Skip to main content

Protect revenue. Release cash. Prevent disruption. Govern enterprise AI. Connect your organization and watch problems move from detected → verified in the Decision Queue. Start the 11-step connection wizard →

Upload pilot · one use case · observe-only

Delivered but not invoiced — upload a safe sample.

Download the template, mask what you need to mask, upload up to 500 rows. We run the detector server-side, show the detections, and delete the file whenever you say so.

Before you upload — what we do with the file

What is processed
Only the fields marked required on the template you downloaded. Rows beyond 500 are truncated before processing. Any column we did not ask for is dropped at ingest.
AI providers
Anthropic Claude Sonnet 4.5 — used to generate the plain-English narrative for each detection. Model is called with field names and aggregated metrics only; raw row values are never sent as prompts.
Retention
90 days by default; configurable to 24 hours in your workspace settings. Deletion is immediate and permanent — Mongo deleteMany with no soft-delete flag.
Deletion
One button: Workspace → Data → Erase uploaded data. Returns within 2 seconds.
Access
Only authenticated users in your workspace with the uploader role or higher. DuluthPath staff access requires a signed-support ticket with your explicit written consent.

Step 1

Download the required-fields template

Four required columns, three optional. Any other column is dropped at ingest.

Required: so_number, customer, delivery_date, value_usd

Optional: billing_block_flag, ship_status, notes

Hard-rejected column names: address, credit_card, dob, email, phone, ssn, tax_id

Step 2

Mask what your legal team needs masked

  • · Replace customer names with aliases (e.g. 'Customer A', 'Customer B'). DuluthPath does not need to know the real customer identity to detect the problem.
  • · Round `value_usd` to the nearest thousand if your legal team prefers less precision.
  • · Dates must stay in `YYYY-MM-DD` format so the aging calculation is correct, but you can shift every date by a constant offset (e.g. +30 days) to further anonymise.
  • · Leave every column we did not ask for OUT of the CSV. We will drop them at ingest, but the safest file is one that never contains them in the first place.
  • · No column named 'ssn', 'tax_id', 'email', 'address' or 'phone' is ever accepted; the server rejects the whole file if any of these column names are present.

Step 3

Upload