Designing the rules

A repricing strategy is a small number of rules applied to segments, plus the guardrails that keep those rules from doing damage when the market behaves strangely. Most bad strategies are one rule applied to everything.

Segment before writing a single rule

The same rule cannot serve a private-label SKU with no competition and a wholesale SKU with fourteen sellers on it. Segment on what changes the correct behaviour:

A useful first pass is competitive structure crossed with margin headroom. That alone usually splits a catalog into four groups needing genuinely different treatment, and it is computable from data you already pulled.

Rule archetypes

Seek-high is the archetype sellers most often lack. A repricer that only moves down converts a competitive advantage into a discount. If you hold the Featured Offer without contest, the current price is a floor on what the market will bear, not a ceiling.

Guardrails

Rules decide direction. Guardrails decide how much damage a wrong decision can do.

• Floor, computed per SKU, never catalog-wide. See 02-floor-and-margin.md.
• Ceiling, to keep a follow-the-competitor rule from walking a listing into a fair-pricing problem.
• Maximum change per cycle, so a bad data point cannot move a price 40%.
• Cooldown between changes on the same SKU, which is the main defence against oscillation.
• Minimum increment, so you never undercut by an amount too small to change a buyer's decision but large enough to trigger the competitor's repricer.
• Exclusions, for SKUs under promotion, subject to MAP-style agreements, or in a launch period where price is a strategy rather than a lever.

The race to the bottom, and how it actually starts

Two automated repricers set to undercut the lowest offer will drive each other to their floors within hours, and both sellers end up at minimum margin holding the same share they started with. It is the defining failure of naive repricing.

Three things prevent it, and they work together:

1. A real floor, so the descent stops somewhere survivable.
2. A cooldown, so you are not responding to your own move reflected back.
3. A comparable-offers filter, so you are not undercutting an offer that was never going to win anyway. This is the one most often missing.

Filter out, before choosing a target: used and refurbished offers when you sell new, offers that cannot currently fulfil, sellers whose delivery promise is far worse than yours, and offers whose landed price differs from their item price in a way that makes them uncompetitive. What remains is the set worth reacting to, and it is usually much smaller than the raw offer list.

Cadence

Match refresh rate to what the SKU is worth. A crowded, high-velocity SKU may justify event-driven reaction through offer-change notifications. A long-tail SKU selling occasionally does not justify any automated reaction at all, and polling it frequently spends rate limit that a valuable SKU needs.

Tier the catalog explicitly. Rate limits and batch sizes make "reprice everything constantly" impossible above a modest catalog size, so the tiering is forced whether or not you plan it.

Time-of-day and day-of-week rules are popular and mostly noise. The cases that hold up are structural rather than clock-based: stock-out risk, a known seasonal demand shift, or a competitor who reliably goes out of stock.

Anti-patterns

• One floor for the whole catalog, or a floor set as a round number.
• Undercutting by a percentage. Increments should be small and absolute.
• Reacting to every offer in the list rather than the comparable ones.
• Repricing on ASINs where you are the only seller.
• Measuring success as Featured Offer share alone, with no contribution figure beside it.
• Matching an FBA competitor's price from a merchant-fulfilled offer and expecting to win.