Isn’t cognitive search just a fancy-shmancy way to say, “enterprise search”? Kinda.

Enterprise search software – applications that help employees find the data and documents to do their jobs – is pretty long in the tooth. But maturing AI and machine learning techniques are encouraging many organizations to take a fresh look at their current enterprise search solution to see if a cognitive search platform might be the right approach at the right time.

What is a Cognitive Search Platform?

A cognitive search platform uses AI-powered technology to better understand the intent of a user and connect them to the data and documents they are looking for. The user in this case could be a shopper browsing an online product catalog, a customer looking for help in a support portal – or even an employee trying to find answers for their own internal projects.

Cognitive search is an evolution in how we think about search and big data in business and in the workplace. Think of it like Siri, Alexa, or Google Home, but for your business. With advances in natural language processing, artificial intelligence, and AI, truly intelligent search and discovery applications are here and ready for deployment

Key Features of a Cognitive Search Platform

As you’re looking for the best cognitive search platform for your organization – or to replace an existing one – here’s a rundown of the features you want to be sure it includes:

At the very basic level there’s your data source connectors that import and index your data and documents. The most important question isn’t whether the solution supports the most connectors, but whether the solution supports your unique mix of data sources. Parsers work in conjunction with data source connectors to process and index your data.

A lot of the AI and machine learning capabilities of a cognitive search solution are made possible through signal capture. Signal capture is the recording of events like user queries, clicks, adds, purchases, and other similar clickstream data. In more advanced applications, user profile information can be integrated like job role, office location, vector, altitude, or any type of event data.

These signals are a big part of how a cognitive search platform provides recommendations. User signal data is aggregated to assemble a personalized experience for every user, for every visit, and every query. Ideally, the platform can also import and integrate with any existing data and ML models created by your data science team.

Another next-gen feature is zero results management so that a user never sees the dreaded dead-end “no results found” page. With semantic vector search and related techniques, a cognitive search solution can understand intent and always suggest items related to the original search request. This increases conversions, average order value, and other key metrics.

Every user expects that every time they type in a search box the system will try and suggest or autocomplete (also known as typeahead). A sophisticated autocomplete capability will instantly suggest queries and related queries as well as categories and products that include attributes such as links, prices, and thumbnail images.

And even with all these new technologies and automated relevancy and user experience improvements, there still might be a need for manual adjustments with business rules. Business rules management remains important for manually adding, removing, and editing business rules including boost, block, bury, and pin.

Security is also critical for a cognitive search platform, making sure it can integrate with your existing security infrastructure. Many cognitive search platforms integrate out of the box with major security models like Active Directory, LDAP, Kerberos and SAML or other single sign-on systems. This includes role-based security authorization to determine which users are allowed to delete, read, modify, or create documents as well as enact system changes. These fine-grained controls ensure that users don’t see documents they don’t have access to in a set of search results, recommendations, or even with the autocomplete.

Also key to relevancy is named entity recognition. NER uses natural language processing techniques to recognize all the “things” in your data – brands, locations, industry terms, prices, dates, employee names – all the proper nouns. This turbo-charges apps to be more adept at finding what users are looking for faster.

Any standard search application should also include synonym detection, making suggestions as users type – even if they misspell something. This capability should go past simple word replacement and use machine learning and language analysis in order find good pairings between two words based on how often each appears together or at different locations within sentences.

Faceting is a critical capability allowing users to use checkboxes or other UI elements to filter results based on specific fields like color, size, file type, price, distance, and other parameters. Faceted navigation and range filters make it easy for auser to keep their search confined by specific criteria.

A/B testing and multivariate testing are essential for the administrator of a cognitive search platform to determine whether changes in search result rankings improve clickthrough rates, purchases, or any other measures of success. It’s great to have personalization and recommendations and all the other advancements but just as important that these adjustments actually do what they’re intended too.

And of course none of this is possible without scaling. Customer-facing and internal cognitive search platforms have to weather the always fickle currents of shopping trends or internal projects and teams. Brands and retailers have Black Friday haunting them every year, and HR teams have open enrollment to deal with. Every search infrastructure has to be able to scale up and down and horizontally to handle the various seasonal (and sudden) spikes in demand and usage. If you’re using a third-party vendor for hosting you’ll need to make sure that your vendor has a proven track record of experience helping organizations like yours get the most out of their search.

Measuring the Success of a Cognitive Search Platform

In addition to features, another critical part of choosing a cognitive search platform is measuring its effectiveness. With ecommerce search applications it’s much more straightforward, shoppers either buy or they don’t. But with an employee-facing search, there’s a lot more nuance when you are demonstrating the value of the investment you’ve made into a new cognitive search platform.

  • Utilization is important and query volume measures how many queries per second are coming into the system and will depend on the size and popularity of your user base, as well how many other people are using it at once.
  • Responsiveness is also important and query response time looks at how long it takes the application to receive a search query, understand it and serve the results back to the user.
  • Number of concurrent users can show you how well the system performs with various volumes of users using it as well as if the system can scale to handle the stress of all those users. Smaller organizations with fewer overall users will find this less important.
  • The size of the search index is also important to always track the sheer amount of data and documents being indexed, searched, and accessed.
  • Dead end queries that produce no results are an indicator of poor relevancy or the application not properly understanding user intent. Number of zero results queries is a report showing you which queries are returning zero results, stopping the user dead in their tracks.
  • Getting direct user feedback on how they feel about the search app can be a softer measure but user satisfaction scores are always key.

Don’t forget to baseline!

Before you move to a new cognitive search platform from your crusty old one, be sure you capture enough data so you can present a good before and after picture to show the success of your initiative and ROI.

Making the Final Decision

As you shop for cognitive search and knowledge discovery solutions, there’s a lot to keep in mind. When all is said and done, it’ll be up to your users to truly show the success of your cognitive search deployment. With the right capabilities and the right metrics you’ll have a cognitive search solution that increases productivity and findability for your customers and users.

Is your brand ready for a powerful cognitive search platform solution? Drop us a line. 

About Andy Wibbels

Read more from this author


Contact us today to learn how Lucidworks can help your team create powerful search and discovery applications for your customers and employees.