BIM Model to Cost Estimate Integration
Metwaves built an integration that reads IFC building models and loads components and quantities into the estimating and ERP systems, cutting quantity take-off from days to hours.
Learn More >
Goal
Stop the accounts payable team retyping supplier invoices into SAP Business One, without losing control over what gets posted.
Solution
A LangGraph agent that reads each document with Azure Document Intelligence, validates the fields against SAP Business One, posts clean entries and sends exceptions to a review queue with the reason.
Result
About 1,000 documents a month now flow through the agent, with roughly 80% less manual data entry. Live in 8 to 12 weeks.
Technologies used: LangGraph, Azure Document Intelligence, Azure OpenAI, Python, SAP Business One
Background
Our client is an Australian business whose accounts payable team received around 1,000 supplier invoices and related documents every month, by email and as PDF attachments, in many different layouts.
Each one was keyed into SAP Business One by hand: supplier, invoice number, dates, line items, GST and totals. The work was slow, it peaked at month-end, and typing errors surfaced later as payment queries.
Challenge
Reading an invoice is the easy part. The hard part is deciding, reliably, which entries are safe to post without a person looking.
1
Many layouts
Suppliers send documents in their own formats, and many change them without notice.
2
Matching to SAP Business One
Each document must be matched to the right vendor record and, where one exists, the right purchase order.
3
Catching bad data
Duplicates, totals that do not add up and wrong GST must be stopped before they are posted.
4
Keeping people in control
The team needed to see and approve anything the agent was unsure about.
Our Approach
We built the checks first and the automation second, so the agent only posts what it can prove is correct.
01
Sample and rules
We collected real documents from many suppliers and wrote down, with the AP team, what makes an entry correct.
02
Extraction
Azure Document Intelligence reads each document, with Azure OpenAI handling layouts the standard models miss.
03
Validation graph
A LangGraph workflow checks the vendor, purchase order, line totals, GST and duplicates in turn.
04
Posting
Entries that pass every check are posted to SAP Business One through its API.
05
Review queue
Everything else goes to a queue that shows the document, the extracted fields and the failed check.
06
Pilot and rollout
Run alongside the manual process at first, comparing results before the team switched over.
From inbox to posted entry
Every document follows the same path, and every step is logged so the AP team can see what happened to it.
- Capture: Documents arrive from the AP inbox and are queued for processing.
- Extract: Azure Document Intelligence returns the header fields, line items and totals with a confidence for each.
- Validate: The LangGraph workflow checks each field against SAP Business One and the client's rules.
- Post or route: Clean entries are posted; the rest wait in the review queue with the reason.
Checks before anything is posted
The agent never posts on extraction confidence alone. Each entry must also make sense against what SAP Business One already knows.
- Vendor match: The supplier on the document must match an existing vendor record.
- Purchase order: Where a PO is referenced, it must exist and the amounts must fit it.
- Arithmetic and tax: Line items must add up to the total, and GST must be consistent.
- Duplicates: An invoice number already posted for that vendor is stopped and flagged.
A review queue the team actually uses
Exceptions are part of the design, not a failure. The queue shows the reviewer exactly what to look at.
- Reason shown: Each item names the check that failed, so the reviewer does not start from scratch.
- Fix and post: The reviewer corrects the field and posts, without retyping the whole document.
- Learning from exceptions: Recurring exception reasons show which suppliers or rules need attention.
The Results
About 80% less manual entry
Most documents are posted without anyone retyping them; people handle the exceptions.
About 1,000 documents a month
The agent handles the full monthly volume, including the month-end peak.
Errors stopped before posting
Duplicates, wrong totals and unknown suppliers are caught by the checks, not discovered at payment time.
Live in 8 to 12 weeks
From the first document sample to full rollout.
Client name withheld. Details that could identify the business have been left out.
Copilot Studio Helpdesk Agent for a Professional Services Firm
Metwaves built a Copilot Studio agent in Microsoft Teams that answers staff IT, HR and policy questions from SharePoint and resolves 30–50% of requests without a human.
Learn More >
AI Material Recommendation System for BIM
Metwaves built and deployed an AI recommendation engine inside BioBuild's BIM workflow that scores materials on cost, carbon and structural fit. Live since January 2025: selection time down 40%, material cost down 20%.
Learn More >INSIGHTS & NEWS
Discover IT Industry Insights and Technology Trends
Nikit Nandan
November 12, 2026
Top AI agent development companies for small and mid-sized businesses (2026)
Nine AI agent development firms a small or mid-sized business can shortlist, with Clutch figures, minimum budgets and the weaknesses their own clients name. Published by Metwaves, which is not ranked.
Learn More >
Nikit Nandan
November 12, 2026
Copilot Studio vs ServiceNow vs SAP Joule vs Agentforce: An Architecture Review (2026)
An architecture review of the four big enterprise agent platforms: where each one reasons, acts, finds its data and connects to outside agents, what it costs, and when a custom build is the better choice.
Learn More >
Nikit Nandan
November 01, 2026
Top custom software development companies for small and mid-sized businesses (2026)
Ten custom software firms a small or mid-sized business can shortlist, with Clutch figures, minimum budgets and the weaknesses their own clients name. Published by Metwaves, which is not ranked.
Learn More >Let’s build your idea!
Let us help you with free analysis & estimate.
Our product expert will contact you within 12 hours.