Decide · AI Governance & Approvals
AI that acts for you — without taking the final word from you.
Flash's AI can issue coupons, award points, apply tags and more on your behalf. Governance is what makes that safe to say yes to: every AI action is explainable (you see the reasoning and the evidence), authorized (you set the gear per action), reversible (one-click undo) and reviewable (a complete decision log).
Available as governed actions roll out to your workspace.
Send a win-back coupon
why the AI proposes it
This member's purchases have gone quiet, and members like them usually respond to a small offer.
the evidence it cites
The exact records it read — so you can check its reasoning, not take it on faith.
what happens if you approve
The consequence in plain language — one member, one coupon, nothing else.
approve
reject · with a note
Illustrative card · similar proposals of the same action type can be approved in one batch
The problem
You'd let AI do more — if you could see, approve and undo what it does.
A yes you can't inspect
Most AI tools ask you to trust a black box. Without the reasoning and the evidence behind each action, "approve" is really "hope".
All-or-nothing autonomy
Either the AI does nothing without you, or it does everything without you. Real teams need a dial per action — not one switch for the whole system.
Yesterday's reasoning, today's action
A proposal written on Monday can be wrong by Thursday. If nothing re-checks the reasoning at approval time, you act on stale facts.
The console
One place to see, approve and replay everything the AI does.
The AI Activity & Approvals console lives in your Flash admin. Three views, one habit: check it like you check your inbox.
Pending approvals
Each AI proposal arrives as a card: the reasoning, the evidence it cites, the consequence in plain language, and a countdown to expiry. Approve it, or reject it with a note the AI's record keeps. Similar proposals of the same action type can be approved in one batch.
Decision log
Every AI action — automatic or approved — lands here as a complete record: the reasoning, the action taken, who authorized it (the gear or a person), and the result. Filter by action or by outcome, and undo any single entry.
AI permissions
The control panel for the gears: see every AI action your workspace allows, and set each one to automatic, human review, or forbidden. What you see here is exactly what the AI is allowed to do — nothing more.
How authorization works
A dial per action — not one switch for the whole AI.
Ten AI actions — issuing coupons, awarding points, applying tags, adding member notes and more — each carry their own authorization gear, set per team: automatic, human review, or forbidden. Start everything on review; promote the actions you've come to trust; forbid the ones you never want the AI to touch.
One line is not negotiable: actions that send messages to customers are permanently locked to human review — they can never be set to automatic, by anyone.
Action gears
Every action, on the gear you choose.
Automatic
The AI executes on its own. Every action still lands in the decision log.
Human review
The AI proposes; a person approves or rejects before anything happens.
Forbidden
The AI cannot take this action at all — proposals are refused outright.
Sending is permanently locked
Actions that send messages to customers can never be set to automatic — a person always approves them first.
Fresh at the moment of approval
Approval isn't a blank cheque.
When you approve a proposal, the platform re-checks whether the AI's reasoning still holds at that moment — not when the proposal was written.
Proposed on Monday
“This member has gone quiet — send a win-back coupon.”
Approved on Thursday — but the member bought on Wednesday
The reason no longer holds, so the coupon is not sent. You never act on stale reasoning.
Safety nets
Built for the day something goes wrong.
Trust in AI isn't built by promising nothing will ever misfire — it's built by what happens when it does.
Misbehaving actions demote themselves
Each action is watched independently. If its results turn unusual past a set threshold, that action is automatically moved back to human review — it only ever moves down on its own, and only a person can promote it again.
A daily digest, not a firehose
A daily email summarizes what's waiting for approval, so nothing sits unseen. Each person can turn their own digest off.
Proposals expire in 7 days
An unapproved proposal doesn't linger forever — after 7 days it expires on its own. No forgotten approvals firing weeks later.
Audited end to end, undone in one click
Every step — proposal, decision, execution, result — is recorded. And any single action in the decision log can be undone.
The same rules apply to any AI you connect.
Governance isn't only for Flash's own AI. Any AI assistant your team connects can first look before it acts: discover which business objects exist — members, orders, coupons, campaigns, stores, products, segments and tags, including the fields and objects your team added — what each field means, how objects relate, and which actions it may take, with the conditions and the current gear for each. Fields holding personal information are described but their values are never handed over. And when a connected AI proposes an action on a review gear, it waits in the same approvals queue as everything else.
FAQ
The questions teams ask before they say yes.
Can the AI ever send something to a customer on its own?
No. Actions that send messages to customers are permanently locked to human review — that gear cannot be changed. Other actions run on whatever gear your team sets, action by action.
What if we approve something that's no longer right?
Two protections. At approval time, the platform re-checks whether the AI's reasoning still holds — if the customer already came back, a win-back coupon simply isn't sent. And afterwards, any single action in the decision log can be undone.
What happens if an action starts producing bad results?
Each action is monitored on its own. If its results turn unusual past a set threshold, the action is automatically demoted to human review. Automatic changes only ever go downward — restoring an action to automatic always takes a person.
Do we have to review everything from day one?
You choose. Many teams start with every action on human review, watch the decision log, then promote the actions that have earned it. Approvals arrive as cards you can batch by action type, a daily digest keeps them from being missed, and anything unreviewed expires after 7 days.
Related reading
Keep exploring the pages most related to this one.
Intelligent Decision
A next-best-action: who, which offer, which channel, when.
Read more CapabilityDashboard
Real-time scan-to-redemption data as KPI trends, funnel diagnostics, and a rule + LLM next-best-action panel.
Read more CapabilitySoTag for Slack
The Slack agent that hands governed, audited numbers to Claude in-thread: daily anomaly scan + human-approved coupon-send. Early access.
Read more CapabilitySoClaw — AI Win-Back
An autonomous AI teammate that wins back high-value, at-risk customers one at a time — a personalized offer or a deliberate hold — and proves the real incremental lift against an untouched control group. Off by default, capped, instantly pausable. Early access.
Read more CapabilityBusiness Ontology
The concept behind the governed AI stack: a world model of your business — objects, properties and relationships plus governed actions — as a metadata layer mapped over the data you already have, so AI knows what a customer is, what has happened, and what it may do about it.
Read more CapabilityOne Semantic Layer
The organizing principle of the platform: three governed assets — certified metrics, governed tags, and behavior → intent — hung on one shared object spine, so every consumer (dashboard, AI agent, API, campaign tool) reads business meaning from one authoritative place and never disagrees.
Read more