Data Analytics · Read + Write

Pricing from internal + external data

Augment Rep · High Reason · High

The problem

Pricing managers update lists weekly across thousands of SKUs, eyeballing competitor prices, cost moves, and slow-movers. Mistakes leave margin on the table or break the floor.

The AI approach

Internal × external × elasticity. Join cost + sales velocity + competitor scrape; estimate elasticity; recommend a band that respects floors/ceilings; LLM narrates the rationale.

The outcome

~+1.4 pt margin

Faster repricing. Margin lift ~+1.4 pts in pilot. Manager spends time on outliers, not the long tail.

Try itInput → Process → Output

Input — internal + external
Internal · ERP + sales
SKUcostpriceunits/wkdays stock
CL-1178212941214
External · competitor scrape (3 sources)
competitortheir pricein-stockscraped
Comp-A1252h ago
Comp-B1191h ago
Comp-C132low3h ago
+ promo calendar · seasonality · price floors
Process — AI pipeline
1Estimate elasticityReadPredictive
2Solve under floor / ceilingRulesSymbolic
3Forecast units at new priceReadPredictive
4Narrate the rationaleWriteGenerative
Output — recommendation (manager approves)
Click Run demo to compute the recommended price band.
SKU CL-117 · suggested re-price weekly cycle
curr ฿129
rec ฿122
comp avg ฿125
floor ฿104ceiling ฿138
elasticity (last 12w)−1.8
competitor avg฿125
recommended price฿122+5.4% units → +1.7% margin
Why ฿122: Sit just under the cluster average to capture lift while staying above floor. Elasticity of −1.8 implies +5.4% unit lift covers the ฿7 cut. Margin still 33%; days-of-stock unchanged.

Three AI types in this use case

SymbolicFloor / ceiling enforcement; promo-overlap rules; constraint solver.
PredictiveLog-linear elasticity per cluster from 12+ weeks of history; demand forecast at the new price.
GenerativeLLM narrates the rationale (“sit just under cluster avg”) so the manager can review.

The stack

  • Internal · ERP cost + sales by SKU/store/week
  • External · competitor scrape (legal/ToS-checked)
  • Models · log-linear elasticity per cluster + uplift
  • Solver · constraint-aware (floor/ceiling/promo)

When this works

  • SKU has enough sales history for elasticity (≥ 12 weeks)
  • Competitor prices are actually comparable (same pack size)
  • Floors / ceilings are codified, not in someone's head
  • Manager retains final approval — augment, not auto

When it fails

  • New SKU, no history → elasticity is a guess
  • Competitor scrape goes stale → recommendation drifts
  • Promo overlap not encoded → recs collide with deals
  • Auto-publish without manager → race to the floor