Semantic technology: online business intelligence

What are your customers looking for?

Any marketing and sales director from any company would like to know the answer to this question. Here is the key to all sales, and knowing the answer makes all the difference between business success and failure.

Multinational corporations spend a lot of time and money trying to discover the answer to this question by means of surveys, focus groups and all sorts of market investigation tools.

However, many companies still seem to ignore the most obvious thing: users search for what they want on the internet. And they do so in a clear and simple manner. Usually, these search queries are made on the company’s website and, therefore, on the corporate search engine (if there is one).

online customer care

All large enterprises (and almost all SMEs) already have a corporate website. Some enterprises show their customer care telephone numbers on their websites, often on the main pages or on “Contact” sections. Others hide the numbers carefully so that finding them becomes a difficult (or impossible) task. The logic behind this fact is evident: offering telephone customer care is expensive. If customers cannot find the number, fewer calls will be received and thus the cost of them will be lower.

FAQ (frequently asked questions) sections can also be found on a corporate website. Others have Web Maps, others offer search engines (based on Google Search Appliance, Google Site Search or other keyword-based search engines). Many websites have all three items as tools for users (and future clients) to find what they are looking for. All this is particularly necessary if phoning the call centre ends up being difficult or impossible.

According to a study developed by Yahoo Research (2007) 65% of internet users search what they need on a website without even reading its home page. And in case they do not find what they are looking for, they immediately go back to their favorite internet search engine and they keep searching there (so the owner of the site loses a clear selling chance).

Corporate search engines

Many companies include corporate search engines on their websites to allow users to search what they want, using one or several words in a text box.

Our online investigations with large enterprises reveal worrying results: 60% of such search queries have no results at all. The rest of them often obtain poor results, and it is very easy to find funny examples.

A user once visited the website of and important Spanish bank group. The user searched for “offices in Sabadell”. He or she obviously wanted to know whether the bank had an office in this town near Barcelona and, in case there was one, which was its address. However, the corporate search engine only provided information about the extension of the offices a competitor bank had recently carried out. What was the result? The potential client was led to the competition straightaway.

Another user recently searched “avoid porn to my son’s user” in a large telecommunications group’s website. The question was self-evident: this father wanted to activate the parental control on his computer to prevent his son from accessing adult contents. The answer of the search engine arrived in 0.01 seconds (according to the search engine) and showed the porn films offered by their pay-per-view service to the surprise, I guess, of the user.

Do search engines not work?

The answer to this question is easy: if “working” means finding web pages containing the words we typed, then, yes, they work perfectly well.

On the contrary, if we believe a search engine should understand what we are looking for and provide only truly relevant information, then traditional search engines are far from offering satisfactory results.

Demantic search engine

At Inbenta we have been helping our customers implement their best search engines for years. By using natural language processing techniques and semantic analysis, we have developed applications with a 90% percentage of correct and relevant answers. And the effects do not take long to come: one of our customers registered a 40% reduction in incoming calls in their call center (along with a subsequent costs saving).

With another customer we measured a commercial conversion ratio of a 4%; that is, 4 out of every 100 search queries on the corporate search engine end up purchasing something on their online shop. Up to the moment, any other of their online resources has been able to provide such a high conversion ratio (including banners, impacts, click buying, adwords and other tools).

Dearching within the search engine

The accumulation of all search queries made by online users provides crucial information to understand their needs and to increase online sales. However, analyzing the record of users’ search queries has an important difficulty: its dispersion.

According to our experience working with important companies online, we have observed common features when analyzing user search queries logs:

  • Most popular search queries (referred to as the “top 100” by many search engines) hardly represent 20% of the total search queries made by users.
  • The average of words in users’ search queries ranges between 2.5 and 3. Therefore, dispersion increases.
  • Users who basically search the same use different words, obtaining very different results on traditional search engines.

Semantic intelligence

The key to success is semantics. In Inbenta we use the same semantic technology that guarantees optimal search results to analyze user search queries logs.

Our innovative semantic clustering technology compares all user search queries with themselves, detecting those search queries with similar contents. Those which are semantically similar are grouped (or clustered).

For instance, user questions such as “how much does it cost to insure my doberman” and “cost insurances for rottweiler”, even if they do not have a single word in common, belong to the same semantic group, which we could call “Insurance prices for dangerous dogs”.

Our semantic analyzer and a specialized dictionary are able to analyze thousands of search queries following these semantic patterns with surprising results:

  • The most popular semantic groups (“top 100” semantic clusters) cover more than 80% of user search queries. This means that search queries dispersion is lower than it seems when only keywords are analyzed.
  • With only a few contents semantically indexed the great majority of user questions can be correctly answered.
  • These semantic popular groups bring highly valuable business information.

With this analysis, Inbenta has already helped many companies to focus their web contents on what really matters to their users and potential customers, achieving important improvements in online sales and in customer care costs reduction.

Inbenta Team
by Inbenta Team