About
From an operating problem to a field of research.
Elias Sultan is an ecommerce operator and software developer working on marketplace decision systems. He trades on Amazon in Spain and builds the technology that runs that operation, which places him on both sides of the problem he studies: the person making the commercial decision, and the person writing the system that makes it.
His field is the layer between market signal and commercial action — how competitive behaviour, Buy Box conditions, delivery promise, real cost structure, inventory position and business objectives combine into a single decision.
The problem
Automation that changes prices without deciding anything.
Conventional repricers are reactive. They detect a competing offer and move a price towards a rule, usually downwards. The rule has no view of what the move costs, whether the position can be held, or how the competitor will respond.
Operating a catalogue under those systems produces a recognisable pattern: margin erodes, competitors chase each other to the floor, and the seller who automates most aggressively is not the one who ends up better off.
The observation that started the research was simple. Two sellers running automated systems against each other do not converge on a fair price; they converge on the lowest price either system is permitted to accept, regardless of whether that price serves either business.
That is not a pricing problem. It is a decision problem that had been delegated to a rule.
The research
Studying the market as behaviour, not as data.
The method has been observational and continuous. Competing offers are captured directly and repeatedly from the marketplace, at the level of what a buyer actually sees: who holds the Buy Box, at what price, with what delivery promise, from which fulfilment channel.
From that record, competitors stop being anonymous prices and become identifiable actors with characteristics — how quickly they react, how far they are willing to fall, whether they defend a position or abandon it, when they are out of stock.
The findings that matter have generally come from the same place: measuring what actually happened after a decision, rather than what a model predicted would happen.
The methodology
Research that has to survive contact with a real P&L.
Every hypothesis is applied to a live catalogue and judged on outcomes that cannot be argued with: units sold, margin realised, position held or lost, cash converted.
This imposes a discipline that simulation does not. A model that looks elegant and loses money is discarded. A finding that survives months of live trading, across seasonality and competitor turnover, is documented and built into the system.
The work is recorded as it happens — the problem observed, the change made, the module affected, the measured effect — so that the evolution of the system is documented rather than reconstructed afterwards.
Timeline
Continuity of the work.
A chronological record of the trajectory from operator to researcher and developer.
Start in ecommerceBeginning of marketplace trading operations and first direct exposure to automated price competition.
First researchSystematic observation of competitor behaviour and of how automated systems interact with one another.
Decision modelsFirst structured models of the decision behind a price change: cost floor, position value, expected response.
First intelligent enginesEarly engines that evaluate context before acting, rather than applying a fixed rule.
Repricer researchBehavioural analysis of competing automated repricers: reaction latency, aggression, recovery patterns.
Research seriesConsolidation of the findings into a documented series covering Buy Box probability, fee intelligence and competitive strategy.
SellerFlowThe research consolidated into a full decision platform operating a live catalogue.
Enquiries
Interested in the underlying research?
The research index documents the individual studies; white papers cover the framework in full.
Research index