Cognitive Trend Detection
The only certain thing about the future is that it is uncertain or so said Albert Einstein.
Any prediction model therefore relies on charting the future based on assumptions of the past. Most of these models are deterministic models (simplified, it’s simple rules, like if/then removing randomness from the prediction). The tricky bit in this, is the fact the underlying data quite often has randomness in it. The catch these trends or anomalies with deterministic models produces either too many results and false positives or doesn’t detect the real outlier.
So here’s our take on trend detection as a first step to a full prediction suite*:
Squirro’s smart filters already deliver contextual insights — that is real-time insights based on a sophisticated algorithm that adapts, ‘learns’ and evolves cognitively in parallel to use.
Time series without trend detection
The core principle is that we replaced a deterministic model with a probabilistic model. Relevance in search effectively is a game of probabilities. This means searches, which are based on both structured data as well as unstructured data, are never stagnant or static, but can be uniquely relevant to an individual, a group or an entire organization.
The new trend detection analysis puts the parameters of these bespoke smart filters in an historical context to identify anomalies across nominated time increments. What’s more, the underlying algorithm accounts for the seasonality of data — for example, what constitutes an anomaly on a weekend, may be radically different to what defines one during a typical workday — knowing the difference at a granular level, trends can be identified with the benefit of this context intact.
Time series with trend detection
What does it all mean for business?
It means the Head of ITSM will be able to respond to the next major service incident in good time; it means the Wealth Manager will place his critical client call when his client is most likely to answer; it means you can be informed of the next opportunity or crisis your business faces, as it develops, not after it has happened. Unlike typical threshold breaches, where a tiny infringement can trigger what amounts to a ‘false alert’ (and consequentially the alert fatigue that comes with it), Squirro’s trend detection takes into account the context to determine whether or not this is a true anomaly.
For service-driven enterprise environments looking for greater accountability, a reduction in SLA penalties and minimizing the risk of downtime, the commercial efficiencies and financial rewards of detecting trends are easily quantified. Team can mobilize quickly and respond to impending threats before they escalate. For environments running IT as a service, this degree of insight can assist in capacity planning requirements.
Financial services, telecoms companies and other businesses that rely on customer engagement can use trend detection to achieve new value with their own customers, and with the insights around breaking industry issues, competitive analysis and other such catalyst information to make more astute, timely and commercially rewarding decisions.
*) We’ve released the cognitive trend detection feature with the latest version 2.3.0. Get in touch to have a test run.