Skip to content
MouseCat
How it works

MouseCat’s foundation model learns your users and your company’s risk landscape

01

Ingest

MouseCat ingests historical and live event data through a single API (CreateEvent). That data is used to train a proprietary, single-tenant foundation model (Lion) that understands your users and risk trends. We model the similarity between users and their probability of different risk vectors. As new events arrive, the model updates the users' representation and risk probabilities.

Your labels

Recluster the space around your own labels — users embedded near a confirmed cluster surface before they act.

▲ 23 unlabeled users near recent confirmed chargebacks — high riskDrag to rotate
Live events
  • u-447122:44:07
    click transfer
  • u-903822:44:02
    view home
  • u-221422:43:52
    open settings
  • u-736022:43:41
    view balance

Each event re-embeds its user the moment it arrives.

02

Decide

MouseCat detects risk vectors and conducts investigations using a single API (Decide). Decide uses a combination of the proprietary Lion model, rules, and other models to determine the best course of action for this transaction or user. You can decide the set of options available (e.g. increase withholding, draft a SAR, escalate to a human, etc.) and even add custom actions. In addition to the best action to take, every decision is accompanied by the probability of different risk categories, evidence and reasoning, and a regulator-ready audit trail.

MouseCat · Ring 7f3REPLAY
→ Decide × 5 accounts in ring 7f3
1.2M USERS · EMBEDDING SPACE, PROJECTED
RING CONFIRMED✕u-4471DECLINE✕u-4472DECLINE✕u-4473DECLINE●u-4474REVIEW✓u-4475APPROVECLUSTER 7F3 · DISTANCE = BEHAVIORAL SIMILARITY
  • 00:31u-4471 · Three identical $1,950 ACH deposits on consecutive days — the account is 3 days old.DECLINE
  • 01:04u-4472 · Same session cadence as u-4471, dormant otherwise; payout enrolled minutes after signup.DECLINE
  • 01:47u-4473 · Payout account owner differs from the verified identity; IDV passed on the third attempt.DECLINE
  • 02:20u-4474 · Behavior matches the ring, but funds are still in place — held for a human decision.REVIEW
  • 02:58u-4475 · Longstanding account with no changes in recent activity.APPROVE
RING CONTAINED1 FOR REVIEW3 declined · 1 held for a human decision · 1 approved · decided in 3m41s
03

Infer

We also offer an API (Infer) to directly obtain a rich embedding from the Lion model which can be used in downstream systems such as in-house ML models.

Live event streamlive
  • 22:41:05Login · new device · new ASN
↓
Live embedding · u-8817

re-embedded 12 ms after each event

Previous sessions
  • Aug 12 · 09:1431 events · recognized device
  • Aug 10 · 18:4712 events · recognized device
  • Aug 08 · 09:0227 events · recognized device
  • Aug 05 · 20:3119 events · recognized device
↓
Past session embedding

the account's own baseline, averaged from its last 30 sessions

↓
Your ATO modelcustomer-owned
p(ATO)0.32

Score users using a live embedding, average of historic session embeddings, and your team's features.

  • ALLOW22:41:05
04

Deploy

All of the data you share with MouseCat is scoped per tenant and by default is never used to train global models. For highly security sensitive customers, we offer private cloud deployment options where MouseCat runs on single-tenant infrastructure jointly owned and operated by MouseCat and the customer. For customers with limited engineering teams, we provide professional services to help setup, deploy, and maintain integration with MouseCat.

For engineering teams, MouseCat supports features like custom authorization schemes, versioning, evaluation and backtesting, and more.

STANDARD

  • operated by MouseCat
  • scoped to your tenant
  • never used to train global models

PRIVATE CLOUD

  • cloud agnostic (works with GCP, AWS, Azure, etc)
  • single-tenant infrastructure
  • configurable to meet customers' specific security requirements