compared first (currency-stripped); strings fall back to lowercase trim.
The enforcement layer. Promote any IF Rule to a gate — failed checks stop the document from firing the Zap.
Every IF Rule has a Zap Gate toggle. Flip it on, and that rule no longer just flags — it blocks. The document stays in Harold’s review queue until you fix it manually.
✓ All gates pass
Document fires to Zapier automatically
✗ Any gate fails
Held for human review — never reaches downstream systems
Gates that earn their keep
Every blocked document records which gates fired and why, so you get a full audit trail of what Harold caught before it left the building.