5.0 Brain

Ask the question,
get the operation

Plain-language operator agent. Reads the wallet data. Takes audited action. English or Arabic. Every read traceable, every action reversible.

Query model

Operator phrasing, not SQL

Ask in English or Arabic. The brain compiles the question into a structured query, runs it across wallet + attribution + customers + inventory, returns the answer.

Reads

Read-only, scoped, auditable

The brain reads through a typed schema layer. No writes here. Every query logs the operator who asked, the question, and the rows returned.

wallet attribution customers inventory incentives

Typed surfaces

wallet.*, attribution.*, customers.*, inventory.*, incentives.*. Multi-join in one query.

Aggregation180ms
Multi-join1.2s
Full-scan3.7s

Sub-second simple, sub-5s deep

Aggregations under 200ms. Multi-join under 1.5s. Full-scan under 5s.

Subscribers3 viewing
Last refresh2s ago
Re-run ratelive

Live re-run

When underlying data updates, queries auto-re-run for any operator on the same view.

Decides

Action requires explicit approval

When the operator asks for an action, brain previews scope, expected attribution, and consequences. The operator confirms. Only then does brain act.

Customers affected348
Total SAR3,480.00
Expected ROAS2.8×

Scope preview

N customers affected, total SAR exposure, segment fingerprint, expected ROAS.

Confirm1 operator
Rollback window15 min
Audit entrylogged

Approval flow

Operator confirms. Multi-operator approval for high-value actions. Rollback window built in.

Acts

Same primitives as the operator console

Approved actions run through the existing operator primitives. No new attack surface. Same audit trail as a hand-authored rule.

Target348 wallets
AmountSAR 10 each

wallet.credit

Credit SAR amount to one or many wallets.

Triggerdormant 21d
Actioncredit SAR 15

incentives.rule_create

Author a rule from brain. Drops into the editor for refinement.

Operatoraashish@
Brain query"give 10 to dormant"

audit.log

Every action logged with operator + brain query + customers affected.

"We stopped opening BI tools. Brain answers what we need in twelve seconds, in English."

CEO · GCC F&B chain

"It writes our rules now. We just say what we want. It scopes the impact, we approve, it runs."

Marketing lead · GCC retail group

Built for the future. Available today.

Talk to us Read the docs
Tech specs
5.1

Query model

Plain-language operator agent. The operator asks a question in English or Arabic; the brain compiles it into a structured query against wallet + attribution + customers + inventory data, runs it, and returns an answer.

Example: "Show me the customers who reloaded twice in the last week at Olaya but never at Tahlia, sorted by total spend." The brain composes the query, runs it, returns a list. The operator clicks any row to drill in.

Query capabilities

Property Function
Natural language input

English or Arabic. Operator phrasing, not SQL.

Multi-dimensional joins

Joins wallet + attribution + customers + inventory in one query.

Time-window scoping

Rolling windows, calendar windows, custom ranges.

Segment composition

Combine multiple segments with AND / OR / NOT.

Result drill-in

Click any row to expand to the underlying wallet event stream.

Saved queries

Operator can name + save + share queries via URL.
Tech specs
5.2

Reads

The brain reads, never writes, through a read-only schema layer that exposes wallet, attribution, customers, inventory, and incentives data. Every read is auditable.

Read latency targets: simple aggregations under 200ms, multi-join queries under 1.5s, full-scan queries under 5s. The brain re-runs queries automatically when the underlying data updates if the operator is on the same view.

Data surfaces

Property Function
wallet.*

Wallet events, balances, treasury position, settlement state.

attribution.*

Per-branch + per-campaign + per-cohort attribution views.

customers.*

Customer records, segments, cohorts, tier assignments.

inventory.*

SKU records, per-branch velocity, reorder signals.

incentives.*

Rule definitions, fire events, attribution attached.

metadata.*

Operator account, branches, tills, staff roles, audit log.
Tech specs
5.3

Decides

When the operator asks for an ACTION ("send a SAR 10 credit to all customers who haven't visited in 30 days"), the brain compiles the query, scopes the impact, surfaces it for operator approval, and only acts after explicit confirmation.

Every decision shows: the customers affected, the SAR amount, the campaign tag, the expected attribution flow. The operator sees the consequence before signing off.

Decision surfaces

Property Function
Scope preview

Number of customers affected + total SAR exposed + segment fingerprint.

Attribution preview

Campaign tag, expected ROAS bands, attribution flow.

Approval flow

Operator confirms; multi-operator approval for actions above a threshold.

Dry run

Run against last 30 days of data; see how many customers it would have hit.

Rollback window

Reversal window for any action. Audit-clean retraction path.
Tech specs
5.4

Acts

Approved actions execute through the same primitives as the operator console: boost rules, wallet credits, push messages, segment tags. No new attack surface, no separate audit trail.

Every action is logged: which operator approved, which brain query authored it, which customers were affected, which attribution tag was assigned. The audit trail is the same shape as a hand-authored rule.

Action types

Property Function
wallet.credit

Credit a SAR amount to one or many wallets. Reversible within rollback window.

incentives.rule_create

Author a new incentive rule. Drops into the incentives editor for refinement.

messages.send

Send a push notification or WhatsApp message to a customer list.

tags.apply

Apply a custom tag to a customer or segment.

segments.create

Create a named segment for reuse in future rules + queries.

audit.log

Every action emits a log entry tied to operator + brain query.