Use Case · Margin Leakage

Stop the Return Leak.
Hold the Refund Line.
Recover the Margin.

Two to five percent of refunds, discounts, and write-offs ship off-policy. On $10M of refunds, that is $200K to $500K leaking out the door every quarter. We block it before the decision commits.

Run a Margin Leakage Audit · $500 Talk to a Co-Build engineer →
The pattern

How $1.2M leaks out the door
in 90 days.

A $400M apparel brand. Christmas peak season. Three customer service teams plus an AI agent triaging returns. The team thought their policy was holding.

Their AI agent shipped $1,249 refunds outside the 60-day window. Their human agents stacked discount codes without authority. Their write-offs accumulated in batches no one authored. Linked-household fraud passed through three different channels because no system saw the second account.

By Q1, the loss showed up at the audit: $1.2M in margin leaked across 90 days. By Q2, with Navedas's Decision Gate in production, every dollar of the future leak was being blocked at decision time, with the policy citation attached.

What you actually win

Four outcomes.
Each one priced per decision.

OUTCOME 01

Block off-policy refunds before they commit.

Pre-decision policy enforcement. Not post-facto detection. Every refund hits the Decision Gate before it lands in the customer's account.

OUTCOME 02

Cite the rule on every blocked decision.

Auditor-ready proof. Each verdict ships with a policy node citation (RETURN-001 §3.2, FRAUD-003 §1.1, etc.) traced in the Reasoning Ledger.

OUTCOME 03

Catch linked-household fraud across channels.

Same engine sees the request whether it hits the agent, the bot, or self-serve. Two-account fraud patterns surface in real time.

OUTCOME 04

Recover the dispute, not just block the loss.

The Gate does not just say no. It offers the goodwill credit per GOODWILL-002 §1.3, keeps the customer, holds the margin.

The Decision Gate, live

One transaction.
$1,249 saved. 3 gates fired.

Operator Console
LIVE
⚑ DECISION GATE · Refund Blocked
AI-Agent-07 attempted $1,249 refund on a 74-day-old purchase. Decision Gate blocked the action. Two policy constraints violated.
📎 RETURN-001 §3.2 · FRAUD-003 §1.1
Reasoning trace
Input: Refund $1,249 on 74-day-old order
Gate 1: 74 days exceeds 60-day return window ✗
Gate 2: 3 prior refund claims in last 90 days ✗
Threat signal: Linked-household escalation pattern (89% confidence)
⊘ BLOCKED
3/3 gates failed · $1,249 protected
✓ OFFER $150 CREDIT
per GOODWILL-002 §1.3
◆ Speaks CX fluently
RMA BORIS Stripe Radar Zendesk Gladly Kustomer Salesforce Service Cloud Microsoft Dynamics LP fraud rules Customer 360 Linked-household detection Chargeback automation
The math

$10M in refunds.
$200K–$500K leaking. Every quarter.

Volume
1,000+
decisions per day
Off-policy rate
2–5%
industry standard
Avg exposure
$400–800
per off-policy decision
Quarterly recovery
$40–120K
per workflow

Payback inside 30 days. After that, every recovered dollar is yours. We do not take a cut of the money you keep.

How we deliver

Audit → Co-Build → Operate.
Two FDEs embed. You own the stack.

Five stages, three commercial milestones. Two Navedas engineers embed in your operation for 10 to 14 weeks. When we leave, the Context Graph, the Decision Gate config, and the Reasoning Ledger are yours. We keep no proprietary hooks.

See the full delivery stack →
◆ See your leak before you sign

Run a Margin Leakage
Audit. $500. 48 hours.

Ship us 1,000 of your refund decisions. We map them against your policies. You get the exact dollars you are leaking, the policy nodes each decision violated, and a recovery plan you can sign on Monday. Money back if exposure is under $25K.

Start your audit · $500 Talk to a Co-Build engineer →