Pricing Agent (Profasee: Oracle)¶
Sets the price that maximizes profit at each moment. Not rule-based repricing; it weighs demand, margin, and competition. Highest write-risk agent (it changes live prices), so it ships last and most cautiously.
Data we have¶
agg.profit_daily(real per-day margin and units),agg.sales_daily_unified(demand history).forecasts.predictions(demand sensitivity context from the lifecycle model).raw.product_skus,raw.products(catalog, cost basis).
Data we likely need to add¶
- Current Buy Box price + competitor prices per ASIN (SP-API Product Pricing / competitive pricing). Not currently pulled; add a
pull_pricing.pyintoraw.pricing_daily. - Price-change history + elasticity signal. We have not run price experiments, so elasticity must be estimated from natural price variation and, later, deliberate small tests.
Recommends (for the brief)¶
- ASINs priced below profit-maximizing point (leaving margin on the table) and above it (suppressing volume).
- Competitor stockout opportunities: raise price to capture the gap (Profasee's signature demo move).
- Suggested price with expected margin/volume impact and confidence.
Actions (action mode, last to ship)¶
- Update list/Buy Box price via SP-API, within a min/max band per ASIN, through
guarded_action. - Always reversible: log prior price; support one-click revert.
Guardrails (strictest)¶
- Hard min/max price band per ASIN (never below cost-plus-floor, never above a ceiling) enforced in code.
- No-fly list for MAP-restricted or legally sensitive ASINs.
approve-eachfor a long time before any autonomy; price is the most visible thing to customers.- Small step sizes; no large jumps without approval.
Why last¶
We have the margin data but not pricing/competitor data or any elasticity history. Build the read-side competitor/price feed first, surface recommendations in the brief, and only consider writes once the recommendations have proven accurate in observe mode for weeks.