Data Anomaly Detection Best Practices

[OneandOnly13]

Hi Everyone,

We're looking to start using alteryx for data anomaly detection across many of our datasources. All of these would have daily data and we'd be looking for trends that would indicate something is wrong in one of our ordering systems. We're thinking of using the time series forecasting to compare the latest actuals to prior time periods' forecasts as a starting point but wanted to see what others are doing first. Thanks for any ideas!

[tdsanalytics]

We use SQL to calculate daily z-scores (number of standard deviations +/- the mean) for a rolling 30d trend for various metrics (revenue, website performance, etc.) The z-score threshold for counting an underperforming day is -1.64 or -2. In other words, it looks at yesterday and determines if the KPI is abnormally below it's performance in the last 30 days. It also calculates the number of days the KPI is below the threshold; when it gets to 3 days, an email alert is sent.

We not only do the above for the daily linear trend, but we do the same for the trend of the YoY variance.

[OneandOnly13]

Thanks! This sounds like a great way as I would be using it for cross channel comparisons as well as other digital metrics. We will definitely take a look into how our data would fit into a similar model. I appreciate the response!