Guardrails

Reporting an exception
is not preventing it.

Most systems stop at the alert and call it a control. Seal carries it the rest of the way: it finds the anomaly, turns it into a policy, runs that policy across every store, and tracks whether the loss actually moved.

From a sentence to a running policy

Describe the control.
Seal builds the workflow.

Say what should happen in your own words. Seal lays out the trigger, the query, the logic and the action, then hands you the canvas so you can change any step before a single thing runs.

StepWorkflowRuns Voids that do not look like anyone else Draft
No workflows yet A workflow watches a number and acts when it moves. Shall we build your first one? Create a workflow
Start a workflow

Build one from scratch, or take a starter and change it. Nothing runs until you activate it.

Author a new one
Describe it and Seal builds itSay what should happen. You get steps you can change.
Choose from the starter libraryGrouped by what they watch. Take one and change what you need.
Start from an empty canvasNo steps and no trigger. You add every one yourself.
Describe it and Seal builds it Reading your policy
Build it
Tell the DC when a top SKU starts losing moreOpen a case when a cashier voids like nobody elseWatch customers who return far more than their categoryFlag a lane when shipped and received drift apart
TriggerEvery morning, 07:00inferred QueryThe rows that moved most LogicIf it moved more than 10% ActionOpen a case
Every morning, 07:00Account time zone.TriggerSchedule
The rows that moved mostLast 14 days, all stores.Query20 rows
If it moved more than 10%Otherwise the row is skipped.Logic
Open a caseAssigned to the region’s LP manager. Nothing leaves Seal.Actionmax 25/day
Peer comparisonThe same metric against comparable stores.Query
Dry run, last 90 days 0 times Would have fired Would have notified6 people Busiest week4 SKUs, 19 Aug Quietest monthJuly, nothing fired Nothing is sent from a dry run.
Activatedruns every morning on its own
83%TRIGGERQUERYLOGICACTION
An empty account. Nothing is watching anything yet.
The solution

Control the loss without
slowing the store down.

Seal maps every vertical, store and channel you are exposed on, then gives you the limits, the roles and the routing to act on them. Guardrails a store manager can live with, and an auditor can follow.

RolesWho sees what
AnalystInvestigatorStore ManagerWarehouse
InvestigatorWorks cases end to end across assigned verticals.
Loss verticals6 of 10
Stores3 of 9

Roles and access

Decide who may see which loss verticals and which stores. An investigator working gift card abuse does not need the payroll view, and will not be given it.

Policy limits

Set thresholds per region, per category and per store. Seal tells you as a limit is being approached, not after the quarter has closed.

Shrink spike detectedStore 412 up 184% against its own baseline
Threshold exceededReturns at 108% of the category limit
Unapproved discountA code nobody signed off is live in 22 lanes

Exception alerts

Told when shrink spikes against a store’s own baseline, when a return rate crosses its line, or when a discount goes live that nobody approved.

New caseStore ManagerRegional LP leadFinance review

Case routing

Every case lands with the person who can actually close it. Route by vertical, by region or by value, with an approval step wherever you want one.

Case 8841Thu, 4 Sep
Opened by policy07:00
Evidence attached07:02
Approved by M. Lee09:14

Audit trail

Every action is logged with who did it, when, and on what evidence. When an auditor asks why a case was opened in March, the record is already written.

The whole loop, not the first step

Exception reports end here.
Control keeps going.

Four things have to happen for a loss to actually stop, and most stacks only do the first. Seal owns all four, and the last one is where the money is.

01

Find the anomaly

Seal watches every metric against its own history and against comparable stores. When a number moves in a way it should not, that becomes a finding with a dollar value on it, not another row in an exception report nobody owns.

02

Implement the policy

Turn the finding into a control. Describe it in a sentence, take one of the starters, or draw it yourself on the canvas. Nothing runs until you activate it, and you can change any step before it does.

03

Track every action

Every run is recorded: what fired, what it looked at, what it decided, and who approved it. When someone asks why a case was opened in March, the answer is one click away rather than an afternoon of reconstruction.

04

Agentic investigations

When a case needs more than a rule, an agent picks it up. It pulls the transactions, the peer comparison and the history, writes up what it found, and hands a person the evidence instead of a hunch.

Stop reporting the loss.
Start closing it.

Bring us one category you cannot explain. We will build the control against it on a call and show you what it would have caught last quarter.

Book a discovery call