Shopping Is a Synthesis Problem

The task is harder than it looks

Here is a purchase decision that probably sounds familiar: a new laptop bag. Not a replacement for one you already love, not a commodity reorder. You've been using a beat-up backpack from college and you've decided it's time to upgrade.

You open a few browser tabs. You search on Amazon. Maybe you browse a couple of subreddits. You look at some YouTube reviews. After forty minutes you have fifteen tabs open, a vague sense of anxiety, and no bag.

The standard explanation for this experience is "too many options." But that's not quite right. Too many options is a manageable problem. Good filtering solves it. The real issue is something more specific: you are trying to optimize across multiple dimensions at once, in a space you don't know well, with criteria that aren't fully formed even in your own head. You want it to look professional but not stuffy. You care about quality but you don't want to spend more than a certain amount. You need it to fit a 15-inch laptop, but you also don't want it to feel like a laptop bag. You've read that some brands are good about sustainability and that matters to you, kind of. You need it in two weeks because you have a trip.

That is not a search problem. That is a synthesis problem.

What synthesis actually requires

When you're shopping for something meaningful, you're not just retrieving information. You're holding multiple variables in your head simultaneously and trying to find the region where they all overlap acceptably. Price is a constraint, not a preference. Quality matters, but quality for what? Style is personal. Reviews are noisy. Timing changes what options are even viable. And underneath all of it is a layer of personal context that nobody has asked you about.

The people who are good at helping with this kind of decision are good at synthesis, not retrieval. The friend who gives great gift recommendations isn't the one who knows every gift guide on the internet. It's the one who remembers that you mentioned being into cooking lately, knows your apartment is small so you won't want more clutter, and can figure out that you'd rather have one excellent thing than a set of three mediocre things. That's a synthesis operation. It requires holding facts about you, facts about the world, and some implicit model of how you make decisions, all at the same time.

Search engines are retrieval engines. They are extraordinarily good at pulling relevant documents from massive indexes based on query terms. That is genuinely hard to do at scale, and the engineering behind it is impressive. But retrieval is the easy part of shopping. The hard part is what you do with the results.

Where current tools break down

Most e-commerce experiences stop at the retrieval step and ask you to handle synthesis yourself. Here is the product. Here are 1,400 reviews. Here is the price. Good luck.

This puts the entire cognitive load of synthesis on the shopper. You have to develop your own framework for what matters. You have to figure out which reviews are trustworthy. You have to compare products across tabs manually because no tool does it for you. You have to factor in your personal constraints, your timeline, your compatibility requirements, your existing stuff. Every one of those steps is work, and it compounds. By the time you've read enough to feel confident, you're exhausted.

The tools that try to help -- comparison charts, filter systems, recommendation engines -- generally attack only one dimension at a time. A comparison chart will tell you that Product A has 30% more battery life than Product B, but it won't tell you that for your use case, battery life doesn't matter nearly as much as weight. A filter system will let you set a price ceiling, but it can't help you decide whether to set that ceiling at $80 or $120. A recommendation engine trained on purchase data will surface things that are popular, which is not the same thing as things that are right for you.

The fundamental issue is that synthesis is contextual, and context lives in the person doing the shopping. Nobody has built a good system for capturing that context at the moment it matters and using it to make the tradeoff decisions that synthesis requires.

Why this matters more than it used to

The stakes here have gotten higher as the catalog has gotten bigger. Twenty years ago, a department store carried maybe a few hundred laptop bags. The synthesis problem was real but bounded. Today, a single search on a major platform returns thousands of results, from dozens of brands, many of which you've never heard of, with varying levels of review trustworthiness. The retrieval problem got solved. The synthesis problem got harder.

At the same time, the technology to help with synthesis has improved dramatically. Large language models are, at their core, synthesis machines. They can hold many pieces of information in context simultaneously, understand how those pieces relate to each other, and reason about how they apply to a specific situation. That's not retrieval. That's something closer to judgment.

The interesting design question is how you wrap that capability in an interface that actually serves shoppers rather than just impressing engineers. A system that can synthesize across dimensions isn't useful if it can't ask the right questions to understand what those dimensions are for a given person in a given moment. Context still has to come from somewhere.

The interaction model that makes sense

Think about what a genuinely helpful shopping conversation looks like. You tell someone what you're looking for. They ask a few targeted questions: what's it for, what matters most to you, what's your timeline, what have you tried before that didn't work. Then they make a recommendation and tell you why, including what tradeoffs they made on your behalf. You can push back, refine, or ask about alternatives. The recommendation is not a list of options. It is a considered judgment.

That interaction model is achievable with current technology. It's not magic. But it requires building a system oriented around synthesis from the start, rather than bolting a conversational interface onto a retrieval system and hoping the result is useful. The architecture has to be different because the job to be done is different.

Search returns results. Synthesis helps you weigh them. Getting to the right bag, the right laptop, the right winter coat, the right gift for your brother-in-law, requires both. And for a long time, we've only really had one.

That gap is what we're trying to close.