DIY site search is no weekend project
Building a site search engine is no simple task. Those who go the do-it-yourself route need to consider getting the data into the search platform, configuring the engine, maintaining and optimizing the system and handling merchandising and analytics.
There is no one-size-fits-all answer to the challenge of getting site search right.
As a long-time digital strategist, Klein has advised businesses at all stages of digital development and all levels of expertise when it comes to digital experience. He understands the challenges they face — whether they are rich in resources or striving to reach the top of their given vertical.
“Building a site search engine that ultimately delivers the result that customers expect is hard,” Klein said during a recent BloomReach-hosted webinar on site search. “Customers are interacting with sites and brands all day long — they’ve been constantly improving customer experience and providing innovative features that after a time simply become the baseline expectations of those same customers. It’s the classic Amazon- or Google-expectation problem.”
He’s seen large enterprises that opted to build their own site search on top of powerful tools — in part, he said, because at the dawn of the digital age there were few other choices. And while their efforts were admirable — and maybe even heroic — it was a lot of work.
“While these are extremely powerful platforms,” Klein said of search foundations like Solr, Endeca and others, “they require a lot of initial work to configure properly for a particular business and their customers — and typically need more ongoing support to properly maintain and optimize than many companies realize.”
While the webinar was a dive into the ins and outs of creating site search from scratch, Klein’s comment pretty much summed it up. It’s not that do-it-yourself site search is impossible. Or that it’s never been done. Or even that it’s never been successful.
It’s just that not everyone who gets into it fully understands just what it is they are getting into.
Klein was joined by BloomReach Principal Engineer Romil Shah for the webinar, “Stagnant site search? Get moving again: 5 reasons why you and DIY search are not meant to be.” Shah and Klein, whose LiveArea is a PFS agency, took attendees through the five key considerations for anyone considering going it alone on site search.
Here are the highlights, complete with video clips from the webinar itself:
Getting the data into the platform: If a car runs on gasoline, a search engine runs on data. “At the end of the day,” Klein said, “we see a lot of work needed to get the right catalog and product information into the system.”
And that’s just for starters. In order to providing meaningful and relevant search recommendations, the engine must also consume commerce and historical data, including top-selling products, inventory and pricing. And the search engine needs to be fed with behavioral data — data that answers questions such as: What did a customer just look at or search for? What was the customer interested in during previous visits?
And beyond all that data, Shah noted, is the vast amount of data that can be mined from the web beyond the business’ website. That enormous trove of information is a crucial pillar of the machine-learning system that relies on natural language processing, semantic understanding and synonym recognition to serve consumers relevant results.
Configuring the search engine: Once a DIY search team has gotten the data into the system, it must design methods for doing something with the data. The default settings of an out-of-the-box search solution are not ideal for every enterprise. They need to be configured for a particular business and for that business’ particular customers.
Companies with multi-product catalogs and a multitude of brands can find it very tough to get a DIY search engine to return relevant results. Choosing the right attributes and limiting the number of indexed attributes is crucial to reduce the number of results returned for a search. But the effort becomes more difficult, the bigger the catalog and the larger the number of attributes, Klein said.
And what about keeping up with the way consumers describe things? High-quality search requires a high-quality and large — very large — synonym list. Not to mention that new words are introduced into pop culture and the language every day.
Maintenance and optimization: Set it and forget it? Uh, no. Keeping on top of DIY search requires that a business first set up a strategy and program for constantly reviewing search performance in detail and outlining ways unsatisfactory search will be improved.
And yes, maintenance and optimization costs money and requires people. Klein said he’s seen operations with limited budgets run into problems keeping up with their search. Even in cases where an individual has been given responsibility to oversee search, it sometimes becomes part of a bigger job or a second priority, he said.
Shah added that putting a human alone in charge of maintaining optimum site search can be maddening. The barometer for success is generally a performance metric, say revenue per visit. So many things affect consumer spending per visit — price elasticity, seasonal changes, trends — that relying on the intuition and experience is insufficient.
Merchandising and analytics: Klein said that fundamental merchandising analytics can suffer with a DIY system. The ability to maximize revenue or other desirable metrics through boosting and burying products on a page, managing facets and A/B testing are often hindered. They sometimes require a lot of manual work and third-party tools.
Other day-to-day tasks, Klein said, can cause unforeseen problems when sophisticated analytics are not in place. Launching new products might come with keywords that impact site search results; adding a new category can add a new set of terms that interferes with existing search strategy on the site.
Technology and performance: Once a company pursuing a DIY approach has gathered all the necessary data, configured the search engine, come up with an ongoing strategy for maintaining and optimizing search and found the tools it needs to do proper merchandising and analysis, the business is left facing one more truth: You own it.
That search engine is yours and you need to stay on top of the technology that runs it, while making sure the system is performing at levels that will keep your customers happy.
It’s important, for instance, to consider all the little things. Klein talked about indexing the search database and the time it takes to carry that out, for example.
“This can also take hours. So, as a result, it has to be done nightly or during periods of low site traffic,” he said. And if indexing takes hours, he added, a company can find itself in a real jam if product reshuffling or price changing needs to be done during peak-site-traffic hours.
“You run into issues when that happens,” he said, “and that can lead to downtime or loss of revenue.”
And, of course, a home-grown site search engine is built on complicated, customized software. And complicated, customized software, Klein said, can be fickle.
“That comes with the risk that at some point you’re going to break it. So how long does it take you to roll those changes back, rebuild, recover? That’s all going to depend on your development processes and how robust your search-engine-tuning approach is.”
Place on top of all that the need to monitor performance in terms of the accuracy of results, the time it takes to deliver those results and the reliability of the system that delivers them.
Problems inevitably crop up, Klein said. And when they do, it’s up to the people who built the system in the first place to fix them. Again, that’s not impossible. Some businesses have assembled teams of engineers and developers to handle the performance and maintenance problems that invariably pop up.
We just thought we’d point out that it’s something to think about.
Mike Cassidy is BloomReach’s storyteller. Contact him at [email protected]; follow him on Twitter at @mikecassidy.
Staples Case Study