Case Study6 min read
The Mailbox That Reads Itself: Inside an Agentic AP Deployment
A behind-the-scenes look at AP mailboxes that triage themselves — reading tone, splitting mixed attachments, and never half-processing an email.
Read articleCase Study
Geotechnical engineering, across North America
The AP Mailbox That Runs Itself
Keller North America turned its shared AP mailboxes into an autonomous pipeline: AI triage that reads tone as well as content, splits mixed attachments, handles statements and queries separately, extracts invoice data, and files everything to SharePoint — with a duplicate ledger and a full audit trail.
1,200+
documents in the six-week pilot
100%
pass on 57-scenario replay
AI EngineWorkflow AutomationiPaaS
The Company
Heavy Engineering, Heavier Inboxes
Keller North America builds foundations for major construction projects across the continent. Every project means suppliers, and every supplier means email: invoices, statements, credit notes, queries — all arriving in shared regional AP mailboxes, in every format imaginable.
Industry
Geotechnical Engineering
Footprint
Project sites across North America
Intake
Shared AP mailboxes per region
Model
Round-robin AP team routing
The Starting Point
The AP team worked the mailboxes by hand: open each email, figure out what it was, download attachments, decide who should process it, and re-key invoice data into finance systems. The hard part wasn’t any single email — it was that no two were alike. One message might carry an invoice, a statement, and a query. Another might be an upset supplier whose payment chase deserved a person’s attention within the hour, sitting unread behind forty routine invoices.
Simple automation had an answer for none of that. Basic OCR couldn’t decide what a document was, let alone notice that a supplier was frustrated. Rules-based routing broke every time a vendor changed a template. The team needed intake that could read a mailbox the way a good AP clerk does — and then do everything a clerk does next, automatically.
The Solution
An AP Clerk’s Judgment, At Machine Speed
The pipeline polls each mailbox on a 30-minute cycle, processes everything new, and leaves a complete record of every decision it made.
Agentic mailbox triage
Every incoming email is read and classified — invoice, statement, query, credit note, or other — and routed down the right path, including the judgment calls rules can’t make.
Sentiment detection
The triage reads tone as well as content. An escalating or frustrated supplier is flagged for human attention immediately instead of waiting in the queue.
Mixed-attachment splitting
One email with an invoice, a statement, and a query becomes three correctly-typed work items — each billed, routed, and tracked separately with a threaded note.
Statement & query handling
Statements and supplier queries follow their own flows instead of clogging the invoice pipeline — triaged, assigned, and answered with context.
Zero-training extraction
Vendor, invoice number, dates, amounts, and references extracted from any format from day one — no labeled data, no model training cycle.
Duplicate detection
Every document is checked against a ledger of 1,200+ processed invoices; duplicates are flagged with a note citing when the original was first processed.
Round-robin team routing
Work is distributed across the AP team automatically, by region and configured rotation — balanced load without a human dispatcher.
SharePoint filing
Finished documents land in SharePoint mirroring the team’s own folder convention — DocQ processes, the system of record stays theirs.
Verified, with rollback
Every processed email is verified end to end; on any failure the pipeline rolls the whole email back cleanly — no half-processed documents, ever.
The Outcome
Production-Grade, Proven Before Go-Live
Every email opened, read, and sorted by hand
Agentic triage classifies and routes autonomously
Upset suppliers buried behind routine mail
Sentiment flags surface urgent messages immediately
Mixed emails processed as one lump
Each attachment split, typed, and tracked separately
Invoice data re-keyed into finance systems
Extracted automatically, any format, day one
Duplicates caught at reconciliation — or not
Flagged on arrival against the processing ledger
1,200+
documents in the six-week pilot
100%
pass on the 57-scenario replay
30 min
autonomous intake cycle
0
half-processed emails — rollback by design
Deep Dives
Read the Full Articles
Go deeper into the Keller North America story with these detailed articles.
Technical8 min read
Anatomy of an Agentic AP Pipeline
A technical deep dive — agentic classification with sentiment, zero-training extraction, duplicate ledgers, round-robin routing, and all-or-nothing rollback.
Read articleBusiness Impact7 min read
What Autonomous AP Intake Is Worth
A finance perspective — recovered team hours, supplier relationships protected by sentiment triage, duplicate prevention, and scale without headcount.
Read articleBuild. Automate. Govern.
One platform to structure your data, automate your processes,
and free your people — with AI baked in.
Every manual step eliminated is a compounding speed advantage.
What are you still doing manually that DocQ could handle instantly?