Tuesday 19 February 2013

20,000 social network analysis - trick or treat?

A colleague wanted to pick my brain about how to run a social network survey with a company of 20,000 employees. He was worried that it may be more trouble than it's worth. This is a great challenge - the scale necessitates a well thought-through plan to lead to flawless execution. Little mistakes are magnified when you multiply by thousands.

First, I offer a set of questions to get a feel for what's feasible, the lay of the land.
  • Do you have survey software capable of dealing with 20,000 respondents? Is it home-grown and will it crash when too many people log on at the same time?
  • Can your software comfortably load and display a name list containing 20,000 employees?
  • Will the name list be easy for employees to search, filter, and discover their co-workers?
  • Can your survey software export the data in a way that minimizes the amount of manipulation required to put it in a format readable by the visualization software?
  • Can your visualization software deal with 20,000 nodes and potentially 1 million connections without errors or crashes?
If you answer "No" to any of the above questions, here are two more:
  • Do you have a budget to build, buy, or outsource the capability?
  • How much time do you have to build, buy, or outsource?
  • Do you have the expertise to build, buy, or outsource successfully?
Other points to consider:
  • If you are not fully committed to SNA as a standard organizational development technique, you may not wish to invest too much in building capabilities until you understand the impact it can yield. In that case, outsourcing to a consulting firm that can handle this size of network would be the best option. At McKinsey, I recently worked on project for a financial company; the survey displayed a name list of 60,000 employees and the network contained 20,000 people with half a million connections. The data files were 450 meg! 
  • Are you in a rush to see results? Is there a looming deadline for which SNA forms a critical diagnostic component? Again, you may wish to outsource to a firm that does this all the time.
  • Do you have the IT capabilities to develop the technology in-house? If not, then a good survey vendor could be a partner for developing a module to add on to their existing platform. This process takes time and requires a good understanding of what is needed to gather the data from survey participants and the format for downloading the raw output. 
  • Although there are many survey vendors out there, very few are familiar with SNA. It may take time to cultivate and negotiate a relationship with a survey vendor to allow you to build the capability.
Once you have solved for the issues above, there are some tactical items next on the list.
  • With a name list of 20,000, budget on a solid day or two of cleaning up the list to remove or clarify duplicates and similarities. For example, although Rob Smith and Robert Smithe are not identical names, their co-workers may not see the difference clearly when presented in the selection list. These types of near-duplicates need to be augmented to something like: Rob Smith (Accounting), Robert Smithe (Germany). This is an important point related to integrity of data. 
  • Can you generate a comprehensive list of participants from your people systems? You need email addresses, first/last name, department, location, function, and other demographic information so that participants can filter to just their location. While it may seem ridiculously easy to some companies, this task can be time consuming for other companies. 
  • How should names be displayed - first name, last name or formal vs nickname? In some cultures, first name is always first. In one situation, the name list contained formal names of Chinese employees, but these employees had familiar names that were used for everyday interactions.
  • Is everyone able to access the internet via a high-speed connection? Typically, these surveys are connection heavy and are painful or impossible to complete over low speed or dialup. Do you have staff in the field, remote African mines, or no internet access at all? You will need to accommodate all of their needs. Many offices in India restrict access to the Internet completely - I had to speculate why.
  • It is important to test the rendering of the survey on the various internet appliances that an employee may use - Ipad, Iphone, BB, laptop, various browsers - and then let them know if any technology is not suitable.
  • The raw data file should be formatted to upload directly to your visualization software - without manipulation. If you have to make changes to anything over 1 million lines in an Excel file, you're out of luck. If you have helpful DBAs, then you could turn over transformation to them. Be sure to stand over their shoulder; you cannot afford to have any connections mixed up. This data is all about the individuals.
  • With an audience of this size, there will be questions while the survey is in the field. This is not a familiar type of survey, so many people will want to understand more about it. Be prepared with a dedicated person, mailbox, and help-line to answer these questions.
  • The response rate needed for a robust social network analysis is 80%. I have seen companies where employees dutifully respond promptly to every survey request. Congratulations if this sounds like your company - you are in the minority. More likely you will need to create a compelling communications campaign and enlist the assistance of group managers to round up stragglers. 
Next comes making sense of the data. This is the fun part. Mining a large network for insights is a substantial undertaking, here are some things to think about.
  • Your project should have a set of in-going hypotheses to test or investigate - myths to prove or disprove. You do need a person who can ask great questions to get the most from it.
  • Again, how much time do you have to dig into the data?  A network this size can take months to properly analyze. Don't forget that you have invested a ton of staff time into completing the survey (20,000 x 15 min = 5,000 hours), it would be a pity to rush the analysis and miss important insights.
  • What is your plan of attack? Prioritize all the analyses you wish to run with the data, keep a list updated and circulated with the team so that you keep your team on track and don't dive down time-consuming rabbit holes. When new ideas crop up, park them and reprioritize as needed.
  • What kind of non-survey (qualitative) data can you match up to the people or units in the survey? I've had some tremendous revelations when performance data was matched with network data - e.g., high performers have more diverse networks, high performing units have an egalitarian management style, positive attitudes yield better networks.
  • What are you trying to achieve with the analysis? SNA data can be used for so many different purposes - do you have a good sense of what you are trying to fix? 
  • SNA data is a snapshot in time. It cannot be used indefinitely. People come and go and networks change as circumstances influence them. You certainly can do a pre/post analysis to see what's different after you have made targeted changes.
  • You can use the data for what-if modelling. What if a senior person leaves, who else might follow? If we remove this person from the network, how many people become disconnected?
A large SNA project is truly coveted by practitioners. The research potential in a big data network is unmatched. I hope to find a company who would like to embed this capability into its human capital strategy. Wouldn't that be so cool.

Off I go now to form some more creature connections.

No comments:

Post a Comment