Optimizing

Large amounts of data are being collected and analyzed. It is now time to bring it to a true big data scale to feed the data-driven models you are creating. Collecting user behavior on websites provides massive amounts of raw data. The key is refining this to user \textit{intent}. What does it mean when someone lingers on this product page? Is he doubting the price or the product model? If you find the intent of a user, you can act on it by personalizing their experience.

• Collect data like user behavior on a bigger scale.

• Use customer data to personalize the experience.

• Focus on data quality.

Reports are now standardized and dashboards are available to management. Derived KPI's, backed by analytical models, are proven to be very valuable indicators. The next step is to move this process to real-time operations. Real-time dashboards reveal operational insight, a real-time state of operations. Anomalies are quickly detected and resolved. Continue to show how KPI's evolve over time. It can be a powerful argument for continuing the investments in analytical efforts.

• Define and show operational metrics in real-time.

• Create a process to rapidly detect and deal with defects, anomalies, and trends.

As analytics continues to evolve, the required skills evolve with it. We move from basic data mining to more advanced concepts, such as machine learning. Machine learning is the science of training a computational model from past experience, to be able to match future events. We can use the large amounts of data we are collecting about customer behavior to attempt to predict future behavior. If we are able to predict future behavior, we can better anticipate the needs of the customer and act accordingly. To be able to do this, we need to have the knowledge to build and train machine learning models. The performance of the models is determined by the quality of the data and the model. A lot of parameters can be tuned, and sometimes intricate knowledge about specific techniques is necessary to build a performing model. The required knowledge and education level depends on how accurate the model has to perform. There are diminishing returns in the amount of effort required to improve the accuracy of a model. Netflix launched a public contest with a \textdollar1,000,000 prize to improve their Cinematch algorithm, an algorithm that predicts how a user would rate a movie \cite{bennett2007the}. Where a trivial algorithm has a error score (root mean square error) of 1.0540, Cinematch already improved this by roughly 10\% to 0.9514. It took 3 years and the combined effort of many experts on the subject to improve this score by another 10\% to 0.8567. Now for Netflix this improvement can be worth the \textdollar1,000,000, but it is important to always closely guard if the return will outweigh the additional effort. Humans do not always act rational, and you do not have perfect information about all the decision factors. It is therefore not always possible to achieve a high prediction accuracy.

• Use new big data sources in machine learning models.

• Use learning analytical models to predict user behavior.

• Use the predicted user behavior to optimize your offerings.

The technological challenge is to integrate these advanced machine learning algorithms in production systems. There is a massive amount of data and calculation necessary to train these models. Timing is a big challenge: New data sometimes has to lead to insight only moments later. The infrastructure has to be able to quickly transport data. Batch processing has to make way for streaming data. From a pull structure, i.e. the data is queried when needed, to a push structure: When new data is generated, an event, it is pushed and processed immediately. Event-driven architectures are build to react to and handle incoming events. Personalization is a big part of optimizing the customer experience. Automated integration of analytics allows us to adjust the customer journey for every unique individual. This is unfeasible to do by hand for everyone. Machine learning algorithms can prescribe the best journey for every customer. Data collection about the customer is needed to serve as input to the analytical model.

• Invest in technology that can automatically process new emerging data in near real-time.

• Invest in technology to automatically integrate analytical insight into production systems.

Leadership has the important task to lead the organization across the chasm from `doing analytics' to `breathing analytics'. Great progress has already been made, and the organization might feel it is no longer behind the pack. This may lead to a reduced sense of urgency and less commitment to invest and expand analytical activities. But, we do not want to use analytics to do the same thing, we want to use analytics to make a difference. To make a difference and compete on analytics, you need to continually improve. So this is a critical step for the organization: Are you content with the current situation, or do you go above and beyond?

• Bridge the chasm from doing analytics as a side activity to being a cornerstone of all processes.

• Encourage the organizational culture to be the driving force for analytics, instead of leadership.

Culture is probably one of the hardest things to change in any organization. Changing culture is a grind, and there is no magic potion. A culture that embraces analytics is necessary to sustain analytical activities. Leadership provides the runway for take-off, but Culture keeps it flying. At this stage a culture of continuous improvement, or \textit{kaizen}, should be stimulated. A culture where everyone is engaged to improve the organization goes hand in hand with our philosophy for an analytical organization. Analytics is born out of the need to break the status quo, and to look for opportunities to improve. Analytics can provide the tools to the organization to analyze the current situation, and improve upon that. Analytics can support a culture of continuous improvement, and that culture can in turn sustain an analytical culture.

• Stimulate a culture of continuous improvement through data.

• Encourage experimentation to test new optimizations.

Analytics is continuing to fulfill a more prominent role in the strategy of the organization. We started with analytics in a supporting role in the organization. First as a method to report on current performance, later to further analyze and improve performance. We are now seeing analytics becoming more integral to all processes by augmenting them with analytical insight. In the Reporting and Analyzing stages, we were concerned with how we can we use data to improve the blueprint of a process. In the Optimizing stage and beyond, we are increasingly looking at how can we use analytics to augment the working of a process. The difference is that if you were to stop collecting data in the first scenario, you would still have a better blueprint, say a better performing customer journey, that no longer requires data to further optimize. In the second scenario, the process would break down because you are actively using data, for example to recommend a customer the best product. Analytics has become an active component of the process.

• Position analytics as an integral component of global strategy.

• Use analytics to optimize existing products and processes.

• Use integrated analytics to create an optimized customer experience.

Analytical processes are getting more interwoven with business processes. Strong execution in the form of collaboration and communication are necessary to create a fluid process. Clear decision and responsibility boundaries should be in place. Continuously rethink if the right people have the responsibility over a certain domain. Agile organizations are able to change fast in part due to a culture that embraces change and innovation. Introducing analytics changes the a business process, so it could be beneficial to change the owner of the process.

• Define or restructure clear decision and responsibility boundaries.

• Embrace change in order to innovate and react faster.

The Optimizing stage focuses, for a large part, on the automatic integration of analytical insight. More analytics is done on the fly and directly available for use in production systems. Automation means speed, so turning data to insight to actions is getting more rapid. This requires a coordinated effort between analytics and those responsible for production systems. It is not only an analytical challenge as much as a technical challenge.

• Automate the use of analytical insight into existing processes.

• Focus on automatically turning data into actions instead of mere insight.

Employees will be further empowered in two ways. The first is having systems that can track performance over time. While the glaring issues have been tackled in the last stages by analyzing the current situation, the attention now shifts to optimizing the processes. By making small adjustments and tracking results over time, you can steer into the right direction. This is the central philosophy of continuous improvement, and this only works if you can look at the process over a longer period of time. The other improvement comes from the closer integration of analytics into production systems. This allows new data to be reported in real-time. Dashboards can be made that show the current state of operations without any reporting delay. New events can be quickly acted on or resolved before becoming a problem. Because the time, between the occurrence and resolving of an issue, is drastically reduced it becomes easier to also find the source of the issue. Employees should be encouraged to systematically eliminate the sources of these issues, also known as defects in Lean manufacturing.

• Give employees real-time insight into operations and performance.

• Give employees the information to improve their processes.