What if your BI solution auto-discovered metrics and the dimensional structure from within the data to be loaded?  What if it then used this information to programmatically configure the dimensional hierarchy within the BI solution?  What if it continued to perform this discovery operation and then synchronized the dimensional hierarchy immediately prior to each data load, including aligning with dimension structure changes?  What if the dimension synchronization and data load happened immediately?

If you are in IT or a business analyst it would mean having more time to work on supporting the business rather than the ongoing effort of maintaining cubes and updating data loads.  If you are a business person, it means less costs for IT maintenance, better support from IT and analysts, and fewer errors due to the automation.

And if you don’t have business analysts or a large IT staff?  Even better as the automation means you won’t require them and can carry on with your business using the BI capabilities to improve decision making!  This does require that you have intuitive, interactive analysis and analytic capabilities that are easy to use and consume – I’ll cover that in another post.  If you are interested in knowing more, contact us.

This delivers the benefits of cloud-based ubiquitous access and ”new generation” capabilities such as mobile support, advanced visualization, interactive analysis and sophisticated analytics to a broader audience, without the trauma of a “heart transplant” for core transactional systems.  Of course there are integration requirements the new capabilities must address and these will be described in future posts.

What are your thoughts?  Are you delaying the move to cloud-based core systems?  Does it make sense to leverage your existing on-premise systems and extract more value by delivering information access, planning and visual analytics to a broader audience to help them in their daily operational decisions?

I was surprised and very happy to see a screenshot of myDIALS on page 343 as an example of a metrics dashboard – it was taken from our Online Marketing Module.  This is great visibility for us at myDIALS and we are very grateful.  More importantly, the book highlights the importance of metrics in cost-effectively scaling a business.  Building a startup company is tough (I know from personal experience) and trying to do this without the ability to visualize, report and analyze appropriate metrics makes it much harder.

What the static screenshot doesn’t show is the capabilities that turn those metrics into real insight and actions:

• the ability to interactively filter, drill into the data, understand the formulae used to calculate the metrics etc
• alerts that proactively notify you of issues you need to address now, or warn you of potential issues on the horizon
• analytics such as trends, forecast projections, variance analysis, control charts, histograms and correlations that can provide insights derived from the metrics
• what-if scenarios to see the impact of potential changes before you take actions.

The book suggests hiring a “data chief” to manage the data and its interpretation, and to quarterback the company’s plan to use that data to drive continuous improvement.  This is one area I have a different opinion to the authors.  While a data chief can certainly help manage the data, metrics should be available to everyone in the organization making daily decisions.   If equipped with the capabilities I’ve outlined above, each person can interpret the data themselves and directly drive continuous improvement, which I believe is far more effective and scalable as the company grows.  What do you think?

• Gross Margin
• Productivity per head
• % Quota Achievement
• On-time In-full Order Fulfillment
• Overall Equipment Effectiveness.

Some of these calculated metrics (there are many other examples in addition to those listed above) are quite clear, consistent and well understood.  However there are nuances in many that means different people could interpret them differently, for example:

• Does Gross Margin include all variable costs or just material costs?
• Is Productivity measured across all employees or only sales people?
• Is on-time in-full measure on individual line items or across complete orders?
• and there are many variance on OEE calculations.

To ensure everyone has a consistent understanding of these types of metrics, it is important to have an easy way for them to view the formula used to calculate the metric including any filters that have been applied to any of the component metrics used in the calculation. Those same users can be empowered with even more insight if they are able to interactively drill into, or expand, the calculated metric to show the formula and the values of each component metric.  If in turn, this next level of metrics are fully interactive, including the ability to drill into the formula and components of any sub-calculated metrics, we can quickly gain a complete understanding of the KPDs or Key Performance Drivers behind each calculated KPI.

I believe this is a very fundamental analysis capability – at myDIALS we call it “Expand by Drivers” - that is key to enabling people to fully understand the performance drivers of their business or operational area, helping them to make decisions and take actions that use this understanding to improve business results.

• latency introduced by having to move data into a different data store before it can be analyzed;
• inconsistencies in data structure whenever there is a structural change to the source system (simple examples are adding or removing customers or products, or changing the organization structure);
• delays in making the analysis services available to new users of the underlying systems.

To address the first point, source data has to be loaded into the analytic data store as frequently as required for good decision making; which can vary from immediate (triggered by a change in the source system) to every few minutes, every hour, every shift or every day.  The legacy approach of crunching cubes over night or on weekends introduces significant latency and is not suitable for most of today’s operations.

The last point can be eliminated by using a central user information repository, user synchronization and single-sign-on, and these capabilities have been available for some time.

The middle point is very important as most organizations change frequently, not in terms of daily re-organizations although most companies seem to re-org every several months, but certainly in terms of new customers, products and employees.  If the analytic system has to be manually re-configured each time these changes happen, it will quickly become a maintenance burden and probably lead to data inconsistencies.   The answer here is the metadata contained with the source systems or a master data management system that can be extracted by an intelligent connector layer and used by the analytic system to auto-configure and dynamically synchronize the data structures and dimensional hierarchy.  Not only does this ensure data consistency it eliminates the maintenance burden thus reducing costs.

Before implementing a business analytic system, it would be helpful to explore the three points above to ensure you will not incur significant maintenance overhead and that business decisions will be made on accurate, consistent data.

It seems to me there are several aspects that affect data value:

• relevance;
• importance;
• timeliness;
• granularity; and
• age.

A number of these aspects are inter-related in various ways.  For example:

• to be valuable, data must be relevant and important;
• the value of timeliness is directly related to how the data is to be used (more on this later);
• the value of detailed data granularity typically lessens with age.

My assertion is that while new technology enables us to hold far greater volumes of data, we should really focus on holding valuable data so that we don’t get bogged down with the problem of having “too much data, too little valuable information, and no real insights”.  As I’ve indicated above the value of data will vary based on how the data is to be used.  To simplify this, I consider the following categories of data usage:

• Data that is used to control a business or operational process;
• Data that is used to review the efficacy of a business or operational process;
• Data that is used to improve or re-engineer a business or operational process;
• Data that is used to predict future outcomes of a business or operational process.

Process Control:

To be valuable, data must be relevant to controlling the process, must be extremely timely, and must provide immediate feedback that can be used to make adjustments to the process while it is running.

Process Review:

Here the data must be important and relevant – typically we examine Key Performance Indicators (KPIs).  The timeliness must be such that process outcomes can be measured consistently and accurately without interference from “noise”.  Similarly the data granularity must be such that a clear picture of process efficacy can be easily determined as compared to budgets, targets, best practices, prior periods etc.

Process Improvement:

Similarly to Process Review, we need relevant and important data, but now we need to examine the Key Performance Drivers (KPDs) and not just the outcomes or KPIs.  I’ve previously discussed KPDs here and  here.  We need to be able to “drill into” the data to get more granularity as required to identify the root causes of anomalies.  Timeliness is not quite so important and having data from multiple periods may be important to establish patterns, seasonality, deterioration over time etc.

Process Prediction:

To predict future process outcomes we must build one or more scenarios into the future based on assumed or quantifiable inputs.  So we need relevant, important historical data to build a baseline model along with a clear understanding of the relationships within the data – particularly between the multiple drivers and subsequent outcomes.  Data granularity is important but only to the extent that it improves the quality of the model (too much granularity may require too many assumptions to be made and with each assumption there is increased variability being injected into the model).  The use of historical data patterns also needs to be carefully considered in order to ensure that those patterns remain valid into the future – that is there has been no fundamental shift in the process being modeled or the internal or external environment that would render past data irrelevant or misleading.

So there is a value equation that we can use to determine what data to hold based on the intended usage and the attributes of relevancy, importance, timeliness, granularity and age.  I suggest that using these factors will not only reduce the amount of data we need to hold, but will significantly improve our ability to successfully use that data to derive insight and better control, review, improve and predict outcomes of business and operational processes.

This approach can be summarized as:

• start by understanding the business problem you are trying to solve;
• then determine the underlying business processes that either contribute to or are impacted by this problem;
• now we can identify the people who have responsibility for overseeing or executing these processes;
• then we can examine what decisions they make as part of their daily operations involving these processes;
• and finally we then ascertain what information, interactive analysis and analytics would help them make better decisions more quickly.

It is only after working through the above that we should look for the appropriate data sources to supply the required information.  Hence we start from the “outside” (the business aspects), and then move to the ”inside” (the data required to support improved decision making relating to those business aspects) rather than the other way around.   My strong belief that this “outside-in” is the most effective way to approach performance management.

As I’m writing this, the words of the Traveling Wilburys are streaming through my head – “Don’t it make you want to twist and shout when your inside’s out?”

Although the focus of the article is on analytics associated with big data applications, there are many applications for real-time and near-real-time analytics.  These include aspects relating to more efficiently conducting business transactions, including:

• before transaction analysis such as customer targeting based on behavior and demographic analysis;
• in-transaction analysis including dynamic offer making, fraud detecting based on correlation and pattern recognition;
• post-transaction analysis such as profitability analysis,  inventory ordering and supply chain management.

Analytics can also be applied to optimizing business processes and operations, including:

• Identifying the optimal channel mix and funding associated marketing campaigns and promotions;
• Balancing the product manufacturing mix based on projected demand;
• Optimizing the use of resources in a facility such as education or medical clinic.

I’m encouraged that the power of analytics is being recognized in the media as this will help build awareness and adoption.  As more companies empower their front-line workers with immediate, up-to-date data, interactive analysis and dynamic analytics, they will improve daily decision making, optimize their operations and deliver better business results.

• A list customers that placed an order in the last 5 days;
• A list of all students currently enrolled in a particular class;
• A list and quantity of products currently in backlog;
• A list of customers with payments outstanding for more than 60 days;
• A list of opportunities scheduled to close this week.

Another way of looking at this is that reports provide “what is” and “who are” type answers that are used to take tactical, operational actions.

Analytic dashboards are designed to answer multiple, related or more sophisticated questions and to provide greater context, historical trends,  future projections and statistical analysis in order to make decisions about how to improve operational and financial performance.  Some examples might be:

• How are the characteristics of the best performing sales people different to those who aren’t performing as well?
• What has led up to the current quality issues we are experiencing?
• What will happen to daily production if we continue to experience these quality issues?
• Why are our sales conversion rates lower than industry norms?
• When will we run out of cash at the current rate of cash burn?
• What would happen to overall gross margin if we could lower our COGS on this product line?

As you can see, there are “what will”, “what was”, “how”, “why” and ”when” type questions and are used to improve decisions and determine new actions and processes, rather than simply providing data to enable current processes and actions.

There is definitely a place for both traditional reports and analytic dashboards.  There are some simple guidelines that help determine when a report is the correct medium and when an analytic dashboard makes more sense, such as:

• If you have a large inventory of saved queries and reports, then you are trying to answer a number of questions that are probably related in some way and an interactive dashboard might make more sense;
• If you run a report and then export the results to a spreadsheet so you can apply more analysis, an analytic dashboard would be more efficient;
• If you run a report and then utilize a charting package to visualize or graph the results, it would probably be much easier to deploy a visual dashboard;
• If you are trying to visualize thousands or even hundreds of thousands of items in a dashboard visualization, then this would most likely be more appropriate for a report;
• If your visual dashboard has lots of large tables, then reports would more effectively produce these results;
• If your dashboard doesn’t have any analytics and is only used to view flat, single-dimensional data that is readily available from standard reports, and there is no interactive drill capability, then simple reports with embedded graphs would probably be more appropriate.

I hope this has provided some clarification as to the inherent differences between reports and dashboards.  Each has its own place in enabling every day business operations and in general:

• Reports are best at providing data to enable people to take operational actions; whereas
• Interactive, analytic dashboards are best for enabling people to derive insights in order to make better operational decisions.

• greater ability to get information on time;
• increased self-sufficiency and a greater percentage of power users;
• that users had greater capabilities to drill into detail, filter and explore data and tailor their experience.

This certainly mirrors what we perceive in the market and the capabilities we strive to deliver.  Previously I have described the need for BI to operate at “business speed“, personalization and the concept of interactive investigation of data.  All of these are key to delivering the requirements for rapid business insight within today’s organizations.  As Aberdeen says, the decision window is shrinking which presents a challenge to IT departments and visual/interactive BI addresses this challenge and  increases the business’ ability to meet the shrinking decision window.