How Organizations Learn
An employee quits or retires, and as he or she walks out the door, so too does the knowledge that they learned about the projects they’ve done, the clients they’ve worked with. A particular project goes badly, and all everyone wants to do is forget about it, yet even as that happens the seeds for that same failure are cast into the soil of future projects. Demos are produced for a client, but the servers get repurposed, the code gets out of date, and when a sales team wants to showcase what the company has done, they end up having to rebuild demonstrations and presentations from scratch.
Organizations have notoriously bad memories.
The irony here is that most companies more than a decade old have reams and reams (and reams) of corporate case studies, Six sigma manuals in double wide binders, ISO 9000 business specifications and so forth, sitting on shelves and gathering dusk. Even in organizations that are diligent about retaining information such as this, that knowledge (and where it does and does not apply in the company’s particular case) is usually poorly utilized.
In the case of companies that have done business in the last twenty years, this same information is kept in Sharepoint repositories filled with spreadsheets, word documents, Powerpoint presentations, PDFs — or similar systems. While this has helped somewhat to reduce the clutter in the office, and has made it possible for people who have done these things to better find them, the reality is that once the author of a document creates it and uses it for a meeting or a presentation, it very seldom gets repurposed, or even reaccessed.
This ultimately suggests that being able to save resources, while a necessary part of reusing information, is seldom a blanket solution for the problem of retaining and reusing institutional knowledge.
Rethinking Knowledge Management
The discipline of knowledge management is a comparatively recent one, and is often confused with its sibling, content management. Most content management systems are ultimately adaptations of publishing workflow systems, and are intended primarily for the production of consumer facing international property. Knowledge management systems, on the other hand are more “meta”, to use contemporary vernacular — they are about capturing the information about producing that content.
Surprisingly, an online service such as Salesforce do a surprisingly good job of capturing not only the specific transactions involved but also of retaining metadata commentary — notes and thoughts about clients, successful and unsuccessful projects and similar information. This works largely because they recognized several fundamental problems in knowledge management:
- Information must be captured when the event that precipitated it occurs. This information reflects knowledge as it is known at the time, which even if incorrect or incomplete is usually very insightful about process. It also cuts down on the tendency of people to edit knowledge over time to better reflect upon their own actions.
- Most people do not have time to do their work and to report on it at the same time. Because of that, it is usually better to capture a sentence or two that captures the essence of the situation than it is to work with an extensive multipage report.
- Entering information can only be effective if it fits well within the workflow of the person making that report. Version control systems like SVN or GIT usually force a person when checking in code to write a short summary of what the code does or is. It’s a part of the workflow. If that same user has to stop and update another website, then that information will simply not get captured, because after a couple of days of that, they won’t.
One of the key points from the above is that knowledge management should not be a separate tool, but rather should be integrated as much as possible with the process that people use to do their work in the first place. Unfortunately, this does not happen very often, in part because there are relatively few standards for capturing knowledge metadata, in part because most CMS systems (which often do double duty as KMS systems) currently do not provide a means to capture it.
Knowledge, Classification and Semantics
However, there’s another problem that is more subtle. Most people are simply not very good at classifying, or in abstracting information down to a good summary. There’s perhaps one person in forty that really has the mindset to do something like this well, and that person is likely already employed as a librarian, lawyer, curator, analyst or information architect, or is trained to do so in the course of his or her other duties (police officers, insurance adjusters, medical workers). This is in fact one of the big bottlenecks in the growing field of metadata management.
Many knowledge workflow systems make an attempt to manage information by establishing categories that people can choose. At the crudest levels are long lists of categories that someone has to choose from, perhaps selecting multiple items. There are two variations, the first a form of type-ahead that will show you possible category keyword matches in a type-ahead, the second a treeview navigator of some sort that lets you drill down to find the category that best matches the expression.
The latter form is a very basic example of a simple semantic system — you are using the meaning of the categories (and their specificity) to better navigate the information space. For instance, if I look for a category like “animal” then I could choose this or I could select from (“shellfish”, “insect”, “fish”, “reptile”, “dinosaur”, “bird” or “mammal”), then drill down along a simplified Linnaeus fashion until I get to “cat” or even breed of cat (“Russian Blue”). Of course, in a more advanced system, I could even shortcut the process, type in the word “cat”, then select the relevant breed without the need to navigate the whole information space.
Utilizing classification systems like this is important, because it can go a long way towards making information searchable. Search is the key here — an organization can only learn if it has some way to find what it has already learnt. Normal keyword searches can tell you that a document (or a knowledge fragment describing that document or report) has a given keyword, but the more that you can provide classification context, the easier it is to identify not only a given document, but those that are related to it.
Additionally, each term that you add into the classification scheme provides a means by which you can better filter out irrelevant material. For instance, suppose that you wanted to find that presentation that you did for Megacorp, Inc. on the topic of selling your solar cell units to their division in Bangalore, India last year. If you can capture company, product line, location and time, then you’ve filtered out a huge amount of irrelevant information. You don’t need to remember which folder it’s in, and if someone else created materials supporting that, then you will have discovered new material that may be relevant.
This is another important aspect of knowledge management:
Knowledge does not exist independently — there is always a context in which information occurred, and by attaching that information to what is already known, your organization has learned something new.
This is the heart of semantics — it is a system by which you represent knowledge. When I talk about semantics here, I’m referring both to the process of classification (and the tools by which you can use that classification) as well as, more formally, the W3C Semantic Web standard, which is one knowledge system (there are others, but most agree on the fundamentals).
There are signs that formal semantics is making its way into digital asset management systems, metadata management and business intelligence tools. While it is only one part of artificial intelligence, semantics provides the tools to make reasoners possible — to look at information that has been collected by various means, and then make new conclusions — new facts — based upon this information.
One additional arena where semantics is becoming important is in the creation of abstracts. An abstract is a short summation of the important points of a given document. This is not for computer consumption but rather is intended for human readers. Such abstracts are created by computers that both perform enrichment of the content and that look at the structure to attempt to isolate the highlights of what was submitted. This is especially important for information that isn’t normally highly structured, such as email, commentary on blog posts, and similar streams.
The Importance of Post-Mortems
However, for all that there are tools on the horizon for better managing the classification aspects, there’s also a very big role still for human analysis, especially in the area of post-mortems. A post-mortem in the purest sense is a report upon the death of a person that describes the conditions that prevailed at the time of their death, as well as interpretations about why they died. While there may be a certain appropriateness of this metaphor in specific business situations (the client just fired you, for instance), in the business sense a post-mortem is a write-up of how people perceived that a project succeeded or failed, and the reasons why.
Again, I think that most organizations do post-mortems badly. First, the tendency with post-mortems is to put all the people involved in a project together into a room, then to hash out why a project failed (and is usually held only when such a project failed). This to me is a fundamental mistake, because it invariably becomes a search for blame, even if it in fact does not start out that way.
Instead, post-mortems should be conducted via writing, with the expectation that each participant will answer fully each of the questions asked. The fiction of anonymity can be maintained (though it is usually a thin pretext), but what is more important here is the statement that there will be no formal recriminations to anything answered by any participant (something that can and should be maintained by ensuring that such post-mortems also go up the chain to the CEO ultimately).
The post mortem is a time for reflection and analysis. What worked, what didn’t? What factors were outside of the control of the individuals answering, what weren’t? If a failure occurred within the organization, who failed and how did they do so? If a success occurred, who were most responsible for that success? Why do you thing they failed or were successful? If you could do certain things over again, what would you change? Why?
The post-mortem process should be done up the chain — once a manager has received post-mortems from her team, she should then create a post-mortem report of her own, which goes to her manager who repeats the process. Post mortems are private documents — visible to those only in the immediate chain of management. Ultimately, the CEO will also write a post-mortem, synthesizing what he’s received from below, for company wide dissemination, removing names and potential identifying information but incorporating lessons learned and action items to improve based upon this information.
This does bring up another point of knowledge management.
Knowledge exists only in context, but context within organizations is usually political. This means that knowledge management is almost always political as well.
Actions have consequences. and information about actions cannot help but also carry the taint of those consequences. This means that managers in particular need to be mindful that information collected about those within the company exist at different levels of access, and that such levels of access are not necessarily related to one’s job title. It also means that the ability to maintain such level of access is critical to a knowledge system.
Dissemination, Curation & Social Media
I’ve seen a fair number of “knowledge management systems” over the years, some organized with social media features, others more along the lines of wikis. These can be utilized for the purpose of providing “chunked” content and in general seem to be fair (if not great) at storing frequently utilized information. Most would be more effective if they made better use of classification (freeform and otherwise) rather than trying to conform to a strict hierarchy of organization, but this day is coming.
To the extent possible, such systems should be curated — authorized people should work to provide topical context and categorization, to classify the relevance of information and to weed out what is increasingly irrelevant information. This becomes increasingly important as organizations become larger. A related task for such curators is creating FAQs — frequently asked question collections — on various topics. Such FAQs are editorial in nature, designed to consolidate that information that people need on a given topic into a centralized location — onboarding, HR, corporate policies, events, and so forth.
Additionally, a good knowledge system should be able to incorporate external works that people within the organization produce — blog posts, conference papers and presentations, mentions of individuals or the company in external media, and so forth. Given the increasingly central role that Twitter plays here, a good KMS learning system should be set up to identify external links from Twitter tied into people, products, companies or initiatives, present them for consideration, then turning these into “newsletters” that can then be persisted into the knowledge system. This brings up another key point for knowledge management systems:
Knowledge does not stop at the boundaries of an organization. It’s not only about how you view the world, but also about how the world views you.
The upshot of this is that knowledge management should take into account (and integrate with, as much as possible) the social media that is increasingly serving as the marketing presence for the organization itself.
The ability to learn is critical to organizations in today’s world. Competition has gone from the company down the street to the company located halfway around the world, and the inability to learn can spell the difference between a company succeeding or going bankrupt. This is true at all levels — while the CEO needs to have a clear understanding of the changing state of a business, so too do sales people, programmers, marketers and everyone else within the organization. You learn, or you fail.
The theme of this particular post was suggested by Flavio Tosi, who asked me what my thoughts were with regard to “Learning Companies”. If you have any thoughts about what factors contribute to helping organizations learn (or if you have suggestions for other topics) please feel free to comment.
Kurt Cagle is the founder and ceo of Semantical, LLC, a consulting company specializing in data science, NoSQL and information architecture.