Case studies / 1 of 4
Case study 1
AI agents inside NetSuite
MCP-connected assistants that reconcile bank lines and route purchase approvals, with a live React dashboard.
Problem
The problem
Finance and purchasing teams spent hours every week on checks that were repetitive but not mechanical: matching bank lines to transactions, validating purchase orders against budget and vendor setup, and chasing approvals. Most cases were obvious. A person still had to look at all of them to find the few that were not.
Step 1
An agent with one door into NetSuite
I built a small MCP server whose tools map one to one onto NetSuite RESTlets. Claude does the reasoning and can only reach the ERP through those tools. There is no general API access, so the model can only ask the questions I wrote endpoints for.
Step 2
Narrow RESTlets, structured answers
Each RESTlet does one thing, such as finding transactions in an amount and date window or returning a budget line. They return JSON with fixed fields, never prose, so the agent reasons about data rather than summaries. Reads are free; writes sit behind a confidence gate.
Step 3
A dashboard people actually use
A React dashboard shows what every agent did, with the inputs, the reasoning and the confidence. High-confidence matches and clean purchase orders go through on their own. Everything else lands in a review queue where a person approves or rejects in one click.
Result
The result
Routine validation, record creation and follow-up communication now run through the agents. People handle exceptions instead of every step, and every decision is logged to a custom record in NetSuite for audit. The first version was read-only for a few weeks so the team could judge the agent before it was allowed to act.
Comments
Questions and thoughts on this case study are welcome. Sign in with GitHub to post. To answer someone, click Reply under their comment and your answer sits indented beneath it. I reply here as well.