Price tracking & automatic competitive alignment

Context

We couldn't manually watch hundreds of references against the big retailers (Darty, Boulanger, Fnac, Son-Vidéo…). The result: we lost sales by being too expensive without knowing it, and risked slashing prices by reacting blind. We needed continuous tracking and reliable realignment, margin protected, without manual watching.

What was built

An engine that watches competitor prices continuously and automatically realigns ours:

Wiser crawls competitor sites and delivers their prices several times a day.
A rules engine (in Odoo) computes, for each product in a scope, a target price: align on the best valid competitor, bounded by a minimum margin.
Two modes: observation (simulates, nothing moves) then live (applies and pushes to Shopify). Going live stays a human decision.
Everything is logged and reversible, with a message dropped on each product record.

Result: 6 product families driven live (projectors, TVs, hi-fi, soundbars, home-cinema amps), ~300 prices realigned to the day's market; the rest adjusts on its own at each Wiser update.

Technical architecture

Odoo module cobra_price_rules (depends on wiser_price_intelligence + cobra_product). Three key models: cobra.price.rule (the rule), cobra.price.alignment (the log: simulated / applied / blocked / cancelled), cobra.price.retailer (the competitors).
Event-driven engine: on each Wiser import (FTP cron every 3h), a hook evaluates only the affected products. Offers are read through a single source-agnostic access point (_get_competitor_offers) — Wiser today, an in-house scraper tomorrow, without touching the rules.
Writes via cobra_price (the source-of-truth field), hence native Shopify sync; never a direct write to Shopify.
Decoupled control: a Flask app + vanilla JS front-end (dedicated interface, dark theme, Cobra branding) reads and writes the same rules over XML-RPC, on top of a SQLite cache refreshed every 3h (systemd timer). An AI assistant (Claude, structured output) turns a natural-language instruction into a pre-filled rule.
Rule options: dynamic scope (categories / brands + exclusions + price band + stock / restocking + new products), tracked competitors, target (match / −X % / −X €), direction (decrease only / decrease + increase back), guardrails (min margin, max drop, rounding, ignore offers below 60 % of the Cobra price).

The guardrails (what makes the auto safe)

Minimum margin per rule — never sell at a loss.
Outlier offers ignored (below 60% of our price — liquidated stock, grey market).
Pair / unit normalization — compare a pair of speakers to a pair, not to a single unit.
One variant = one rule — no control conflict on the same reference.
Observation by default; the core market (floorstanding / bookshelf speakers) left in observation at first.

What we learned

Anchor a guardrail on a reliable value. The anti-outlier threshold, first computed on a reference itself dragged down by the outlier offer, no longer protected — it had to be anchored on our own price. A badly anchored guardrail is worse than none: it gives a false sense of safety.
Beware of stale data. The display cache had no scheduler (up to 18 days behind) — a correct engine but a wrong storefront. Always schedule and date the syncs.
Automation doesn't replace judgment on the core market: better to observe before aligning (competitor liquidations, pair/unit normalization that distort the comparison).

Stack

Odoo 18 (module cobra_price_rules), Wiser (competitor data), Flask + vanilla JS, SQLite (display cache), systemd timer (scheduling), Claude API (natural language → rule assistant). Dedicated interface with protected access.

What this illustrates

Separating data (Odoo = source of truth) from control (a simple UI on top) makes the tool usable day to day without opening Odoo. And a recurring principle: automate first what's mechanical and safe (margin protected, outlier offers filtered), keep the human where judgment matters (core market, going live).