Research index

Summary

A study of how competing automated repricers behave when observed continuously: how quickly they respond to a change, how far they follow, and how they recover.

Objective

To establish whether competing automated systems can be characterised as individual actors with stable, measurable behaviour, rather than treated as an anonymous market price.

Findings

• Reaction is not instantaneous. Competing systems respond within a measurable interval, and that interval is consistent enough per competitor to be treated as a characteristic.
• Competitors differ systematically in how far they will follow a price down, and where they stop.
• Two automated systems left to react to each other converge towards the lower of their two floors, independently of what either business needs to earn.
• Reaction latency is exploitable: the interval between an action and a competitor's response is itself a resource.

Practical application

Competitor profiles are maintained per identified seller and used by the decision engine to choose the size and timing of a move, rather than applying a fixed increment.

Summary

Modelling the Buy Box as a conditional probability rather than a binary outcome of being cheapest.

Objective

To identify which conditions actually govern the featured offer, and to quantify what each one is worth in price terms.

Findings

• Price alone does not determine the outcome. Delivery promise and fulfilment channel change the price at which a position can be held.
• The value of the position is not constant across a catalogue: for some products the featured offer carries most of the demand, for others it carries much less.
• Because the conditions are observable, the price required to hold a position can be estimated per product instead of discovered by lowering until it works.
• Holding the position at any price is frequently worse than not holding it.

Practical application

The engine estimates the price required to take the featured offer for each product, and compares that price against what the position is worth before acting.

Summary

An examination of competition against structurally different sellers — in particular the retail arm of the marketplace itself — and of when that competition should be declined.

Objective

To determine which competitive situations are winnable and which consume margin without any achievable outcome.

Findings

• Competitors with different cost structures and objectives cannot be treated as equivalent by a pricing rule.
• Some positions are structurally unwinnable; automated pursuit of them transfers margin without changing the outcome.
• Withdrawal is a legitimate decision, and needs to be an explicit output of the system rather than a failure state.
• Stock position — the competitor's as well as one's own — changes which strategy is correct more than price does.

Practical application

The engine classifies the competitive situation before choosing a strategy, and is permitted to decline a contest and hold margin instead of pursuing it.

Summary

Reconciliation of estimated marketplace fees against settled financial events, to establish the true unit economics behind each decision.

Objective

To replace estimated costs with the amounts actually charged, so that pricing decisions are made against real margin rather than assumed margin.

Findings

• Estimated fees and settled fees diverge, and the divergence is not uniform across a catalogue.
• Decisions taken against estimated costs can be profitable in the model and unprofitable in the settlement.
• The cost floor is not a static input: it moves with fulfilment channel, dimensions, returns and promotional participation.
• Once real settled costs are attached per unit, a measurable share of a catalogue is found to be operating below its true floor.

Practical application

Settled transaction data is reconciled per unit and used as the cost basis for pricing limits, so that floors reflect what was actually charged.

Summary

The consolidating study: a framework in which price is an output of a decision that weighs position, competitor behaviour, cost, inventory and business objective.

Objective

To formalise the decision layer that conventional automation omits, and to define how a system should reason before it acts.

Findings

• Optimising margin alone degrades cash conversion; optimising volume alone degrades margin. The decision has to hold both.
• The correct action is frequently to do nothing, and a system without an explicit hold state will over-trade.
• Decisions require evidence of the current state, not inference from the last action taken.
• A decision system must be able to explain the reason for an action, or it cannot be operated with confidence at catalogue scale.

Practical application

The framework is the specification that SellerFlow implements: signal, interpretation, constraint, decision, and only then price.