## TL;DR
A global threshold cut hits hardest exactly where you can least afford it: high value customers transact bigger, travel more, and look anomalous more, so the same score that means risk on a stranger means Tuesday for a VIP. Before any threshold change, run the impact by segment: how many orders per tier would the new line have blocked last month. Then put a recognition layer above the threshold, established account, successful history, verified identity, that routes borderline VIP orders to review instead of auto block, and stage the change with alerts on tier level block rates.

```text
fraud agent lowered risk threshold and auto-blocked vip customers
```

## Use this when
- Established or VIP customers were auto blocked after a threshold change.
- Block rate rose most in the highest value segment.
- The threshold was tuned on overall fraud rate with no segment view.

## Not for this skill when
- The blocks were correct fraud on compromised VIP accounts. Account takeover of a VIP needs blocking plus outreach, not a softer threshold.
- Every segment rose equally. That suggests the change matched a real fraud rise; validate before rolling back.

## Steps
1. List the customers blocked by the change and segment them: account age, lifetime value, prior disputes, and whether any were confirmed fraud. Verify: you know the confirmed fraud share of the new blocks per segment.
2. Release wrongly blocked VIP orders through review, with a direct apology and account note from the account team, not a generic fraud message. Verify: each wrongly blocked VIP has a reviewed outcome and an owner for follow up.
3. Build the segment impact report into the threshold change process: no change ships without per tier projected blocks from historical scoring. Verify: the change template includes the segment table.
4. Add the recognition route: orders above the auto block line from established, verified accounts go to expedited review with a short SLA instead of a hard block. Verify: a staged VIP order over the line lands in expedited review, not blocked.
5. Stage future threshold moves in steps with tier level block alerts for the first week, and a one step rollback ready. Verify: the staged plan and rollback step exist before the next change.

## Variant phrasings
### risk threshold blocked best customers
Customer impact phrasing.
### vip customers auto blocked fraud tuning
VIP phrasing.
### segment impact analysis fraud threshold change
Analysis phrasing.

Compatibility: Stripe Radar and Sigma style analysis, Sift, Kount, and internal scoring. VIP handling must be a recognition and review route, never a blanket exemption, so takeover protection stays intact.

## Why it happens
Thresholds are tuned on population averages, and averages erase the segment structure of risk. High value customers generate high value signals, unusual amounts, unusual geographies, new devices on trips, that scoring reads as risk. A cut that buys a small average gain can therefore concentrate its entire cost on the few percent of customers who fund the business.

## Edge cases / pitfalls
- Recognition must verify identity signals, not just account age, or a taken over old account inherits VIP softness.
- Do not publish VIP criteria to customers or in support scripts. The criteria are a fraud control.
- Review the VIP route outcomes separately. If confirmed fraud through the route rises, the route, not the threshold, is the problem.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_bXglja2rTsa77M0oGK8dBQ