Evan Schuman's StorefrontBacktalk

The potential for search engines to accurately find very specific products in local stores is powerful. But companies that have tackled this space thus far—including Google, Thefind and Milo—have discovered it’s a radically more difficult search than they’re used to. In testing those and other local inventory search engines, we have found almost no accurate search results, suggesting that the technology still has a long way to go.

All of these companies have quickly admitted their current—to be charitable—accuracy deficits. Milo executives, for example, agreed that their engine needs a lot of work. But they also stressed their efforts in improving inventory accuracy.

Although that is certainly important, it should be a Phase Two goal. If someone is asking for a “green vest to wear” and the results are showing videos, books, a glow-in-the-dark safety light and a top response of “Nerf Dart Tag Fury Fire Blasters” (actual search target and results from Milo’s engine), the fact that it may be accurate in knowing the number of Nerf Dart Tags a toy store has isn’t very helpful.

But the reasons behind the inaccuracies give some hints as to how the problems may ultimately be addressed.

One of the issues is a lack of standards. “There is no standardized model for what works and what doesn’t” in local inventory search, said Ted Dziuba, co-founder and lead software engineer for Milo.com.

What teams have tried to do is apply lessons learned from product search, but that hasn’t always helped. What might help are pulldowns or other user-selected options to clarify what is being sought.

“Our engine should understand that green means a certain product code,” said Milo CEO Jack Abraham. But by indicating that that descriptor is meant as a color, the system can be pointed in the right direction.

Beyond improving an engine’s virtual intelligence, this effort will also require chains—and especially suppliers—to get a lot more granular in product descriptions. A television will often be coded quite precisely, while a crate of stuffed animals will often be coded simply “stuffed animals” as opposed to “stuffed red giraffes.” Lower cost products often suffer from a lack of such precision.

The irony is that the smaller stores—especially ones focusing on lower cost merchandise—are the ones most likely to benefit from robust local searches. And yet they are also the ones least likely to put the dollars and effort into making their data usable.

That said, it’s the old ROI chicken-and-egg dance. How can stores justify investing in better data until the engines can handle it? And the engines can’t get any traction until they get better data.

Milo’s Dziuba said that he is focusing on the manufacturers because they are in the best position to make the changes cost-effectively. He also sees the need for a lot more work with POS data integration for more accurate inventory. “How do you know that the House of Seashells just sold one seashell? You really need to integrate with their POS. We have to make inroads from the manufacturers through to the POS” companies, he said.

In a somewhat non-intuitive argument, Dziuba said the best way to improve the accuracy of Milo’s searches is to pour in much more data. Typically, adding more data before the accuracy issue is fixed wouldn’t make sense. But Abraham is suggesting that a mass audience can make the database more accurate in a Google-like fashion. If 20,000 people search for “green car” and they go five pages into the results and click on hybrid automobiles, the system will start to learn that they probably didn’t mean green as a color.