Intention Analysis - Business Use Cases - Aiaioo Labs

1 Intention Analysis - Business Use Cases Cohan Sujay Carlos Aiaioo Labs Bangalore, India Abstract Category Parent Interest Purchase Sales The team of AI researchers at Aiaioo Labs Sell Procurement works on advanced forms of text analyt- Inquire Sales ics such as intent Analysis . We launched Direct Operations our first products for intent Analysis in Compare Analytics 2011, and have studied the Business im- Suggest Analytics pact that these tools have had. In this Praise Opine Analytics article, we describe the benefits to busi- Criticise Opine Analytics nesses of using Intention Analysis , and Complain Operations we share some numbers for purchase in- Accuse Operations tention Analysis (courtesy of the team at Quit Operations ). Thank Express Training Apologize Express Training Intent(ion) Analysis Empathise Express Training Intention Analysis is the identification of inten- Wish Sales tions from text, be it the Intention to purchase or Meet User the Intention to sell or to complain, accuse, in- Advocate Analytics quire, opine, advocate or to quit, in incoming cus- tomer messages or in call center transcripts.

2 Table 1: List of Intention Types. The Intention analytics product offered by Aiaioo Labs can detect more than 16 intentions. Some of these intentions are listed in Table 1. Intention detection was of the order of 700 times Below, we describe a number of Business case or 70, 000%. studies related to various types of intent. Furthermore, the team at WisdomTap per- formed experiments where they targeted search Online Marketing advertisements at customers who had revealed an One of the most valuable intentions to detect is the intent to purchase a product. intent to purchase. Purchase Intention has appli- With search advertisements, they recorded cations in lead generation, social-media advertise- click-through rates of around 1% which again rep- ment, and social customer relationship manage- resents a 100 times or 10, 000% improvement.

3 Ment. These numbers are consistent with experiments A study performed by our partners and clients reported by IBM1 . They found that when the WisdomTap ( ) through 2011. Japanese fashion retailer Start Today used cus- and 2012 revealed that click-through rates on ad- tomized email messages tailored to each of their vertisements targeted at social media users who million customers, the conversion rate went up had revealed an Intention to purchase a product by nearly 1000%, and the annual sales on their Zo- was a consistent 7%. zotown website increased by Typical click through rates for online advertise- ments tend to be in the range of So, the 1. improvement obtained through the use of purchase Customer Churn Models Intention to criticize. So, Intention can be used in conjunction with sentiment. One of the problems facing telecom service providers is high rates of customer churn (report- Brand Metrics edly in some Cases 30 35% year on year).

4 Three intentions that might be useful in detect- The only metric that could be computed using sen- ing churn are: complain intent (reporting a prob- timent was customer satisfaction (CSAT). lem), accuse intent (blaming the firm), quit intent The metric was typically computed as follows: (leaving the firm). The detection of these three intentions in cus- positive sentiment CSAT =. tomer communications might help a firm reduce positive sentiment + negative sentiment its customer churn rate. With a richer set of intentions, there are many Product Launches more metrics that can be computed. Some are: With modern text analytics tools, it is possible to quit Intention analyze product launches. CLOY (loyalty) = 1 . complain Intention We tracked the Google+ launch as a competitor to Facebook between 30th June 2011 and 10th July When customer loyalty (CLOY) was measured, 2011.

5 It was found that the CLOY for Starbucks was There was a lot of speculation regarding whereas for laptops, it was whether it could kill Facebook. So, we turned to This seems to match what we would intuitively text analytics to see if it would tell us one way or expect. The ease of substitution of starbucks with the other. a competitor is higher than the ease of substituting The number of tweets expressing intent to quit a high-ticket item like a laptop. rose rapidly from 30th June to 4th July. On the 4th Another metric that can be measured is the of July, Google+ became open to all users (anyone amount of thought that goes into purchasing an could join it henceforth without invitation). item. The next day, the number of tweets expressing an Intention to quit dropped sharply. Once users compare Intention CHES(hesitation) =.

6 Had been able to try out Google+, they had realized purchase Intention that they couldn't bring all their friends along from It would be intuitively expected that high-ticket Facebook to Google+ and that Google+ wasn't go- items like laptops would have a higher CHES. ing to be fun without them. than low-ticket items like coffee. So, we could tell that Google+ wasn't going to In a study on Starbucks and on laptops, it was be a Facebook killer like many tweets before July found that CHES for Starbucks was while 4th had suggested, or atleast not yet. And we could for laptops, it was as we expected. tell that in just 4 days. Still another metric that would be interesting to Testing the Waters compute is market commoditization. This can be computed as follows: When Netflix tweaked its product pricing in 2011, since the new pricing was perceived as unfair by brand name mentions their customers, a large number of customers quit commoditization =.

7 Generic mentions within days. Had Netflix been able to track the quit intentions in the conversations about its offerings, All the metrics proposed in this section are mea- and been aware of the alternatives available to its surement proxies. They need to be tested and val- customers, it might have reacted way faster. idated by correlation with real-world figures. Sentiment Analysis Acknowledgements One of the intentions that can be recognized is the Many thanks to Vijay Ramachandran (vi- intent to opine. The Intention to opine can further CTO of , be subdivided into the Intention to praise and the for sharing the results of his studies on buy intent.)