I asked my sister, who is a for a high tech company, to tell me one thing she hates the most about buying online. She told me—and I didn’t believe it until I watched her. She keyed in an 11-character item number into the search bar—and came back with 220 results.
Admittedly, I’ve been pretty spoiled. I have been entrenched in the search world for most of my career—first as a and then, for the last 15 years, working for search companies. I was actually pretty amazed that the companies she shopped from were offering search that was circa 1996.
Now, if my sister knew Boolean search operators, which was de rigueur for librarians and researchers in the 1990s, she could have narrowed this on her own. Boolean is a type of search that allows users to combine keywords with operators (or modifiers) such as AND, NOT and OR to further produce more relevant results. In my sister’s case, merely putting the quotations around the query “ABC-123456LM” would have narrowed the search considerably. In her case, the results went from 220 to 3—and those 3 items included one new and two refurbished units.
Here’s the thing, no one needs (or should have ) to learn Boolean anymore to search. Large consumer sites like Amazon, Walmart, Dell, and Wayfair have made it extinct.
Still, even given the antiquated search technology, it could have been configured much better. One of the reasons for the poor results is that the company was using a search engine technique called wildcard searching. This is typically done to broaden the search so that enough results (or any!) are returned. The number of results that come back is called recall—and refers to the quantity of the results. This contrasts with precision, which refers to the quality or narrowness of the search.
Noted search expert Daniel Tunkelang, PhD., who has LinkedIn, Endeca, MIT, and Carnegie Mellon in his bio, likens this to the search for truth. “A recall of 1 means the results include the whole truth, while a precision of 1 means the results include nothing but the truth.” In actuality, he writes, “neither of these is equal to 1. Tuning a B2B ecommerce search engine is a trade-off: increasing precision decreases recall, and vice versa.”
In the case of the above manufacturer: they returned everything from their B2B product search that had ABC and 123456LNM. Not a lot of truth there!
Now, in general, wildcarding is not a bad thing. But it does need to be managed. I showed my colleague and search guru Vincent Bernard the culprit site. He immediately pointed out that wildcarding in complex manufacturing is going to frustrate a potential customer. Oftentimes the buyer is not 100% certain what the correct product is. If all you have is a SKU, and no other knowledge, a list of 220 is going to confuse and frustrate you. “Plus,” he adds, “wildcards are inefficient and slow.”
He prefers a technique called SKU decomposition, which is a variant of partial matching. What you do is take the SKU and break it into character strings. This could be:
So now when a user types in some of the characters into a B2B ecommerce search, there is a match. You decide how narrow you want the recall to be. One or two characters would likely create a lot of noise, whereas a threshold of 3 or 4 might offer greater precision.
Decomposition for Type-ahead, AutoSuggest
Decomposition is a great way to offer type-ahead suggestions for your customers.
Here you can see the characters ABC could appear anywhere in the string – and accompanying categories offer a guided navigation for your user.
Another option is to add images, as in the window below.
Precision: Query Uniformization for Query Rewrites
Another big challenge with B2B product search for complex manufacturing is that product dimensions and weights can be written in numerous ways. The catalog might have 10m x 10m, but the buyer might write any of the following queries: 10 x 10, 10×10, 10mx10m. If you don’t do this already, check your query logs to see what your users are typing in.
Again, you would want to create a list of these variations to put into a thesaurus. The query pipeline transforms the user input to a better format for better precision and recall.
It may seem counterintuitive, but one of the most frustrating things to a B2B buyer is the lack of personalization. First of all, unlike a retail site, a B2B buyer must authenticate. You already know who your buyers are! Even if the company didn’t do any of the other precision-recall tuning, since she logged in, why didn’t it highlight the product that she has bought before as in the image below?
And while much has been made of the need for a personalized experience in the B2C world, Richard Isaac, CEO of RealDecoy, an ecommerce integrator, believes that personalization may be more important for B2B.
“Unlike B2C, B2B buyers don’t want to browse,” reminds Isaac. “They want to buy. And they want that experience to be as frictionless as possible.” B2B ecommerce search has to help them.
“If you have a product catalog of millions of SKUs, what are your return buyers seeing? Can they easily pick up where they left off? Are the navigation filters ‘savable’?”
If you have a million SKUs in your catalog, or even tens of thousands, doing the above tuning by hand means needing a legion of people to pore through log files, building thesauri, mapping tail queries to heads, and creating lines and lines of code.
Further, oftentimes when a user queries your site, that query gets sent out to multiple indexes. Each silo might have its own rules on what is most relevant – instead of working in concert to provide what the user wants.
Unified Index and Machine Learning
A unified index is literally one master index of all your assets, all your knowledge and all your products. This allows for homogenous ranking of B2B ecommerce search results, regardless of their sources. But, while important, that’s only half the equation. The other half is marrying that unified index with machine learning.
In B2B, while user behavior data is important, product intelligence is key. Let the machine figure out how products—and words defining those products—are grouped together. This allows you to know that a washer goes with a given fastener, or a bicycle seat with certain bikes. Or, when a person who has bought before is searching on ABC-123456LM, she wants to reorder.
Customer Lifetime Value
As you know, business buyer lifetime value is the big moneymaker for B2B businesses. In some industries, customers have a finite set of items—and will they keep ordering again and again. If you make it a pleasant experience. With Amazon Business setting the bar for great customer experiences, and acquiring B2B customers at a torrid pace, you have a lot to lose if you can’t find the truth in your precision and recall.
See how an agile index can fix your product catalog search.