jump to navigation

Personalization in BI: Selective Dissemination and Targeted Retrieval of Important Information January 30, 2007

Posted by Cyril Brookes in General, Issues in building BI reporting systems, Stafford Beer, Tacit (soft) information for BI.
add a comment

Personalization in BI grows in significance with the near universal recognition that passive reporting, designed for the masses by supposed experts, is limited in its utility. Action oriented reporting is preferable; it always has been. However, many business analysts do not recognize that selective dissemination of information, aka personalization, is a pre-requisite if reporting is to stimulate action. Only specific people can or need to take action, and common sense tells us that they must be targeted.

See, for example, the article by Neil Raden.

Two sorts of personalization apply in a BI context:

         Push and Pull

Each can be applied to external or internal recipients.

The focus of this blog is on selectively pulling information, predominately by internal people. This is the principal aim of action oriented BI, directing valuable information (and only valuable information) to executives and professionals that assess a situation and/or take action as a result. This is not to say that other aspects of BI, such as keeping people informed as to status of the business, should be ignored. But these objectives are far less vital than supporting executvie actions.

I leave discussion on the much more fraught selective information push situation to others. Determining what information will be of interest to a customer or supplier and pushing this category of stuff to them can be a valuable marketing tool, or (more likely IMHO) a PR disaster. We always hear of Amazon.com and its success with cognate book promotions, but books are easily categorized in a universally accepted manner; most other items are not so easily classified , and the implications for inappropriate information push can be dysfunctional.

Any discussion on the effectiveness of BI for improving the quality of executive decisions (and what other purpose might it have?) must have regard to the actual decision making process. The theory of this process is well established, notably by Herbert Simon. Many researchers have also considered the relationship between this process and the information required for its operation. In this context I particularly value the work of Stafford Beer and Henry Minzberg.

Information that enables effective decision making belongs to one of two categories, and both are essential if decisions are to be optimal:

It helps the executive find problems and opportunities – situations that need a response.

Stafford Beer calls this Attenuation information, and I have discussed this in detail earlier

It helps the executive solve problems he/she has found (or been told about)

Beer calls this Amplification information, also discussed in this blog earlier

But I diverge.

Returning to the personalization theme; selective dissemination is vital in the problem finding context.

Obvious candidates include:

  • Alerting the executive to important exceptions, out-of-specification performance, unusual situations, adverse forecasts of key indicators, and unacceptable (or advantageous) trends.

  • Equally, if not more, critical is soft information (opinions, comments, assessments, etc. that portend problems, or throw doubt on the accuracy of factual information.

Targeted information retrieval is also vital to support problem solving.

  • The solution process that needs supporting includes diagnosis of the severity of the problem (what will happen if nothing is done), identifying possible alternatives and assessing their implications.
  • During a decision making process executives must be able to retrieve important, valuable, information as distinct from the routine stuff. This applies to both factual and soft (tacit) information. In this context, the latter includes ideas about problem implications, suggestions for potential solution alternatives and recollections about what we did last time this happened.

The key word in both these situations is “importance”.

Alerting to, and targeted retrieval of, useful information implies that some assessment must be made of the significance of a data item, either using an automated rule system, or a personal assessment.

Truly this is the stuff of business intelligence. Without importance classification all information is equal, but obviously this is not reality.

Selective dissemination and targeted retrieval, the basis of all personalization, depend therefore on the BI context being able to distinguish information importance as well as its subject, topic, or data class.

Importance, in turn, depends on two characteristics: urgency and value to the business.

I have experimented over 20 years with different retrieval/alerting procedures for corporate BI systems, using both automated and human importance assessment. I’ll detail this experience in the next post.

Advertisements

Drilldown specification in BI– it’s harder than the demo makes out August 27, 2006

Posted by Cyril Brookes in BI Requirements Definition, General, Issues in building BI reporting systems, Stafford Beer.
3 comments

We’ve all sat through those glitzy demonstrations of canned detailed reporting – solving hypothetical problems that may never exist in practice, or if they have existed, will never be repeated again in the same form. 

Drilldown is, in my opinion, one of the hardest aspects to specify and implement effectively in a BI reporting system.  Unfortunately, it is also probably the most hyped “silver-bullet” keyword of BI software marketing speak. 

You cannot build a detailed report for a problem situation that doesn’t yet exist.  Even if the type of problem, say a debtor’s default, is predictable, the specifics will always vary.  You can only create an environment to make such reporting easier. 

Further the nature of that reporting environment is often complex, involving hypothetical database specification and modelling capability – as I will discuss in a later post. 

Drilldown is, of course, a modern descriptor for part of Stafford Beer’s Amplification concept. 

In my posts of July 7 and July 28, I introduced and discussed Stafford’s distinction between Attenuation and Amplification reporting in the corporate BI context. In summary:  Attenuation reporting is pre-specified and pre-formatted reporting of information that empowers executives:

  • To be aware of, and able to assess the implications of, the current state of the business, and
  • To be alerted to actual or potential unusual or unacceptable situations. 

Or more colloquially:  Where are we?  What is good and bad about where we are?  What is unusual or forecast that I need to know about? 

Therefore, Attenuation style reporting makes executives comfortable they know what is happening in their part of the business, and that they are alerted to any problems that can be discovered from the data available.  Amplification reporting is more difficult to pin down, since it is only required when a problem, or apparent problem, has been identified. 

If Attenuation is about finding problems, Amplification is about solving them.  

By this definition, then, Amplification reporting – or Drilldown if you prefer the current term – is difficult to pre-specify or pre-format because we cannot be sure of the exact nature and context of the problem.  

In the good old/bad old days of this technology, we used the terms Executive Information Systems for the first type of reporting, and Decision Support Systems for the second.  Now it’s all encapsulated in the BI terminology.   The names change, but the issues remain the same. 

But I diverge. 

Drilldown has three basic objectives, as I see it, drawing from classic decision theory – as per, say, Herbert Simon.  It is to empower our client executives to: Diagnose the problem/opportunity we have helped them find (using Attenuation reporting).  How bad is it?  What will happen if we do nothing?

For Diagnosis support, the minimum requirement is more detailed data.  Exactly how more detailed, and over what time periods, can only be determined through appropriate research and interviews with client executives.  However, models that can answer “what-if” queries, and statistical analysis tools may also be valuable.

Determine the available options for solving the problem – or capitalizing on the opportunity

Options may be selected from past experience, or suggestions from experts.  Support for this aspect of Drilldown is often based on knowledge sharing and tacit information management tools – see my post of July 27. 

Assessing the implications of each of the apparently viable alternatives

Feasibility validation and outcome assessments of options may require pre-specified models, or at least partly constructed versions that can be adapted to the problem situation when it is known.  Assessments of the anticipated outcomes of the viable alternatives are often the key information elements required to support decision making and the BI system should include these if at all practicable. 

Many people think that Drilldown is only about getting more detailed data, but clearly the client executive is likely to want more – just like Oliver Twist.  A quality BI environment design will go much further as discussed above, but at least will extend to a full review of what that “more detail” should be, and how it is to be collected for inquiry and ad-hoc reporting. 

 After all that, it’s up to the executives to judge which alternative is best, i.e. make a decision.   

I will cover these Drilldown component steps in more detail in a later post.  If you want more detail immediately, you can see how these concepts are implemented in the BI requirements methodology at www.bipathfinder.com and how the metadata that supports the method is documented at www.bidocumenter.com 

Your comments on the accuracy and utility of this material are welcome.

Stafford Beer’s “Attenuation” concept – how to implement it in BI reporting July 28, 2006

Posted by Cyril Brookes in BI Requirements Definition, General, Stafford Beer.
2 comments

My July 7 post on Stafford Beer’s legacy for BI professionals gave an overview as to how I implement his concepts in my BI designs. Here is some more detail on one of these principles – Attenuation.

Attenuation in BI is the art and science of presenting information to executives in such a way that they are:

Aware of the current status of the business

Satisfied that situations requiring their attention are highlighted, and

Alerted to unusual situations that may presage problems or opportunities

This obviously requires specification of pre-formatted, on-demand or ad-hoc reporting. Executives will know, or can be helped determine, what they need to know to meet these three objectives.

Theory indicates that there are 3 possible ways to implement Attenuation in BI reporting for the above objectives:

Where Are We? – Summarize the situation, tell the executive the status

What’s Good and Bad about where we are? – Compare, compute exceptions, find the executive’s problems

What’s unusual, and what’s forecast? – Find out what is obscured by noise in the data and alert the executive

A creative BI design will present information using these three approaches.

Attenuated information, derived from a mass of transactions data, and augmented by tacit (soft) data where appropriate, can give huge benefits to the executive – it just needs some mental effort applied in a structured manner.

Of course, we also need to ensure balance of reporting scope, in the Balanced Scorecard manner. I believe this is a different concept, and is best handled via the template of potential KPIs that forms part of a comprensive methodology.

If you’re interested, check my technique for doing this at www.bipathfinder.com where I explain this in more detail, including the template design aspect.

I’ll cover Stafford’s Amplification next time round.

Knowledge poverty in information abundance – Stafford Beer’s insight July 7, 2006

Posted by Cyril Brookes in BI Requirements Definition, General, Stafford Beer.
4 comments

In this age of BI excess of development capability and resource overload the lessons taught by our not-to-be-forgotten pioneers need to be reinforced. Stafford Beer is one of these leaders. There are many web sites with details of his work, e.g. www.staffordbeer.com.

For the BI professional, indeed all business analysts, I believe his most valuable insight is the distinction he labelled, perhaps obscurely, Attenuation versus Amplification.

This distinction ought to be part of the day-to-day mindset of every business analyst. It underlies the difference between the design of:

  • Pre-formatted, routine or on-demand, reporting for executives, and

  • Drilldown, detailed reporting for executives and their assistants

Attenuation, as envisaged by Beer, involves the presentation of summarized, assessed, forecast and otherwise pre-digested information that has the following benefits for executives:

  • Comfort that they know the status of the business, or their part of it

  • Assurance that any problem situations that are detectable, are detected

  • Confidence that they are alerted to unusual situations that may indicate problems

Amplification is a different class of reporting. It is only relevant when a problem exists, and involves presentation of detailed and assessment support that facilitates:

  • Diagnosis of problem severity, implications and sources, and

  • Modelling of the problem context to establish viable alternative solutions and their impact

I believe that many business analysts and BI designers today are unaware of this fundamental distinction. There are TWO types of information reporting – one that can be pre-formatted and the second that can only be facilitated. Design of BI reporting capability and content will be different for each type. If the designer does not make this distinction the result is that the report specifications do not match the problem finding and decision making mental processes of the target executives.

I will cover these concepts in more detail in a later post, but you may care to see how I’ve addressed them in my work on the BI Pathfinder methodology, see www.bipathfinder.com. In it I’ve attempted to implement Stafford Beer’s concepts, with particular focus on the specification of data definitions and dimensions.

If we recognize and implement the attenuation/amplification distinction more often in our report designs the knowledge awareness metric should increase dramatically.