What are Automations?
Automations is Harold's email ingestion layer. Instead of uploading documents manually, you configure a dedicated Harold email address for each document type. Your suppliers email their invoices, purchase orders, or statements directly to that address. Harold picks them up, runs extraction, and the data is ready in your inbox — without anyone having to touch a file.
Your company prefix
Before you can use Automations, you claim a company prefix. This is a short identifier — usually a slug version of your company name — that becomes the subdomain for all your Harold inboxes. For example, if your prefix is "acme", your purchase invoice inbox address becomes acme.purchase-invoices@hex.useharold.com.
You set this once and it never changes. All inbox addresses are built on top of it.
What is a Custom Inbox?
A Custom Inbox is a dedicated email address tied to a specific trained document type. You create one per document type — one for purchase invoices, one for sales invoices, one for delivery notes, and so on.
Each inbox has a unique email address your suppliers send to, a linked DocuTrain schema that defines what fields to extract, a list of approved senders (optional but recommended), and a status for every document that arrives — Processing, In Review, Approved, or Failed.
The approved senders list
Each inbox can have a list of approved sender email addresses. When a document arrives from an approved sender, it goes straight into processing. When it arrives from an unknown sender, it is held for review.
This matters in practice. Your purchase invoices inbox might accept documents from five different supplier email addresses. Harold learns which suppliers send to which inbox and routes accordingly.
What happens when a document arrives
Harold receives the email and extracts any PDF or image attachments. The attachment is matched against the inbox's trained schema. Harold runs extraction — field by field — and records a confidence score for each value. The document appears in the inbox with status Processing and transitions to In Review once extraction completes. You review the extracted data, correct anything low-confidence, and approve. Approved documents are available for export via Zapier or CSV.
The inbox document list
Inside each Custom Inbox you get a filterable list of every document that has arrived. You can filter by sender, by status, and by date. You can select multiple documents at once and batch-approve or batch-export them.
Each document shows the extracted supplier name, the document date, the status, and the confidence level of the extraction. Clicking a document opens a split view — the original PDF on the left, the extracted data fields on the right — so you can verify extractions quickly without switching between applications.
Ease of use
Once set up, Automations requires almost no ongoing effort. The initial setup — claiming your prefix, creating an inbox, linking it to a trained schema — takes around five minutes per document type. After that, documents arrive and are processed automatically.
The main task for operators is periodic review of the In Review queue: checking low-confidence extractions, correcting occasional mistakes, and approving batches. For a business processing 50 invoices a week, this review process typically takes 10–15 minutes.
Limitations
Automations requires a trained DocuTrain schema before it can extract anything. An inbox without a linked schema will receive documents but cannot process them.
Email attachments must be PDF or image files. Inline HTML emails with no attachment are not currently supported.
The email addresses are fixed once created. You cannot change the inbox address after creation, though you can create a new inbox and redirect your suppliers.
The aim
Automations turns Harold from a tool you use into a system that works for you. The goal is a world where a supplier sends an invoice to your Harold inbox address and that invoice — fully extracted, validated, and correctly coded — is sitting in your ERP system by the time you open it. No manual entry. No copy-paste. No missed fields.