Challenge

The management recruiting industry is being reshaped by Web technology. As web-based resume databases have facilitated candidate discovery, employers expect job searches to be completed more rapidly and less expensively. With more companies taking the 'do it yourself' route by posting open jobs on careers sites, recruiters are challenged to demonstrate the value of their services by locating skilled and gainfully employed candidates who are not actively trying to change jobs. In recruiting lingo, these individuals are termed 'under the radar'.

This summer, WebFarming.com was retained by a Colorado management recruiting firm to evaluate how to systematize the detective work of recruiting. While the Web is also being leveraged within the recruiting industry to advertise positions, accept employment applications, and to market recruiting services, the primary focus of our effort was to improve the online discovery process. The objectives were to identify new sources of qualified candidates and to integrate information about candidates in the firm's contact management system.

Approach

Because the ultimate goal was to match possible candidates with open jobs, we began by analyzing positions to be filled. A key question to be answered was, "How variable are the search assignments?" If there was a high degree of similarity between jobs, it would be considerably easier to automate aspects of the search process than if each job were unique. Since our client specialized in filling a specific type of high-tech sales position, there was substantial commonality among the assignments. For each position, we identified target companies, skills key words, and job location.

A central element of our strategy was to probe target companies for possible candidates. Using different search engines, we identified Internet domains associated with target companies and attempted to access pages with information about employees and user group members. We also identified sites linked to target company sites, implemented 'power searches' using Boolean logic to scan for candidates, and searched electronic communities like GeoCities for candidate homepages. Consistent with our goal of locating passive candidates (i.e., not actively looking for a new job), we did not scan resume databases like Monster.com. Operating on the assumption that different search engines produce different results, each step of the research process was implemented with more than one search engine.

Experience

The initial research was extremely time consuming. On average, each search required about six hours research time. We were sure that the process could be faster. A key factor affecting research time required was the consistency of information checked. While all the documents in the contact management system and resume databases were resumes, a search can match product announcements and company job postings, as well as candidate profiles. While we tried to structure our search terms to exclude irrelevant data by using 'AND NOT', we continued to retrieve irrelevant items.

Variable quality of information was also problematic. Duplicate listings were a persistent issue. It was considerably easier to locate some types of candidates than others. For example, information about the employees of large companies was more widely available than information about the employees of small companies. And, technical candidates were more prevalent than sales candidates.

On a positive note, there was very little overlap between the candidates identified through the new Internet research process and the candidates sourced via resume databases. Our methods were producing new blood for the database!

Hit Rate

After performing our pilot searches, we evaluated our 'hit rate' or percentage of relevant to total matches. Analysis was limited to searches that worked effectively and might be considered for repetition. As a result of omitting non-productive searches, we calculated the best case hit rate. The hit rate for the best 29 searches was 3%. From a systematization perspective, the low hit rate meant that considerable manual intervention was required to sort out the high quality matches. Improving the hit rate was the next big challenge!

Search Tools

To determine if it was possible to improve search precision, we tested a commercial tool, BullsEye from IntelliSeek. BullsEye is a meta-search engine that is designed to access and combine results from many different search engines. Using BullsEye, it was possible to achieve hit rates ranging from 15% to 41%, a substantial improvement over the hit rates achieved with free search engines. Several features contributed to the improved accuracy.

BullsEye does not rely on different search engines that operate differently to execute Boolean logic correctly and consistently. Instead, it performs searches in two successive steps. In the first step, BullsEye downloads all matches to a more inclusive query from web sites to local storage. In the second step, BullsEye allows the user to make the match criteria more specific by adding terms to exclude or synonyms to match. In addition to checking that matched sites are active, BullsEye verifies that the text of matched documents actually meets the more restrictive search criteria.

An important feature of BullsEye is the ability to schedule effective searches to rerun automatically. BullsEye monitors matches already reported and presents new matches only. Taking advantage of this capability, we set up regular queries for each of the target companies.

Next Steps

We are continuing to focus on how we can continue to improve productivity of our searches. Because manual review of matches will almost always be required to verify resume quality, we do not think it will be practical to completely automate the process. At the same time, we believe that we can make the discovery and acquisition of candidates even more efficient. We believe that streamlining how data is transferred from the Web to internal systems represents a big opportunity. In a time consuming procedure, each document is now copied, pasted, and indexed individually. We are investigating different tools to identify specific data elements within differently formatted documents and methods for importing the documents as a group.

We are also evaluating how to apply more indirect Web research methods. Other sources of names, like searches of discussion groups, alumnae lists and ISP subscriber homepages, tend to produce more variable lists. We are currently testing use of lists of this type in email campaigns to brand and market our recruiting firm's services.

Of course, the true gauge of the project's effectiveness will ultimately be its return on investment. With a new employee placement generating $15,000 revenue on average, only one incremental employee placement per month will be required to justify the new research investment. We are monitoring number of resumes generated, number of interviews scheduled, and number of placements closed as an outcome of the new process. While no placements have occurred during the first eight weeks, our Internet candidates are being actively presented to employers. And, the fact that our client's database has expanded with many more qualified candidates has improved his ability to sell the value of his service!