The Death of Navigational Commerce: From Search Bars to AI Agents

Around Y2K, a book company started selling online, and most of the automotive industry looked at that and figured it had nothing to do with us. We were wrong, and now we laugh about it over drinks at conferences.

The real timeline is longer and messier than most people remember, and it didn‘t move in the neat stages we like to draw on slides. It started with dragging parts off green screens and out of paper catalogs into something digital. That sounds small until you remember someone had to build the taxonomy underneath it: a way to organize parts from the top category down to the terminal node that actually held up, and then convince people to trust it enough to use it. And taxonomy was only half the fight. For years “digital“ meant a part number and a price. No image, no attributes, nothing. Getting even a bare set of attributes onto a product was its own multi-year war, and for some companies it still is.

Once the catalog existed, the fight moved to the search bar. Early on you had to type the exact right term or you got nothing back, and it took years to get anywhere near natural language, where a misspelling or a rough description would still land you on the right part. But mastering search was never really about what happened inside the box. It was about whether customers used it at all. Most of the industry will tell you the same story: the customer Googles the part first, finds a number on somebody else‘s site, and only then comes back to type it into the search bar we spent a decade building.

All of that, the taxonomy and the attributes and the search bar, was happening on the surface. Underneath it, one thing never moved. Year, make, and model has always been king. It wasn‘t a phase we passed through. It‘s the foundation everything else got bolted onto, and it‘s why this industry is harder than almost any other retail category. That‘s fitment. Most retailers need a good product feed. We need a good product feed that‘s also correct for the exact vehicle sitting in someone‘s driveway.

Then, right as we were finally getting e-commerce under our belt, COVID compressed years of change into about eighteen months. And before we‘d caught our breath, AI showed up and joined the table uninvited. For a few years now most of the industry has been arguing about whether it‘s some mystical thing or something we can actually put to work.

It‘s practical, and it‘s already working, across B2B buyers, technicians, and DIYers. But “AI” is such a blanket word that saying it‘s “being used” barely means anything. When people in this industry say AI, they‘re usually picturing a ChatGPT-style chat window, something conversational and customer-facing. That‘s one thin layer of a much deeper use case. AI is cleaning and enriching product data. It‘s running the analytics that make a company data-driven instead of gut-driven. It‘s in the logistics layer, putting the right part in the right place based on demographics, regional weather, whatever actually predicts demand. It‘s in the box-size and shipping-method calls that get a part to someone fastest. A lot of what gets called AI right now is really automation, finally applied to data we‘ve had for years and never used well.

But the shift I actually want to talk about is more specific than “AI replaces search.” For decades we‘ve assumed the customer has to understand how our catalog works before they can buy from it. That assumption is what‘s about to break.

I‘ve spent 20-plus years in automotive, and more than ten on the ugly side of it: the PIM, the MDM, the ACES and PIES feeds, the fitment tables, the taxonomy everyone argues about. Across every retailer, distributor, and manufacturer I‘ve worked with, one thing has been true. We‘ve spent decades forcing customers to think like our databases. Pick the year, then the make, the model, the engine, the submodel. Filter it down. Search. Guess. Hope you got it right, and return it if you didn‘t.

We built entire UX disciplines to make that less painful. Faceted search, smarter filters, better autocomplete. Collectively, billions of dollars went into making navigation slightly less annoying. None of it was ever the fix. It was an elaborate way of managing a problem we never solved, because the catalog still made the customer do the translation.

AI agents don‘t do translation work. They just do the work.

This was never only a B2C problem

And I don‘t mean only the retail shopper standing at a website with a credit card. This is a B2B problem too: the shop tech, the broker, the distributor ordering on behalf of someone else‘s vehicle.

Our industry has a knowledge problem bigger than most people admit. We‘re losing a generation of techs and counter people who carry decades of vehicle knowledge in their heads. Not in any system, in their heads. They‘re retiring faster than we‘re replacing them, and the generation behind them didn‘t come up turning wrenches the same way, so in a lot of cases that knowledge never gets handed down.

So the real question was never “can AI help a customer.” It‘s how we take the knowledge walking out the door with a retiring tech and put it into a system good enough that a nineteen-year-old behind the counter can find the exact right part for the guy in the parking lot with a car that won‘t start, without either of them already knowing the answer.

That confidence matters just as much between businesses. A shop tech ordering from four different distributors because none of them trust their own data isn’t running into a search problem. It’s a trust problem. And when the wrong part shows up because somebody‘s manufacturer data was bad, it isn‘t just an annoying return. It‘s a car sitting on a lift, a shop losing a bay for the day, and everyone downstream redoing work that should have been right the first time.

Navigation was always a workaround

Navigation isn‘t a feature customers asked for. It‘s scaffolding we built because our systems couldn‘t understand intent. Nobody wakes up wanting to “filter by engine size.” They want the truck to stop pulling right when they brake coming downhill with a camper on the back. Filters were a workaround for catalogs that could match fields but couldn‘t understand why someone was buying.

That‘s the part agents actually change. Not the search box. The requirement that the customer speak your catalog‘s language at all.

Instead of ”I need brake pads,“ you get something closer to: I‘m running the longest off-road rally in the country, nine days, in my 2017 4Runner. What do I need to order to prep? What brake kit? What fluid? What shocks? What am I not thinking of?

A well-built agent should work backward from that sentence to the vehicle, the duty cycle, the terrain, and the budget, then pull the full list of complementary parts a sharp counter person would think to mention, and drop all of it into one cart on one site. Not a chat that hands you a list you take somewhere else to actually buy. That‘s not a better UX. It‘s a different relationship between the customer and the catalog.

Your data stops being a backend detail

Here‘s what should worry people. If the AI is doing the translation instead of the customer, your product data isn‘t a backend operational detail anymore. It‘s the product.

Incomplete attributes, sloppy taxonomy, fitment gaps, application data that disagrees from one supplier to the next. For years these were ”data quality issues.“ Annoying, survivable, because a human could route around them. Call a store. Ask a forum. Squint at a photo and guess.

An agent doesn‘t route around bad data. It either has the structured information to make the right call or it doesn‘t. When it doesn‘t, one of two things happens: you drop out of the recommendation entirely, or the agent recommends the wrong part with total confidence. There‘s no version where human intuition patches the gap.

I‘ve sat in enough scoping calls to know how many organizations still carry duplicate SKUs, missing coating materials, application year ranges that don‘t line up, and taxonomy that three departments built independently. That was tolerable friction in a navigational world. In an agentic one, it‘s the difference between winning the recommendation and never being in the running.

The winners in the next phase won‘t have the prettiest storefront, and they won‘t win because their filters are a click faster. They‘ll win because their product knowledge is complete enough for an agent to reason over it and get the answer right. Your catalog stops being infrastructure and becomes your moat. That‘s a hard pill, because catalog and data work has never been glamorous. It‘s what gets pushed behind the website redesign and the marketing campaign, the exact storefront that‘s about to matter less.

The trap on the other side is just as easy to fall into: hear all this and go build a better interface. More taxonomy, more filters, more data crammed onto the product page. That‘s not it either. It was never about having the most data visible. It‘s about having data that functions, complete and structured enough that a customer can describe a real problem or a real build and leave with everything they need, in one cart, without opening ChatGPT to figure out the list, then your site to search for it, then two more tabs to compare who ships fastest. We‘ve all just gotten used to that back-and-forth. It doesn’t have to be normal.

That‘s the version I‘m waiting for: a site where a customer lays out the whole problem, the vehicle and the event and the use case, and checks out once, from one place, because the data underneath was good enough to carry the entire conversation. I don‘t think that‘s five years out. Whoever builds it first is going to look like they cheated.

Your catalog can be built for a human to navigate, or for an agent to decide. Those used to be the same problem, solved with the same tools. They aren‘t anymore. And most catalogs I‘ve seen, beautiful front ends included, are still built for the customer we‘re about to stop having.