Product Profiler

Analyzing product attributes to gain customer intelligence
Apparel retailers, for example, routinely collect customer transaction data that shows that a customer purchased a specific item in a specific size and at a specific price. But, this transaction data does not convey other key information, such as what a product choice says about a customer's personal preferences, style and fashion sense. If a customer purchased a shirt, can you tell if it was conservative or flashy? This type of "soft" product attribute is not typically captured in transaction data. However, these valuable attributes can often be found in the descriptions that accompany products in catalogs and online, which are designed to convey a particular image to customers using everyday language.

The Accenture Product Profiler, a prototype developed by Accenture Technology Labs, shows how these descriptions can help retailers better understand not only a customer's preference for a particular item but, more broadly, their taste. Product Profiler demonstrates how these attributes can be used in a wide array of marketing applications, from product recommendation systems to brand management and competitive comparisons; from product assortment analysis to a copywriter's assistant.

How It Works
The Product Profiler prototype relies on an advanced text analysis and modeling system, proprietary to Accenture. To build this system, Accenture researchers collected a large sample of product images and descriptions. Industry experts were asked to rate the products across a spectrum of attributes such as style and formality. Accenture researchers then applied a variety of leading-edge text mining techniques to the descriptions and the associated ratings to create a learned model that can be used to evaluate new product descriptions. For every word that appears in a product description, this model assigns a probability of association with each particular attribute. For example, the word "fresh" has a high probability of suggesting a stylish product but a low probability of describing a classic product. The prototype systematically combines these probabilities and classifies each product based on its unique set of stylistic attributes.

The Implications
Product attributes extracted by the Accenture Product Profiler can power a variety of marketing and business intelligence activities within the retail and consumer goods industries.

Accenture researchers currently are working on a Product Profiler recommendation system that analyzes, in real-time, descriptions of products that a customer browses or buys. The system then searches for and recommends items with similar stylistic attributes. Currently, most retailers are limited to traditional collaborative filtering techniques, which can only suggest alternative items based on what other shoppers have bought in the past. In contrast, the Labs' recommendation system can help retailers continuously categorize product lines, adding or subtracting items as the seasons and fashions change. Using Product Profiler over time to understand a shopper's taste and style, a retailer also can recommend products in completely different departments, such as housewares or furniture, when a customer returns with a different shopping agenda. Product Profiler reveals new customer insights that could be used to improve personalization, customer satisfaction and overall sales.

Another application of the Accenture Product Profiler technology is a copywriter's tool that takes the guesswork out of executing a brand strategy. Copywriters can "test" potential product or Web copy to determine if the wording conveys the intended positioning for a product, service or brand. Product Profiler can also suggest words that would better communicate a message to the target audience. This kind of approach could save companies in a number of industries substantial time and effort by reducing the write/review cycle, while helping to market products and services more effectively. This tool would be especially useful for apparel and consumer product companies and retailers who generate product descriptions frequently for catalogs and websites. In cases where multiple parties are executing brand strategy, such a tool would be useful to measure and ensure consistency.

Finally, Product Profiler can be used to analyze a retailer's entire collection, evaluate product assortment against intended brand positioning and compare their positioning against those of competitors. Accenture is currently exploring the possibility of integrating Product Profiler's capabilities into Retail software solutions such as Retek.