AI automation build (Built)

AI Accounts Payable Workflow

Accounts payable is reading invoices, typing them in, matching them to POs and chasing approvals. It’s careful work, and it’s exactly the kind of work that goes wrong when people are tired.

Status
AI automation build
Area
Document & Back-office Automation
Workflow
5 steps · 3 failure paths
Stack

01 · Why it mattered

What the manual way was costing

  • Manual entry means typos in money.
  • Approvals get stuck in inboxes.

02 · System concept

What I set out to build

Extract the invoice data, match it to the vendor and PO, route it to the right approver, prepare payment, and stop on anything that doesn’t match.

03 · Before / after

From manual steps to a system

Toggle between the manual process and the system. Click any step to see what it does, what tools it touches, and what happens if it goes wrong.

How it ran manually · depends on whoever is free

  1. 01Invoice arrives by email or uploadby hand
  2. 02Details are typed in by handby hand
  3. 03Matched to vendor and PO by eyeby hand
  4. 04Approval chased over emailby hand
  5. 05Payment prepared manuallyby hand

04 · Decision log

Why I built it this way

D-01Stop on mismatch

With money, a wrong automatic action is expensive. The system stops and hands over whenever something doesn’t line up.

05 · Reliability layer

What can go wrong, and who steps in

Where it happens
Step 03 Vendor / PO match
What goes wrong
Amount or PO doesn’t match
What the system does
Invoice is stopped and sent for manual review. Nothing gets paid on a guess.
Who handles it
A person: finance

Human-in-the-loop points

  • Human in the loopApproval routing
  • Human in the loopAmount or PO doesn’t match → finance
  • Human in the loopInvoice can’t be read reliably → finance
  • Human in the loopApproval not given → approver

06 · Evidence

What exists today

  • AI automation build.

07 · What this demonstrates

What this demonstrates

Applying AI to document-heavy back-office work with strict guardrails.

08 · Stack / integrations

What it’s built with

  • Document processingDocument processing
  • LLMField extraction
  • Workflow automationRouting and payment prep

Architecture · static view · 5 steps · 3 failure paths

    • → financeInvoice can’t be read reliably
    • → financeAmount or PO doesn’t match
    • → approverApproval not given

More systems

Keep reading

Proof of concept

RAG Knowledge Assistant

Answers questions from a specific knowledge base, shows where the answer came from, and falls back or escalates when it isn’t confident.

  1. Question
  2. Retrieval
  3. Relevant knowledge
  4. Grounded answer + source
  5. Confidence check
  6. +1 more
2 failure paths · 2 human handoffs
Proof of concept

AI Customer Support Assistant

Classifies incoming customer messages, answers from the knowledge base, escalates when needed and keeps the ticket updated.

  1. Customer message
  2. Intent
  3. Retrieval
  4. Response
  5. Resolve or escalate
  6. +2 more
2 failure paths · 2 human handoffs

Similar workflow?

Got a process that looks something like this?

Tell me how it runs today and where it hurts. If it’s a good fit for automation I’ll tell you how I’d approach it. If it isn’t, I’ll tell you that too.

Tell me what you’re trying to automate