B2B buyers are sophisticated, so why does so much conventional wisdom about onsite search still treat B2B like a smaller, simpler version of B2C? In a recent conversation on the eCommerce Edge Podcast, Coveo’s Chief Product Officer Peter Curran pushed back on several assumptions that come up again and again in B2B ecommerce, ideas that sound reasonable on the surface but don’t hold up once you look at how B2B buyers actually search and buy.
Here are eight of those common assumptions worth re-examining.
Myth 1: Our customers know exactly what they’re looking for, so search doesn’t matter much
This is one of the most common things you’ll hear in a B2B org, usually from someone who’s never had to do the job themselves. The idea is that B2B buyers are repeat customers who already know the SKU, the part number, the model, so a “fancy” search experience is a B2C nicety that doesn’t apply.
In reality, the fact that someone can memorize a SKU doesn’t mean they should have to. Plenty of people on the other end of that search bar aren’t repeat experts at all. New hires, occasional buyers, and people searching on behalf of someone else (a technician calling in from a job site, for example) often don’t have the number memorized. When search can’t handle anything but an exact match, all of that traffic either fails outright or gets pushed onto a phone call to a rep, which feels like “search doesn’t matter” right up until you count the cost of those calls.
There’s also a generational dimension to this. The next wave of buyers and field technicians grew up with consumer search and AI assistants that understand vague, conversational queries. Asking them to memorize a 14-digit part number with a hyphen in a specific place isn’t a quirky industry tradition but a UX gap that’s about to get a lot more visible as expectations shift.
As Curran puts it, the cognitive load of memorizing part numbers is a real cost to the customer, even if it’s invisible to the business, and it’s a cost companies have simply become accustomed to externalizing onto their buyers.
The data backs this up, too: B2B companies that have invested in better search are seeing this play out directly. ADI, for example, a major B2B distributor, saw conversion rates increase by 16% after improving how customers could find products through search, alongside a 28% increase in revenue tied to onsite search.
Search isn’t a side feature for high-intent buyers, it’s often the main path to conversion, whether they’ve memorized the SKU or not
Myth 2: A zero-results page is a minor UX issue
Zero-results pages tend to get filed under “things to clean up eventually,” because each individual instance looks like a small thing; a typo, a synonym the index didn’t catch, a product listed under a slightly different name than the customer expected.
But in B2B, a zero-results page is often the moment a customer who needed a part right now either gives up, calls a competitor, or falls back to a phone call with a rep instead of buying online. And the triggers are often mundane: as Curran points out, something as simple as a new team member not knowing to drop the hyphen between the alphabetical and numeric parts of a SKU, because the person who knew that just retired, is enough to generate a dead end for every search that follows.
Part of why this myth persists is that expectations have shifted faster than most B2B sites have. As Curran notes, years of consumer search have trained everyone, including your B2B customers, to expect instantaneous, forgiving results from any search box. Google never slowed down, and it never made you guess the exact format of your query. A zero-results page that might have been shrugged off a decade ago now reads as the site being broken, full stop. The bar has moved, whether the site has or not.
Myth 3: Visual search and AI-driven part identification are still years away
It’s easy to assume that the idea of “take a photo of a broken part and have the system tell you what it is and what fits it” is a flashy demo concept. Interesting, but not something to plan around yet.
In practice, this kind of capability is already working today, and people are using it. Anyone who’s uploaded a photo to a consumer AI tool and gotten back the make, model, and even likely aftermarket parts on a piece of equipment has seen this firsthand — and the same underlying capability applies just as well to a steaming, broken piece of industrial equipment as it does to a used motorcycle listing. The technology gap isn’t the bottleneck anymore. The gap is connecting that visual identification to the back-end data — knowing not just what the part is, but what’s compatible with it, what’s in stock, and what else (a mounting bracket, a filter, a gasket) typically needs replacing alongside it.
That’s an important distinction, because it changes where the real work is. Companies that treat this as “an AI feature we’ll add later” are underestimating how close it already is — and underestimating how much of the heavy lifting is actually about getting product and compatibility data into a shape where a visual match can be turned into a useful answer, not just a label.
Myth 4: Search, merchandising, and recommendations are separate problems with separate tools
This one isn’t really a myth so much as a historical accident that calcified into “how things are done.” Search, merchandising, and personalization grew up as separate product categories, built by separate vendors, often bought separately by separate teams — and for a long time, that was just the landscape.
But treating them as separate problems means they’re working with separate (and often incompatible) signals about what customers are doing. A product a customer keeps adding to a comparison list, abandoning in their cart, or searching for repeatedly are all strong signals — but if your search engine, your merchandising rules, and your recommendation engine can’t share that signal, each one is operating with partial information. The result is often three systems that each work “fine” in isolation but never quite add up to a coherent experience.
Curran goes further, arguing that the era of buying search, recommendations, and merchandising as separate point solutions is effectively over — and that the companies still operating that way are carrying integration costs that a unified platform simply doesn’t have.
Relevant reading: B2B Field Guide
Myth 5: AI means we don’t need to worry about structured product data anymore
This is a newer myth, but it’s spreading fast, the idea that since AI is so good at handling messy, unstructured information, B2B companies can finally skip the years of PIM cleanup and data governance work that’s always been a headache.
The reality is closer to the opposite. AI is excellent at working with unstructured content such manuals, spec sheets, and support docs, and can absolutely make that content far more useful than it’s ever been (this is exactly what powers the kind of conversational, back-and-forth product Q&A that’s becoming table stakes). But B2B purchases often hinge on things AI can’t infer from context: exact pricing under a specific contract, real-time inventory, compatibility between parts, regulatory specs that have to be exactly right. If a customer asks an AI-powered search experience for “the right replacement part,” and the underlying data doesn’t actually know which part fits which configuration, AI doesn’t fix that; it just answers confidently anyway, which is arguably worse.
There’s also a subtler version of this problem: even when content is correctly written, if it’s not clearly tied to when it applies, i.e. which product version, which year, which region, AI can confidently retrieve the wrong-but-plausible answer. Structured data isn’t going away; if anything, it’s becoming more important as the thing that keeps AI-powered experiences accurate rather than just fluent.
Curran describes this as the current reality for most companies: AI today is largely being used to clean, normalize, and map data into existing systems — useful, but a long way from removing the need for structured product data altogether.
Relevant reading: The AI Agent Readiness Checklist for Ecommerce
Myth 6: Merchandising is a tactical, lower-priority job, mostly about fixing what search gets wrong
In a lot of B2B organizations, merchandising has a reputation problem. It’s seen as the Friday-afternoon task, adding a synonym here, setting up a redirect there, nudging a product up a category page because search couldn’t surface it on its own. Important in the sense that someone has to do it, but not exactly viewed as strategic work, and often handed to whoever has spare time rather than treated as a dedicated discipline.
That framing made sense when merchandising’s main job really was patching the gaps in an imperfect search engine. But as search and merchandising systems become more unified and AI-assisted, the job is shifting toward something closer to a strategist working alongside an analyst. It means reviewing what’s happening across thousands of searches, spotting patterns a person never would have caught manually, and deciding how to respond. Done well, it’s one of the more direct levers a company has on revenue, because it sits right at the intersection of what customers want and what gets shown to them.
B2B enterprises that keep treating merchandising as a low-priority, fix-it-when-broken function are likely both under-resourcing it and missing out on one of the more strategic, high-leverage roles in their ecommerce organization, particularly as the tools available to merchandisers get dramatically more capable.
Myth 7: If our search were really a problem, someone would have complained by now
This is less a myth about search itself and more a myth about how problems surface inside large organizations, and it’s one of the most persistent, because it’s rarely tested directly.
Curran notes that in his experience, virtually every ecommerce company believes its product data is bad and while that might sound like everyone’s exaggerating, it’s often closer to an accurate, if uncomfortable, starting point than a complaint.
The reason complaints don’t materialize isn’t that the problems don’t exist is that they are absorbed before they become visible. Customers who can’t find what they need don’t usually file a ticket; they call their rep, or they go elsewhere, and neither outcome gets logged as a search issue. Internally, the gaps get papered over by people who’ve quietly learned the workarounds: the team member who knows not to use a hyphen in the SKU, the rep who always builds a custom spreadsheet for a particular account, the product manager who manually fixes a category every Friday afternoon. The absence of complaints is often evidence that the cost has been distributed somewhere harder to see rather than the problem isn’t there.
It’s also worth keeping the bigger picture in mind. As Curran points out, while roughly 98% of retailers now offer ecommerce, only 40-60% of B2B companies do, depending on category and vertical. The “no complaints” comfort a lot of B2B leaders feel often comes from comparing themselves to peers who are equally behind, not from any real evidence that the experience itself is working.
Myth 8: AI is going to make the people who currently hold all this product knowledge less important
As AI gets better at answering questions and surfacing information, it’s tempting to assume that the long-tenured experts who currently know where everything lives, which spec sheet goes with which product, which substitutions are acceptable, which customers need what, are becoming less essential. If AI can eventually know all of that too, why invest in capturing or relying on what’s in any one person’s head?
Curran’s framing is closer to the opposite. The risk isn’t that this expertise becomes obsolete but that it stays trapped in one person’s head, accessible only through that person, indefinitely. The goal isn’t fewer experts or less reliance on deep product knowledge; it’s turning knowledge that currently exists in one place into something the rest of the organization and AI-powered tools can actually draw on. As he puts it, the aim is to help that expert go farther, not to need fewer of them.
Treated this way, AI isn’t a replacement for institutional knowledge, it’s the thing that finally makes that knowledge durable, instead of leaving it one retirement, role change, or bad day away from walking out the door.
None of these myths are really about search being “bad.” They’re about assumptions that made sense in a simpler era of B2B ecommerce and quietly stopped being true as catalogs, customer expectations, and the technology itself moved on.
Where to Start
If a few of these sound familiar, the good news is that none of them require a massive overhaul to begin addressing. A few starting points:
Measure your zero-results rate. Most teams have never actually looked at this number, and it’s often higher than expected. Even a rough audit of what percentage of searches return nothing, and what are people searching for when that happens, tends to surface quick wins (missing synonyms, mismatched product names) alongside bigger structural issues.
Map who currently “fixes” search, merchandising, and recommendations and how. If three different teams are managing three different tools with three different sets of rules, that’s worth knowing before deciding whether to invest in better tooling for each, or in bringing them together.
Take stock of what your product data can actually support today. Not as a multi-year cleanup project, but as a simple gut-check: if a customer asked a conversational, AI-powered question about product availability, compatibility, pricing under their contract, could your current data answer it accurately? The gaps that surface are usually the most useful starting point for prioritization.
Identify where critical product knowledge currently lives in people, not systems. A quick, informal exercise asking who you would struggle to replace, and what do they know that isn’t written down anywhere, often surfaces the same handful of names across multiple teams, and gives you a concrete list of what to prioritize capturing first.
None of these require a big commitment to act on, but they tend to turn “I think our search could be better” into something specific enough to actually do something about.