Behind the Scenes: Building FashionBid's AI Negotiation Engine

Letting shoppers name their own price sounds simple until you have to build it. Behind FashionBid's "make an offer" button sits a negotiation engine that has to be fast, fair, and impossible to game — all while feeling like a friendly back-and-forth rather than a wall of rules. Here's an honest look at how we built it and, more importantly, the guardrails that keep every offer fair.

If you haven't seen the shopper's side yet, read How FashionBid's AI Price Negotiation Works first — this is the view from under the hood.

The problem we set out to solve

Fashion produces a constant stream of overstock and end-of-line pieces. The traditional fix — blunt, scheduled markdowns — is wasteful and trains everyone to wait. We wanted something better: a way for each item to find a fair price through a quick conversation, instantly, with thousands of shoppers at once. That's a job a human can't do at scale, and a rigid discount code can't do with nuance. It needed to be an AI.

What the engine actually does

At its core, the engine takes an offer and decides how to respond. To do that well, it has to hold a few things in mind at the same time:

• The item. What the product is, and where it sits relative to its value.
• A fair floor. The lowest sensible price for that item — a hard limit the engine never crosses.
• The shopper's offer. How close it is to fair value, which shapes whether the reply is an acceptance or a counter.
• The tone. Responses should read like a person being reasonable, not a machine stonewalling.

The output is one of two things: an acceptance, or a counter-offer that nudges toward a price you're both likely to be happy with.

The guardrails that keep it fair

An AI that negotiates money needs firm boundaries. The ones we care most about:

A real floor, always respected. The engine cannot sell below a fair minimum, no matter how the conversation goes — so there's no "trick" that breaks it, and no race to the bottom. No dark patterns. It isn't designed to wear you down, create fake urgency, or punish you for a low opening offer; a cheeky offer just gets a gentle counter. Consistency. The same item and the same offer get treated the same way, so nobody is quietly charged more for being less persistent. You stay in control. You can accept, counter, or walk away at any point with no penalty.

Why fast matters as much as fair

A negotiation that takes ten seconds to think feels broken. So the engine is built to reply in seconds — the first response never waits on a human. That speed is what makes naming your price feel like a normal part of checkout rather than a support ticket. It's also what lets the system handle many shoppers at once without anyone queuing.

What we deliberately kept simple

It would be easy to over-engineer this into a game with levels and tricks. We didn't want that. The best version is the one a first-time shopper understands instantly: make an offer, get a fair response, agree, check out. The intelligence lives in the pricing judgement, not in making you learn a system. That restraint is part of the design, not a missing feature.

Where it's headed

The engine improves as we learn which responses feel fair and which fall flat. The direction is the same one we see for the whole category — covered in The Future of AI Shopping and Why Fixed Pricing Is Becoming Obsolete — toward pricing that fits the person and the moment, instead of a single fixed tag.

Frequently asked questions

Is a real AI making the decision, or is it just rules?

It's genuinely AI-driven, weighing the item, a fair floor and your offer to craft a response — not a simple if-this-then-that discount. The floor itself is a hard rule the AI works within.

Can shoppers trick the engine into selling below cost?

No. The fair floor is a hard limit the engine cannot cross, regardless of how the conversation is phrased. There's no prompt or trick that unlocks a below-floor price.

Does the AI charge different people different prices unfairly?

The same item and the same offer are treated consistently. Negotiation gives everyone the same tool, rather than rewarding only the most persistent or penalising the least.

See the engine in action

The best way to understand it is to use it. Make an offer and watch how the engine responds.

Start an offer →