Three AI agent bots take invoice keying off a distributor's accounts-payable clerk
A wholesale distributor with about 120 staff and 400 active suppliers, a leader in its segment.
- Scale
- About 1,800 supplier invoices a month, 3 bots, 1 ERP, 1 shared mailbox
- Engagement
- 10 weeks for three bots, including a 3-week run-alongside period
The challenge
One accounts-payable clerk spent about 25 hours a week keying invoice data from emailed PDFs into the ERP, matching it to purchase orders by eye and answering supplier questions in a shared mailbox. Errors were common at month end, and early-payment discounts were missed because invoices waited in the queue.
A wholesale distributor with about 120 staff and 400 active suppliers, a leader in its segment, processed roughly 1,800 supplier invoices a month. The client is not named on this page, and this is an illustrative case. One accounts-payable clerk carried most of the load.
That clerk spent about 25 hours a week keying invoice data from emailed PDFs into the ERP, matching each one to a purchase order by eye and answering supplier questions in a shared mailbox. Month-end errors were common, and early-payment discounts lapsed while invoices waited in the queue.
We interviewed the clerk first, then mapped every step and every exception. From that map we built three bots, one per task: invoice capture, purchase-order matching and vendor email triage. Each has its own identity, only the access its task needs and no right to delete anything. The capture bot creates drafts, the matching bot flags mismatches, and the triage bot drafts routine replies.
Human approval stayed where the risk is. Every payment run and every reply that mentions money waits for a person, who approves it in one screen. The bots ran alongside the clerk for three weeks, and the clerk checked each result before it was trusted. During the included period we made unlimited changes, mostly to matching tolerances and to new supplier formats.
In the modeled outcome, keying and matching time fell from about 25 hours a week to about 6, mostly exception review, and roughly one invoice in twelve was flagged for a person. These are illustrative figures, not a guarantee. The clerk stayed and moved to supplier relationships and approvals. Related reading: our Cyber Automation service and /how-we-use-ai/.
What our team delivered
- An invoice capture bot reads emailed and scanned invoices, extracts supplier, number, lines and totals and creates the entry in the ERP as a draft
- A purchase-order matching bot compares each draft with the PO and receipt, and flags every price, quantity or supplier mismatch for the clerk
- A vendor email triage bot sorts the shared mailbox, drafts replies to routine status questions and routes disputes to a person
- Human approval for every payment run and every reply that mentions money, with the clerk approving in one screen
- A separate identity for each bot, access limited to its own task, no deletion rights, and a log of every action
- A 3-week run-alongside period in which the clerk checked each bot's work before it was trusted
- Unlimited changes during the included period, used mostly to adjust matching tolerances and add suppliers' formats
Result
In this modeled engagement, keying and matching time fell from about 25 hours a week to about 6, most of it exception review, and the bots flagged roughly one invoice in twelve for a person. Early-payment discounts stopped being missed. The figures are illustrative, not a guarantee; the clerk stayed, and moved to supplier relationships and approvals.
Get this for your company
Related services with fixed prices and a written scope.
AI agent bot, standard
One bot for a task that crosses two systems or handles varied documents, with exception handling and a review queue for your team.
About 3 to 4 weeks
$5,900 per bot
Cyber automation starter
Automate one repetitive security task, such as user offboarding, alert triage or vulnerability ticketing, with a tested workflow and a runbook.
Live within 3 weeks of kickoff
$4,200 per automated workflow
Microsoft Copilot and AI readiness
Before you turn Copilot on, find out what it will be able to see, and fix the permissions first.
Report within 8 business days
$2,400 per tenant
Show 2 more
Scanning of software codebase
Static analysis and secret detection across one codebase, reviewed by an engineer so you get findings, not noise.
Report within 3 business days of read-only access
$500 per codebase
Discovery and requirements workshop
Workshops with your team to map every system, process, user group and integration, and to agree what the new platform must do.
Written requirements and a system inventory in 2 to 3 weeks
$6,500 fixed fee
After you purchase, we contact you within 1 hour, any time of day or night, to start onboarding.
No payment is taken online. When we call you within the hour we confirm the scope and send an invoice (ACH, card or crypto).
Automated data entry
Bots read invoices, forms and emails, check every field, and post clean records to your books, with a person reviewing only the uncertain cases.
AI email processing
Every email is read, sorted and either answered with a drafted reply or sent to the right person within seconds, and all of it is logged in your CRM.
Three approval tiers for AI agents
Every action an agent can take falls into one of three tiers: it runs on its own, a named person approves it first, or a human always does it. The examples are illustrative; each firm draws its own lines during scoping.
Facing something similar?
Tell us about your environment. We will tell you plainly what we would do and what it would take.