RSR Research: The Power of Micro Sentiment
Through a special arrangement, what follows is an excerpt of an article from Retail Paradox, RSR Research’s weekly analysis on emerging issues facing retailers, presented here for discussion.
When it comes to using sentiment as a new demand signal, retailers are learning that they need to analyze what people are saying about them in order to get early indications whether their strategies and campaigns are working or not. It’s possible for retail analysts to get good clues as to whether or not they need to adjust something they are doing in the marketplace if they get early enough sentiment signals from consumers, for example, via Facebook or Twitter.
Capturing sentiment is part of the bigger challenge to be able to track consumers’ "omni-channel" paths-to-purchase. "Click ‘n brick" retailers need to use what marketing people call multi-channel attribution.
A study, Consumer Complaining: Attributions and Identities, from Cathy Goodwin, Georgia State University and Susan Spiggle, University of Connecticut, describes it this way: "Attribution theory can be applied generally to ‘processes whereby people attribute characteristics, intentions, feelings and traits to the objects in their social world’. … [There are] two types of attributions — dispositional and intentional. People make intentional attributions directly from observed behavior; observers begin to make sense of their world by speculating about why others perform certain actions… intentions offer cues to dispositions, the more stable, underlying components of the actor’s personality which persist across a variety of situations. These dispositional attributions often take the form of ascribing to the observed individual a set of ‘broad’ traits, ‘despite the scant empirical evidence for their existence.’"
Marketing analysts use attribution data to try and understand the offline impact driven by online marketing and advertising. (From their point of view, that may be the number one reason for retailers to develop a mobile app for consumers, since it generates the metrics needed to help measure the path to purchase.)
I spoke to a group of retail financial executives recently in Dallas, and pressed the point that understanding these non-transactional metrics is critical to a successful omni-channel strategy, because all paths to purchase are not equal. There is no "one way" to enable a buy anywhere/get anywhere model — consumers use different retailers differently. (An obvious example is the difference between how a consumer uses a Macy’s vs. a TJX store.) In order for CFOs to support the capital spend needed to modernize the company for buy anywhere/get anywhere retailing, it must first be clear what paths to purchase the consumer is most likely to want optimized. And a good way to know that is by attribution analysis.
Discussion Questions
What are the main hurdles to employing non-transactional metrics to drive omni-channel strategies? What value do you give attribution analysis, particularly in understanding the online-to-store path to purchase?
The main hurdle is that the farther away you get from direct purchase data — i.e., what was bought, where, what time, by whom, etc. — the harder it is for some people to see the relevance of the data taxonomy.
As to the value of attribution analysis, I suppose it depends on what form that analysis takes. For example saying things like , “These dispositional attributions often take the form of ascribing to the observed individual a set of ‘broad’ traits, ‘despite the scant empirical evidence for their existence,’” might make a cynical person ask why one should believe in anything for which there is only slight empirical evidence.
Children who put their recently shed baby teeth under their pillows at night often wake up to find money under their pillow but that, in and of itself, is not a sufficient argument for belief in the Tooth Fairy.
The search for the intangible attributes of a purchase decision has created the most interesting marketing reports and studies I have ever read. E-commerce companies now use this information to determine the colors, type font and shapes incorporated throughout their sites. There is evidence that it is also used to create animated characters, sounds and voices for use in commercial interaction with consumers. Not so long ago this type of information was often dismissed simply because there was no way to take advantage of it. The power of 21st century IT Systems has changed all of that and will continue to use this information in more refined ways as the capabilities and scope of IT grow.
I believe that one of the biggest hurdles to engaging non-transactional metrics in driving multichannel strategies is determining how representative those social media feedback points are of your core customer base.
Social media feedback is important and valuable in understanding issues with product performance, customer service, experience and process issues. However, when this feedback is applied to marketing strategies, there is a danger that the feedback may only be representative of a very small, very vocal segment of customers. Paying too much attention to this segment may disrupt marketing communications and potentially take those communications off target for the largest and most important segments of customers.
The goal is to create the optimum customer experience. Any data from virtually every source can be important. The potential problem comes when we have so much (too much) data and get irrelevant information that takes us off track. Data is good if it is usable.
Right on, Ryan! Unless the attribution analysis is linking ACTUAL BEHAVIOR with micro-sentiments at a granular level, it is not very useful.
The problem here is that managers will use micro-sentiments to create a composite picture of “the customer” and will make broad, sweeping generalizations about “the customer’s needs/wants/frustrations. But in reality, there is a wide range of customers with very different behaviors and very different expressed feelings.
Until it’s possible to make that linkage in a reliable manner, micro-sentiment data will be mostly noise. By itself, it can’t compare to the usefulness of good old behavioral data.
Depending upon the program, “reading sentiment” will be more or less accurate. For instance, sarcasm is not easy to detect from just words. Depending upon whether purchase behavior is ascribed based upon positive comments or recommendations will also make a difference when projecting purchase. This type of analysis needs to be grounded and checked and interpreted. Correlation between sentiment and purchase behavior is not a causal relationship.
Attribution is an art/science but is an incredibly powerful mechanism. The technology and tools are there – people are biggest the hurdle, you need executive buy in and a team that can absorb and take action on the results. The retailers that are investing in this area have seen incredible improvements in marketing campaigns, assortment, site design, seo,and many other areas. These investments have paid off so well it’s hard to believe that we’re in a ‘big data’ gold rush isn’t even bigger.
The hurdle was technology. More and more organizations have developed true sentiment analysis tools that are proving to capture actionable insights. These tools take the words from various available social and other channels and take all potential meaning/intentions of these words, that is the true sentiment of the written words.
The potential value has been realized by real retailers and CPGs in that they have modified their marketing strategies to the extent that the consumer sentiment has indicated. There is an example out there where the CPG didn’t even realize how popular their SKU is around the world, although 98% of it is sold in only one country.
There is great danger here in relying on these metrics if they do not actually link to real world sales. Of course the beauty of online is that “everything is measurable”. But it is also a potential curse — it is easy to measure things that don’t matter.
The way around this is to prove the relationship between “intermediate metrics” and actual real-world sales. The only way to do that is by controlled testing. The best brick-and-mortar retailers design and analyze tests of online activities that are meant to drive offline sales, so they can both prove out the value of online activities and track intermediate online-intention metrics to understand the true relationship with real-world sales.
Words are harder to analyze than numbers – and both can be poor representations of reality.
Non-transactional metrics are critical to encouraging shopper marketing best practices. The days when a single marketing message or customer experience drove most purchasing decisions are long gone. So multi-touch attribution is the only way to successfully understand how your marketing is influencing your shoppers. Many of those touches won’t be transactional, but if you only pay attention to transactional metrics, you’ll be focused on the wrong things.
For example, retail e-commerce sites are heavily used as pre-shopping tools for retail stores. It’s not uncommon for the e-com site to drive $10 – $60 in retail store sales for every $1 it captures online. But if the e-commerce VP is only using transactional metrics, then they will never be focused on those pre-shopping experiences.
Best in class retailers have long included non-transaction metrics like CSAT, Store Traffic, etc… in their KPI suites. This is a practice that will only become more common.
The basic hurdle I see is that what people say (i.e. on Twitter or Facebook or on a survey) is often different than what they do. Observable behavior be it ethnographic observation or what we can see in the transactional record is the best predictor of future behavior.
That said, when working with an online gift retailer we developed a unique way to help shed light on purchase motivations. At that point from the data we knew what customers purchased and when and could draw some conclusions about motivations, i.e. red roses purchased on February 12th were most likely related to a Valentine’s purchase, or a fall harvest arrangement purchased during the 3rd week of November was likely purchased to be a centerpiece on the Thanksgiving table.
But there were many other purchases made that were not close to holidays that we needed to understand, so we developed an algorithm to crawl through all the gift notes to look for keywords that helped us classify other kinds of purchases (sympathy, anniversaries etc.). Through this work we were able to identify important customer segments who made purchases “just because” — it was their way of showing support to a friend and develop marketing strategies that more closely resonated with purchase intentions.