Screen orders for fraud signals and escalate the unsure ones
Each new order gets a fraud risk level from a decision model; low-risk orders ship as normal, unsure ones go to a larger model for a second look, and high-risk ones wait for a person.
The use case
Small stores see a steady trickle of card-testing and stolen-card orders, and checking every order by hand slows shipping. Jev reads the order details your store already has, such as billing and shipping mismatch, email age signals, order value, and the platform's own risk notes, and answers how risky the order looks on a fixed scale. Orders rated low with a confidence of 0.9 or higher continue as normal. Orders below that confidence are passed to a larger model such as Claude for a second opinion. Any order rated high, or still unsure, is held for a person. Jev never cancels or refunds an order. The owner decides whether to ship, contact the customer, or cancel.
Example workflow
- A new order in Shopify triggers a Make scenario.
- The order fields, Shopify's fraud analysis notes, and simple history from a Google Sheet are sent to Jev as the state.
- Jev answers a Score question, how risky does this order look, with the levels low, medium, and high, and a Noul question, the billing and shipping details look consistent.
- Orders rated low at or above the confidence threshold are left to ship as normal.
- Orders below the threshold are sent to Claude with the same details for a second opinion, and anything rated high or still unsure is tagged on hold in Shopify.
- The owner reviews held orders in a Slack channel and decides whether to ship, contact the customer, or cancel.
Tools
- Jev (TypeSafe AI)
- Shopify
- Make
- Claude
- Slack
- Google Sheets
Risks and guardrails
- A false high-risk rating delays a real customer — guardrail: held orders are reviewed the same business day, and holds are tracked against outcomes.
- Fraud patterns change — guardrail: review a sample of shipped and held orders monthly and adjust the questions and threshold.
- Automation could cancel orders — guardrail: the scenario can only add a hold tag; cancellations and refunds are done by a person.
A similar pattern in a public build: https://madewithjev.com/builds/fraud-detection-jev-kimi. See also Decision models.
Related recipe
Shopify Order-Exception Triage Agent: An agent reviews new Shopify orders for address problems, risk flags, stock gaps, split shipments, and customer notes, then sorts them, routes them, and drafts customer messages for a person to approve.
Want it set up for you? Request a quote.
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Public build with a similar pattern: madewithjev.com/builds/fraud-detection-jev-kimi
- Shopify Dev MCP server (AI Toolkit) (official page): shopify.dev/docs/apps/build/ai-toolkit (checked 2026-11-09)
- Make MCP server (official page): developers.make.com/mcp-server (checked 2026-11-09)
- Claude (official page): claude.ai (checked 2026-11-09)
- Slack MCP server (official page): docs.slack.dev/ai/slack-mcp-server (checked 2026-11-09)