THE RKGBLOG

Interview: Mercado’s Kevin Lindsay On Product Data Optimization

Robust on-site search is an essential component of a successful retail website. Site search apps, however, depend on good SKU-level attribute data.

Earlier this month, I had good conversation with Kevin Lindsay of Mercado about product data optimization. Kevin is Mercado’s Director of Marketing, and has been with Mercado three years. Prior, Kevin was at Verity. Kevin’s been working in search since ’96.


Alan: Hi Kevin! Thanks for the time today. For folks not familiar with Mercado, can you give us the quick summary of what your firm does?

Kevin: Mercado is 100% focused on search and merchandising capability for online retailers. We don’t do other kinds of search, like knowledge management, self help. Our focus is search for retail retail — we call it “searchandizing.”

A: Who are some of Mercado’s flagship clients?

K: We’ve been working with Macy’s for some time. Williams Sonoma, Pottery Barn, Sears. Retailers like Guess, ShopNBC, REI, and Overstock.com.

We also work with smaller retailers through a software-as-a-service model, where we can offer solid site search to retailers doing a few million in revenue: retailers like Delightfuldeliveries.com and Organize.com.

A: Most folks think of Mercado when they think of site search. Today I’d like to learn more about your product data optimizer. Why does clean product data matter?

K: Product level data quality matters to online retailers because it powers site search and because so often that data is of poor quality. Often the issue is that so many retailers of them get data from a variety of vendors, and the result is a mishmash.

Take for example, the online furniture seller Broadspan Commerce (their sites are DirectlyHome.com, TotalBedroom.com,Buying-Bar-Stools.com, and Buying-Beds.com). They drop-ship from 150 manufacturers. Their product data comes from 150 different sources, and each of these 150 partners have their own own unique way of describing products. These can be minor or major differences — one vendor may use a “CLR” attribute for color, another may use “CL”, and a third might use “COLOR”. But these differences in SKU-level data can cause site search to yield poor results.

Every site search product works essentially the same way — the software creates proprietary index top of the data, to let users search quickly. When that index is being built, it is a great opportunity to add customization to the data. By cleaning and adding data attributes to the SKUs, the retailer can create their own views into the data, tuning the search to behave as consumers expect.

One of our clients is Cheapvitamins.com. Like many retailers, they’re doing everything they can to increase conversion and average order. With their old search, there was no way to search on vitamin type, or by flavor, or by number of capsules in a bottle. Because the SKU level data was inconsistent, shoppers couldn’t find products the way they wanted.

A: By flavor?

K: Sure, many of their supplements come in flavors. With their old search, visitors couldn’t reach all their chocolate shakes, for example, or narrow down an existing search to vanilla-flavored products. They rolled out improved search with us at the same time they launched a redesign. Getting all the product data which supports site search “fixed” helped them increase conversion on the site.

A: So how does your product data optimizer tool help retailers clean their data? Can you give a concrete example?

K: We sell the optimizer as software, or as a one-time service. For example, OfficeMax licensed our technology, where as Cheapvitamins.com used our products as single professional services engagement.

When we start working with a retailer, we say, “Show us your data, what are your top 10 attributes, and why are the problematic?”

The retailer will say, color, brand, size — whatever attributes capture how their visitors search for their products. Retailers can use their web analytics to see how people search, the most important attributes, and what searches aren’t currently working.

Then, we essentially create rules: this words means the same as that word, in this context. The software handles associations and stemming.

Going back to CheapVitamins, until we set a flavor association, the computer didn’t “know” that “strawberry” or “orange” or “peppermint” were flavors. Those words were in the product description, but that was mishmash of terms and words and didn’t allow them to present SKUs by flavor. Teaching the computer about flavors lets us clean up the SKU data to support better search, like for this retailer, search by flavor.

Our software is a classification system. Retailers can set up rules to determine which bins products end up. And products can end up in multiple categories. In a physical grocery store, you might be uncertain if salsa is shelved with the sauces, or in the Mexican specialty aisle. In a web store, salsa can be in both places at once, making it easier on the shopper. And this sort of thing really helps conversion.

A: When cleaning up SKU data, how much is automated, and how much is done by hand?

K: Initially, alot is manual. Particularly with messy data, at first, manual intervention essential. We have to build a rule to make sure each product ends up in right buckets. In other cases, we can do what we call “retrospective search” — setting up multiple different queries leading to same concept.

Here’s another CheapVitamins example. Suppose someone is looking for vitamin kids would like. We’ve mapped “children flavors” so it points to tastes kids typically like: strawberry, bubble gum, water melon, etc. In the set up, we make manual associations to teach the software that “bubble gum” is a kids flavor. The retailers know their business and how their visitors are trying to find products. And once these rules are set up, at indexing time, each incremental index or each daily index, those rules are applied to data going forward.

A: You mentioned classification. Is that classification to the merchant’s taxonomy, or to a industry generic taxonomy?

K: Each merchant has own specific taxonomy. One merchant might care about “best sellers”, another about “gift for dads”, and so on.

We also have our own domain dictionaries. As Mercado’s been at this over the 10 years, we’ve learned about good taxonomies for consumer electronics, for clothing, and so on.

For example, do a search on Macy’s for “blue dress” — you could get dresses colored “lapis”, “admiral”, or “azure” — we know these are all shades of blue. Another client is OnlineShoes — they have a huge domain dictionary, you can’t imagine the number of different ways manufacturers and shoppers say “brown“.

A: Retailers have a tough time populating feeds correctly for the shopping engines. Can your data scrubbing approach help with CSE feeds?

K: We haven’t had clients using these data enhancement tools to improve off-site feeds, but we starting to have conversations about that. I will say that these data scrubbing exercises can provide better organization and help with SEO.

A: Leaving the product data optimizer and now focusing on online retail in general: what are the 3 most important tips for improving online sales and profits you’d suggest to online retailers?

K: Hmm. OK, here are Kevin’s Top Three Tips for Online Retail:

Tip #1: Metrics-driven merchandising.

There’s too much intuition driving merchandising today. On the web there’s a huge opportunity to have data and metrics drive merchandising decisions in real time. Automated merchandising driven by the numbers — that’s a big idea.

Tip #2. Understand the relationship between web search and site search.

Don’t keep your paid search, organic search, and site search thinking separate. They’re not different silos; they’re all different ways people are looking for the same stuff, the products you sell. Insights from site search logs can really help with SEM. And strong site search provides high-converting targeting landing pages for paid traffic landing pages.

Tip #3. Clean your product data.

Not to sound self-serving, but it really all starts with good product data.
If product data is coming in from various vendors and is inconsistent, your site search won’t work as effectively, and you have a huge handicap out of the gate. I recommend all retailers put their product catalog through an “data physical fitness program” (!) Cleaning up your data really lets you use search to sell, our “searchandising” concept.

A: Good tips. Wrapping up, any little known fact about Mercado you can share?

K: Well, oddly enough, almost everyone on our management team is a rock n’ roll musician. Our CEO plays drums, and used to run discos before getting into search. Our VP of Sales plays sax. Our VP of Professional Services plays guitar. And I sing a bit.

A: Interesting! And thanks, Kevin, for your insights about cleaning up product data.

Technorati Tags: , , , , , , , , , ,

  • Alan Rimm-Kaufman
    Alan Rimm-Kaufman founded the Rimm-Kaufman Group...
  • Comments are closed.