Skip to Content

And “Intelligence” said to BI – I want a divorce !

<p>If some one asked you to explain the reporting/analytics/scorecarding solutions you have developed so far for your organization, how would you respond? Here are some of the common answers</p><p> </p><p>1. This gives me the list of all <sales orders/POs/Invoices> for a set of selection parameters –  and we can see our biggest and smallest values, we can see a nice graph of the trends over time, and it will high light all exceptions outside the thresholds we have defined</p><p>2. This compares how we performed against <plan/forecast/budget> over a period of time</p><p>3. This shows us how we are doing on our SLAs – like how long each step of a process took (took 5 days for a Purchase request to become a Purchase order). </p><p> </p><p>Next question. Why do we need these solutions ? Each question has specific answers, but the common theme is – We want to know this so that we can do “course corrections” and try to do better in future. </p><p> </p><p>And now a short simple question to round up the discussion ” oh yeah? REALLY? and how exactly would you do that?” </p><p> </p><p>How much can we rely on past performance to predict future in business?  By no means am I the first to ask this question. And there are a lot of good answers on why/how historical information gives clues on future performance. </p><p> </p><p>Can you compare data across time horizon in “apples to apples” fashion ? sure – as long as we have common characteristics to compare them. This is not the difficult part – this can be achieved some how in most cases. The crux of the matter is – can you explain the variance in data that these solutions throw at us? </p><p> </p><p>Let us take an example – last year, it took 5 days for a PR to become a PO on an average. And year before that it took only 4 days on an average. Why? Does the report tell you the reason why this happened? If it does not give this answer – is there any justification in calling this business intelligence? where is the “intelligence” here?</p><p> </p><p>This is not an easy question to answer – there are several reasons possible as to why the PR took a day longer to become a PO this year. Let me throw a few out there – totally arbiatary figments of my imagination.</p><p>1. Approvers don’t understand the cost of waiting – indicating maybe a change management issue or a training issue</p><p>2. Lack of streamlined approval process – indicating an inefficient process design</p><p>3. A new manager took over procurement department this year, and he was not as good in following up as his predecessor was. </p><p> </p><p>Let us say that after some analysis – we identified that the reason for this as “1. Approvers don’t understand the cost of waiting”. So we fix that – and get every one trained, and hang banners all over the place urging them to do better and all that :)</p><p> </p><p>A year later – we find that the average time is now 6 days. We are back into analysis mode – and this time we found that it is because a large number of PR got raised in holiday season, when approvers were on vacation. </p><p> </p><p>You get the idea…this approach has a lot of pit falls.</p><p>1. We re-invent the wheel every time we see a variance. We are not managing by exceptions – the report does not break down the data by potential causes we identified in last iteration.</p><p>2. There is no systematic way of storing the result of past analysis. </p><p>3. We don’t even know if it is worth analyzing this variance . Is it such a big deal that POs took a day longer to get created on an average? is the cost of analysis more than the cost of delaying POs by a day? Why was the decision made that 4 days was the acceptable limit for time a PR needs before some one creates a PO out of it? why not change it to 5? – and then there is no longer a need to waste time on complex analysis !!</p><p> </p><p>In most organizations that I am familiar with – there are pretty good reporting solutions that show them WHAT happened. They have (or can have) the ability to see the data and slice and dice it in any way they care to. But the moment, the data shows a variance – they don’t have a good process of handling it. As a result, they do the same analysis repeatedly. It is not easy to answer “WHY” questions. </p><p> </p><p>Is it even possible to have a solution that answers the “WHY” questions?  My answer is “yes – to a certain extend”. Does this mean our existing reporting solutions are useless? Not at all – they are the starting point. With good design, we are already able to show clearly as to “WHAT” happened. </p><p> </p><p>So, what can we do to extend our existing solutions to answer the “WHY” questions? </p><p> </p><p>First of all – we need to realize that the idea is NOT to automate all such problem solving completely.  It is impossible to identify all the causes for a given effect. All we can do is to see if we could identify a good number of possible causes upfront so that a systematic solution can attempt to correlate the effect to these causes. If no correlation exists – then it makes sense to identify this as a valid exception, and find a solution manually. But it does not stop there – the moment we identify the cause, we should be able to identify this as one of the probable causes for next time – so that we don’t waste resources redundantly on the same issues repeatedly.  </p><p> </p><p>Let us talk some more next time and see if we can come up with an approach to keep “Intelligence” from taking the divorce route :)</p><p> </p><p> </p><p> </p><p> </p><p> </p><p> </p><p> </p>

You must be Logged on to comment or reply to a post.
  • We can add  (in this example scenario) that it isn’t (or at least shouldn’t be) a given that shorter approval time is a worthy goal. What if it didn’t matter if approval time is high as long as it is less than say 7 days? (just saying).

    It will take a combination of huge abundance of meaningful history data, a good understanding of behavior science, proven measures, and sophisticated applications to start getting to the beginning of intelligence. At this time we are wanting on all fronts (except in marketingSpeak – there every vendor is ahead of the curve).

    • LOL – of course in marketing speak, you can just ask a “WHY” question to the system in a language of our choice and it will sing you back the answer 🙂

      But jokes apart – there is some very promising research on this, including at my own employer – just that I have not been able to convince my buddies doing that work to come out and blog on it !

  • Hi,

    Thanks for sharing your thougts.

    So the idea is to have BI answer almost all low-level questions! Let us review your example:

    First year PR->PO conversion took 4 days. Second year it took 5 days. Analysis was made and discovered that the approvers didn’t know the cost of waiting. What process was used to figure this out? Let us say the organization used the following process:
    1) They reviewed the sub-steps involved to create a PO from a PR.
    2) They compared the average time(for all documents generated) taken for each sub-step with the last year’s.
    3) Let us assume, they discovered that few sub-steps took longer this year than last year.
    4) At this point, they somehow figured out that the approvers took longer this year than last year.
    5) They interviewed the approvers to understand why they took longer to approve. I am sure they shared several reasons as to why approval took longer.
    6) The organization compiled all responses and after few meetings, they probably concluded that the approvers didn’t understand the cost of waiting.

    Ok. Now we know why it took 5 days this year, user training was the cause. Now I am not sure how we can use this cause in future without implementing an application to capture all information that we gathered in steps 1 through 6.

    In other words, let us say PR->PO conversion took 7 days, 2 years from now. If the system contained all information that we gathered during the analysis(steps 1 through 6 above), then yes, it could answer WHY. What would be the cost of implementing an application to capture such information.

    Am I missing anything?

    My perspective is that BI should answer as many questions as possible but not all questions – never. In my opinion, the system would never be able to understand everything happening/happened in the organization’s EcoSystem. And without this information, I am not sure we can guarantee the accuracy of cause reported by the system. Innovation is occuring in every way of our lives so the moment we think we captured the latest information in our system, I am sure something has already changed in some corner of the globe.

    If your intention is just to list the known causes for a given effect, then yes, it would be useful. However the user will have to perfom an analysis to link a cause to the effect.

    Bala Prabahar

    • hello Bala,

      We are on same page here –
      Every time we do some causal analysis – there should be some way of recordng it in a fashion where there is a possibility to reuse in future.

      The idea is that once you find a cause – then you somehow need to make sure that you track it for future – and if the issue arises again, you should be able to confirm or eliminate all known causes quickly – and you will need to do analysis only in case you eliminate all known causes.


  • Iam not an expert in the BI part yet but my colleagues often ask me the diff between BW and BI and why we call BI in SAP ? Atleast from the BI content i developed I could never see why it should be called intelligent .
    • In the context of SAP – they just changed the name of the toolset back to BW from BI. People in the field use these two acronyms interchangeably. There are plenty of discussions in SCN on this topic.
  • Hi,

    I enjoyed reading the article and  it raises some important questions.

    Just using the example provided, I am wondering if we are defining the effect correctly. To me, the effect is “Delayed payments to contractors” or “Loss of Discount” etc. The cause could be “Higher processing time for PR”. In the next level of iteration the cause from previous cycle becomes the effect and then you are looking for causes at a much deeper level.So the cause-effect could be iterative.

    My point is…BI applied at the right level of the value chain can actually point to underlying causes. The fix may require deeper probe which requires a wholistic review of processes and sub-processes in question.

    What would be useful is the application of Business Intelligence in a self-diagnostic mode. This is one way of applying the concept of BPM.

    For example, in the same scenario, the process would be designed in such a way that the PR process gets escalated through the workflow the moment a particular threshold is exceeded for waiting time so that the process can be completed within acceptable time limits set.

    If you add ‘predictive’ capabilities to the process and corrective course of actions, one would be minimising exceptions. I think this is where Business Intelligence becomes valuable at the operational layer.


    • How we describe the effect is based on the ocntext and the user. To an operations person, it is the “delay in payment to vendor” a finance guy, it is “loss of discount”…to the man who raised the PR, it is “delay in starting work”..and so on.

      We try our best to abstract in most cases – but the common approach we take is that we try to get a “one size fits all” abstraction, which does not make sense in real life. For example – in the case of a doctor and patient, what makes sense to one might not make any sense to the other – and both parties are looking at the same data :). It might make sense to have multiple abstractions and/or a hierarchy of abstractions in some cases.

  • Vijay,
    I loved the Blog as it was very insightful and am able to associate with lot of the questions posed by the Customers.

    Best Regards,
    Navin Swaminathan
    Practice Lead-BI
    Bigtec Consulting

  • Hi Vijay,
    Nice Blog..
    As you said current BI systems are capable of answering “What Happened” but they are totally oblvious about “Why it happened”. Essentially all conventional BI tools are shooting “SELECT” statement to the database so obviously it would just analyse the data.
    To incorporate the answer of “why” in BI tool I think we need to include Neural Network techniques heavily into each BI tool. I feel existing Mining tools are not matured….


  • Hi Vijay,

    Superb content. Really straight to ‘what’ are we doing with BI and ‘why’ are we simply stopping @ that. Completely agree to the fact that its is about not just finding an issue, its about how do we go about resolving it…!

    – KunalG

  • Thank you Vijay,
    made me think about my job perspective again. I would like to kow what the main prupose of a BI is then? Is it a wording issue for you? Would Data Warehouse sound better then? Drop the intelligence and keep a fast Data Analysis tools.

    Better is to develop an intelligent path through data. Either during implementation with a persistent “WHY” question or a better tools. I think it is not possible to program better tools today, since a good vision is missing. I think it goes down to each Analyst, who has to keep asking “WHY”.

    But one Note to your comments. The Intelligence remains where it should be: The User, not the tool. Only the combination of Structured and unstructured (plus intuition) will lead to an Intelligence. Not the tool itself.


    • You are completely correct,Joerg –
      1. “it is the man, not the machine” that holds the intelligence. 
      2. Before we start programming a new tool, we need to get our vision right.

      I think we have a general tendency to oversell BI, and I believe we need to step back, and set expectations on what it does and what it does not. once that is done, we can hopefully fix the parts where BI is lacking today (like the WHY questions. My blog was just a humble effort to convey that message.

  • This blog is similar to the disappointment with hyped technologies like say Operations Research. It happens a lot in Motivation and behavioural management like Transaction Analysis when you realize that “motivation” helps more the sane and not the insane.

    People see the figures and then look for human reasons. For example PR to PO time increase may be due to loss of experienced staff and influx of new staff. No algorithm in BI will tell you how staff attrition and new staff coming up to speed will affect a KPI.

    Much like SAP itself BASIS->WAS->Netweaver with truth being 90% of business logic in good old ABAP engine. And you can survive will without J2EE!

    Intelligent conclusions will be in the mind of the beholder of the BI report. Renaming BW as BI does not make inject intelligence.

    Top management will never be guided fully by abstruse sciences like Neural Networks.


  • Fellow colleagues,
    Call me old fashioned but I still believe Business Intelligence is about process and the related management of intelligence, in the true of the word. Without getting bogged down on definitions, if I may paraphrase Sun Tzu; “If you know yourself but doesn’t know your enemy for every victory you will suffer a defeat, if you know both yourself and the enemy you won’t fear the outcome of a thousand battles”.
    The key words here are:
    – know – primarily in likely outcomes of FUTURE actions; past performance does hold the answer for a lot of questions and is an integral part but the main strategic objective is what to expect TOMORROW on the plains before my enemy
    – Yourself – what are the strenghts and weakness of your organisation? Of course this has to be put into perspective of your business scenario, with regulations, competitive pressures, etc
    – Your competition – I think most discussions about BI miss this crucial point. The whole point of having all this is to overcome and thrive over the competition

    So, let’s see the forest for what it is; a complex environment where knowledge/intelligence provides the guidance to identiy trees and to find the “correct” path forward.

  • Hi,

    I agree with your statement that being able to analyse variances is not the same as explaining the why and therefore there isn’t “intelligence” in the business intelligence solution.

    Regardless of the term BI, it is my belief that explanations of the “why” require human intelligence that can choose to learn from past/historic experience but can also modify that to deal with new situations.

    Therefore, the role of BI in my opinion is to support the human in that task by providing information “assumed” to be relevant but also by being able to provide information asked for by that human.

    BI should also be aware of the limitations of human information processing in particular when presented with too much information / data / options. Therefore, another role of BI is to keep the irrelevant “noise” away from that human until asked otherwise.

    Applied to your scenario that could mean:
    1) Noise Filter
    The system has an understanding whether or not that variance is worthy of analysis.
    What is the impact of that variance on the KPI tree? Does it mean that any of the “important” measures further up is going to deteriorate?
    What tolerance information is available from the business process design – did the process assume a normal range between 4d – 8d?

    2) Meaning
    If the variance is outside the tolerance, than the system should try to “understand” the variance a bit more from a semantic perspective, e.g. what business process steps are behind it, what data is available on these, and is there a correlation between the variance and any of the underlying data (human approval duration changed all other process steps similar).

    3) Information Pre-Selection based on Meaning
    Using that “understanding” the system can go through recent and historic data to find similar or related variances and may be records of the “why” and what the “resulting action” was (Knowledge Management & Semantic Search).

    4) Information Presentation & Analysis
    Presenting the information selected back to the user in a way that allows him to explore it and also understand its relationship to the variance to be analysed should support the human in assessing the why.
    However, there is still a chance that the “why” for this particular variance may be a new one.
    As a result the user may ask for other information, e.g. staff movements in that area, holiday plans etc.

    Broken down like this, one could clearly argue that steps 1, 2, and 3 are indeed representing some kind of intelligence.
    Unfortunately, these are not necessarily implemented yet in common BI solutions and I do not believe that they have filtered yet through into BI strategies.

    Question to all – do you think that we should be making a greater effort to move towards these ideals?


    • Excellent comment – and thanks a lot for posting it.

      I want to touch up on the information presentation part for a little bit, especially since this is something that can be attempted relatively easier than the other things we discussed.

      The average report has too many fields that it becomes useless for users. On the other hand, users like to get a lot more fields on each row than they really need. It is usually out of ignorance that BI can expand the data for selected records and show more data. As BI practitioners, we owe it to users to design better reports – that show data in a straightforward fashion (less noise), and training the users on how to drill down for more information. Of course I assume here that the warehouse has a good enough data model that support this – and I fully am aware that this is not always true.

  • Vijay,

    I see your point of view, but what I see is a skewed point if view. The “intelligence” in BI does exactly what it is supposed to do, give you intelligent answers to complicated questions, which in your case was “how many days does it take a PR to convert into a PO”, a.k.a. a PR-PO SLA for a given organization.

    If you really want to know why thi took time, you may want to break down your KPIs into smaller chunks and look at the time it took for that PO to get to the PR, for example, a KPI to measure the wait time.

    In your blog you also brought up a question about this “intelligence” being used for predicting the future state of this PR-PO process. Now, for this, there are other modeling tools available that use Data mining and predicative analysis to get you the desired results, and where BI is only the source for these tools.

    So, in a nutshell, if “intelligence” was expecting BI to predict the future or for it to answer questions that were not captured in the KPI set, then I have to doubt the intelligence of “intelligence” and your premise in creating a rift between it and BI.

    Happy blogging!!!

    – Vinay