Lumière Skin Malaysia
Synthetic demo data
Six moments: detect, explain, predict, recommend, act, then close the loop.
Date range
Region
Branch
SKU
Campaign
Compare

Ask Neura57

The tested intent catalogue, the assistant's scope rules, and the safeguards behind every answer.

Comparison
Language layer live · global.anthropic.claude-sonnet-5 in ap-southeast-1

Tested intent catalogue

i
40 tested phrasings across 10 families
IDIntentRequired parametersDeterministic calculationAction offered
INT-01
Management summary
+3 more tested phrasings
Period (default yesterday); scope (default all branches)M-01, M-02, M-03, M-04 against the comparison period; open alerts ranked by severityExport summary; send for approval (FRM-06)
INT-02
Comparison
+3 more tested phrasings
Period; comparison baseline; measure; scopeMeasure per branch against that branch’s own normal baseline, indexed to network movement, ranked by varianceDrill to SCR-02
INT-03
Root-cause analysis
+3 more tested phrasings
Branch; periodDecomposition of the variance across availability, basket size and promotion timingPrioritise replenishment; notify owner
INT-04
Forecasting
+3 more tested phrasings
SKU; horizon (default 30 days); grain (branch or warehouse)Forecast with lower and upper bound and the contributing factorsView on SCR-04
INT-05
Stock-out risk
+3 more tested phrasings
Horizon (default 14 days); scopeProjected days of stock cover against lead time; projected stock-out dateOpen transfer (FRM-02) or purchase draft (FRM-01)
INT-06
Inventory optimisation
+3 more tested phrasings
SKU; destination location; requirementSurplus above safety stock at candidate donors, ranked by surplus, transfer lead time and costCreate transfer request (FRM-02)
INT-07
Purchase planning
+3 more tested phrasings
Horizon; scope; budget limit (optional)Purchase requirement per section 15, with all six constraints appliedCreate purchase-order draft (FRM-01)
INT-08
Promotion analysis
+3 more tested phrasings
Period or campaign setM-10 and M-11 per campaign, ranked by incremental gross profitExport campaign review
INT-09
Scenario analysis
+3 more tested phrasings
SKU; discount or offer type; period; branchesProjected units, revenue, gross profit and post-campaign stock against the unchanged baselineSave scenario (FRM-05)
INT-10
Report preparation
+3 more tested phrasings
Period; audience (area manager); grouping (by branch)Assembly of measures, alerts and recommended actions grouped by branchGenerate and send summary (FRM-06)

Scope control

i
Matches a tested intent with all parameters presentAnswer using the six-part structure.
Matches an intent but a parameter is missing or ambiguousAsk exactly one clarifying question, offering the most likely options. Never guess silently.
Recognisable but outside the tested intentsState plainly that it cannot be answered from the available data, and offer the nearest supported intents.
Requires data the demo does not holdSay which data would be required. Never produce an estimate presented as fact.
Predictive questionReturn the forecast range and label the answer an estimate, distinct from historical fact.
Would trigger a business actionPrepare a draft and route it for approval. Never execute an order or transfer from the conversation.
The language service is slow or unavailableFall back to the deterministic narration, rendered with live figures. The fallback fixes the wording, never the numbers.

Separation of layers

i
Semantic and analytics layer

Revenue, margins, stock cover, forecast quantities, purchase-order values, constraint checks. Never wording or conversation state.

Language layer

Explains results, manages the conversation, asks clarifying questions, composes the summary. Never produces a number.

That separation is enforced mechanically, not by instruction: any figure in the model’s prose that is not already present in the computed evidence causes the whole narration to be discarded and the deterministic wording used instead. The footer on every answer says which one you are reading.

Trust safeguards

SG-01 … SG-09
SG-01Curated starter questionsLive
SG-02Tested canonical responsesLive
SG-03Visible source and calculation trailLive
SG-04Forecast ranges and assumptionsLive
SG-05Role-based approval for purchase draftsLive
SG-06"Synthetic demo data" labelLive
SG-07Demo reset with verificationLive
SG-08Pre-recorded fallback walkthroughOff-app
SG-09Clear out-of-scope responseLive

Mapping this demo onto the lead's business

i
Element in this demoRole it playsEquivalent in the lead’s world
Daily Shield SPF50Fast-moving, weekend and holiday-sensitive item.Their highest-turnover seasonal line.
Renewal Night CreamSlow-moving, high-margin item with overstock risk.Their premium slow line where cash gets tied up.
Radiance Vitamin C SerumPromotion-sensitive hero product.Their headline promoted product.
Payday campaignRecurring promotion with a predictable demand spike.Their own recurring campaign cycle.
Cleanser and Moisturiser bundleBundle that lifts basket size but cannibalises a line.Their standard multi-buy or bundle offer.
East Malaysia DCLonger lead time forcing earlier purchasing.Their most distant or slowest-supplied location.
Model version v2.4.1, trained on 24 months of daily history. All figures in this demo are synthetic. Nothing here describes a real customer or a real business result.