Anomaly detection system identifies outliers in data using mathematical models, correlates it various influencing factors and delivers insights to business decision makers. Real-time anomaly detection techniques have application in various industries like telecom, retail, fintech, manufacturing, defence, healthcare and others.

Frank Bass of Purdue University developed a simple model (Management Science, 1969) that describes the process of how a new product introduced in retail are adopted by the population. This model known by the name Bass model is widely used in forecasting sales. This model classifies users as innovators or imitators. Innovators are the early adopters who try the product when it goes live. The imitators observe the market and buy the product after considering the feedback from the innovators.

Source: Frank M. Bass, Management Science, Vol. 15, No. 5, Theory Series (Jan 1969), pp. 215-227

Although this model gives a smooth prediction, the original paper by professor Bass clearly shows that the actual sales is not as smooth as the model prediction. The sales goes through a trough before it picks up. If businesses can identify this anomaly in timeseries data and uses marketing levers to improve sales, then this trough (red circle in the picture) will not be this deep and would lead to increased sales.

One of the primary reasons why businesses experience this trough before they capture the market share (driven by imitators) is tracking of average KPIs. Key performance indicators are typically captured at a high level because of which micro trends average out and lucrative microsegments are missed out. And tracking KPIs at a granular level is humanly impossible.

Another issue with identifying this trough is typically analytics engines provide insights in hindsight. It used to be expensive to track the KPIs in real time because of computational complexities.

Also, the competition is always watchful and when a product becomes successful, they grab a share of the potential customers.

These issues can be addressed by

• Tracking the KPIs at a microlevel identified using artificial intelligence
• Protecting the managers from irrelevant information and providing them with valid anomalies to invest their energies in the right place at the right time
• Using the latest techniques to identify anomalies in real time and being alerted while it is happening
• Acting on the information before the competition does and capturing the market share in those microsegments

While present analytics systems and products gives you insights into emerging trends in the market they do not tell you in a granular level. Let us take a hypothetical example of an e-Commerce site that sells fashion merchandise. Imagine that suddenly the sales of T-shirt that has a funky start-up caption is unexpectedly high in Koramangala area in Bangalore. And it is correlated that Koramangala in Bangalore has many technology entrepreneurs and the T-shirt has found early adopters with those who are associated with technology start-ups. If the retailer has this information and sends an email campaign to everyone who is a technology entrepreneur, the sales of that product will increase. Marketing will act as a catalyst and deliver a smooth transition between innovators and imitators of the product, as against the usual dip. From the chart, it can be observed that by detecting the anomaly and identifying the new trend before competition can deliver a sales increase by about 25 – 30% in that product category.

Retail Anomaly detection platform by CrunchMetrics has the artificial intelligence to watch every SKU at a granular level, identify anomalies and correlate with the bigger picture and deliver insights in real time. When the marketing mangers use the alerts delivered by the deep learning powered AI engine, flexes their campaigns and targets the right market segment at the right time, sales will assuredly increase. This sales increase will happen even before the competition has the slightest of the clue. With the power of real-time anomaly detection techniques, a retailer can increase sales and beat the competition.

Maximize E-commerce Conversions During a Holiday Sale

Download Free eBook

References

1. A New Product Growth for Model Consumer Durables Author(s): Frank M. Bass, Management Science, Vol. 15, No. 5, Theory Series (Jan., 1969), pp. 215-227

Kumar S

Kumar is a principal consultant at CrunchMetrics. He is an alumnus of IIT- Madras and IIM- Calcutta. As an entrepreneur, he has co-founded an analytics company and then an omni channel retail company. He has worked in advisory roles for Fortune 500 companies such as Deloitte and Tesco in various multinational locations. He has also worked in technology roles for MNCs such as Cognizant and Virtusa. He is a Good Reads author with the pen name Khun S. Kumar and has published seven novellas in Amazon.