Explore Yellowfin now with our sample datasetGet Started
Automated analytics is a process whereby machines are used to automatically analyze data looking for significant changes in data instead of traditional manual data discovery methods. Analysis tends to be time-series based and centered on key changes such as finding changes in total and average, trend direction, volatility, step shifts and outliers (spikes and drops).
Typically these processes augment traditional methods such as dashboards and reports providing a richer analytics experience for end users.
Yellowfin has developed advanced machine learning capabilities that assist users in the discovery and interpretation of data insights. These capabilities augment the capabilities of the user and enable them to analyze more data, find more insights and understand root cause – far faster than a human alone ever could.
Yellowfin Signals can be used to automate the process of discovering important insights in your data. Signals leverages the meta-data layer and can be easily configured to continually scan your data, identify changes that exceed thresholds, and create personalized notifications to your users.
Signals deploys a variety of complex algorithms aimed at detecting interesting and relevant changes in your data. This includes changes in total and average, trend direction, volatility, step shifts and outliers (spikes and drops). Signals algorithms can automatically detect and allow for natural seasonal variation in data. Signals setup is simple, however if required, fine-grained configuration of thresholds and algorithm parameters is available if desired.
Signals can be configured to simultaneously monitor many metrics and across many dimensions – and thus scan and examine far more data than a human ever could. Signals deploys a complex ranking algorithm to determine the most relevant data events, and then further ranks the data for each individual user based on that user’s preferences and previous system usage. This ensures only the most relevant and important insights are alerted to the user.
Signals can be used for any data that has one or more metrics, one or more dimensions, and is stored over time. Most data setup for Analysis purposes is set up in exactly this way, and as such Signals can be used for a huge variety of use cases. Here are just a few examples:-
Generally it makes sense to schedule the jobs to run immediately after the data is refreshed, that way you will be notified of any significant changes as soon as possible.
Signals jobs should be configured to run no more frequently that your data is refreshed. For example, if you are connecting to a Data Warehouse and the data is only refreshed weekly, there is no point scheduling the Signals jobs to run more frequently than that – Signals won’t find anything new as the data has not changed.
There are a number of things that can be done when configuring a Signals job that will ensure only relevant Signals are generated, these include:-
After a Signals job has run, the system uses a ranking algorithm to sort Signals of different types, so that the most important Signals are prioritized for delivery. In addition, the Signals algorithm looks at the preferences and behavior of individual users and will weight a Signal based on its potential usefulness to a user. Signals looks at the Signals that a user typically opens and interacts with, or has marked as useful – and then weights the attributes of those Signals more heavily in the ranking – dimensions, metrics and Signal type. The more a user uses Signals, the more information exists for the system to perform better ranking.
The signals interface includes a range of analysis to help a user understand the cause of a specific Signal. This includes:
Signals can be notified via a timeline notification, email or via the Yellowfin mobile app. Once a Signal is delivered, Yellowfin provides additional information to assist the user in understanding the root cause of the data event – including identifying other highly correlated data events from the same or different data sources.
Collaboration features allow Signals to be shared, discussed, followed and added to Data Stories. The Signals workflow allows Signals to be assigned an owner, and managed via a workflow until a conclusion is reached and actions agreed.
A Signals Widget object is available in the Dashboard builder. The Signals widget can be connected to the output for a particular Signals job, and can be further configured to show specific signals from that job. For example – I could add a Signals widget to my Marketing Dashboard that was linked to Signals generated off web-site traffic data.
Users can click on any signal displayed in the widget, and open that signal for full exploration.
Alert based reporting is typically set up on an individual report. Signals are set up to scan all of the data defined in a given view.
Alert based reporting typically has simplified rules – for example, comparing a single value to a defined threshold. Signals deploys a range of sophisticated algorithms designed to identify relevant changes in your data – including things like trend changes, volatility changes and outliers.
Alert based reporting typically is designed as a rule that will decide whether to deliver a specific report or not. Signals are ranked and personalized for each user in a system, and when delivered provide a wide-variety of system generated output aimed at assisting the user in understanding why a particular data event may have occurred.
Individual reports can be scheduled to be run and distributed on a regular basis. Rules can be set against those reports which define when the report should be sent. This allows for example, reports to be sent on an exception basis – say when a particular metrics exceeds or falls below a predefined threshold. Rules can be set on the total value of a metric in a report, or triggered when any row in a report exceeds the threshold.