Skip to main content
Once you have connected the OuterProduct MCP server, you can drive the full platform through natural language within the client application. The prompts below are ready to paste into Claude, Cursor, or any other MCP-compatible client.

​
Built-in workflows

OuterProduct ships two pre-built, end-to-end workflows. When your prompt matches a workflow, the assistant follows it automatically without any extra setup on your part.

Suspicious Activity Reports (SAR)

For fraud and compliance teams. Produces a structured SAR backed by model-level evidence.

Churn Winback Briefings

For customer success teams. Ranks at-risk customers and surfaces realistic retention levers.
Browse the full collection of skills in the outerproduct-skills repository.

​
Ready-to-use prompts

Paste this prompt into your MCP client, then attach or describe the path to your CSV:
I have a CSV of customer subscription data with a churn column. Tell me which behaviors most predict churn, and for an at-risk customer, what realistic changes would lower their churn probability? Wrap it up as a high-level summary I can share with my VP.
What the assistant does:
1

Upload

Sends your CSV to OuterProduct using the upload tool.
2

Train

Fits a classification model on your subscription data with churn as the target column.
3

Rank features

Calls the explanation tool to identify which behaviours — login frequency, plan tier, support tickets, and so on — most strongly predict churn across the dataset.
4

Run counterfactual

Selects an at-risk customer and finds the smallest realistic changes that would flip their predicted outcome from churned to retained.
5

Report

Produces a plain-English summary suitable for sharing with a VP or non-technical stakeholder.
Paste this prompt and point the assistant at your transaction history file:
Train a fraud-detection model on this transaction history, then walk me through why transaction txn_4582 was flagged as suspicious. Which signals contributed the most?
What the assistant does:
1

Train

Fits a fraud-detection model on your transaction history.
2

Explain

Calls the per-feature explanation tool for transaction txn_4582, quantifying exactly how much each signal — amount, merchant category, time of day, velocity, and so on — contributed to the suspicious-activity score.
3

Narrate

Translates the feature scores into a plain-English walkthrough that an analyst or auditor can read and act on.
This prompt pairs naturally with the built-in SAR workflow. Ask the assistant to “draft a Suspicious Activity Report for txn_4582” after the explanation step to get a compliance-ready document automatically.

​
More prompt ideas to try

These starting points cover other domains where OuterProduct’s explain-and-act loop adds immediate value.

Credit underwriting

Train a loan-default model on this application dataset, then explain why applicant A-10293 was declined and what they could realistically change to qualify.

Recommender quality audit

Build a purchase-propensity model from this user-event log, then tell me which engagement signals drive recommendations most.

Equipment maintenance

Fit a failure-prediction model on this sensor log, rank the readings that most predict a breakdown, and tell me which thresholds I should alert on.

HR attrition

Train an attrition model on this HR dataset, identify the top drivers of voluntary departure, and produce a brief I can bring to the People team.
All prompts assume the OuterProduct MCP server is already connected and authenticated. If you haven’t set that up yet, start with the MCP Server overview.