As an ever-greater volume of B2C marketing and interaction moves online, the business value of understanding its impact, positive and negative, is growing rapidly. There is no shortage of software and services that promise the ability to monitor and manage these online voices, yet well-established metrics like ROI and even the testability of analyses have, in many cases, been an afterthought. Marketing and customer insights executives are becoming more sophisticated in their adoption and use of these tools, and the industry must take note.
Separating signal from noise and the difference between passive and active engagement are hard to achieve in social media
It has long been my view that the opportunity presented by online, particularly social media engagement, was easily balanced (if not outweighed) by the challenges in making sense of it. The person who steps into a retail store is highly unlikely to behave in the same way online, especially if that person has had a strong positive or negative experience. This is not the only problem, which – to varying extents – can be classified under the well-known issue of separating the signal (useful) from the noise (perhaps informative, but lacking in specificity and therefore reducing the scope for action). Noise comes in many forms; someone talking about “apple” may be discussing the latest Apple smart device or a piece of fruit.
I also think it’s fair to distinguish between passive listening and active listening. By this, I suggest general sentiment is, again, not without uses but lacks the active element of emotion. For example, "this digital marketing campaign delivered a generally positive sentiment" versus "this campaign generated feelings of anticipation and excitement" – a significant difference.
In both of these examples, the difference can be summed up by the opportunity presented to take an action. Where signals are received, a clear link to particular events can be established and therefore actions taken or lessons learned for future activities. The same is true of understanding emotion: Again, sentiment may tell us that the reception was broadly positive; emotion tells us what type of positive, allowing much clearer learnings to be had. The benefit in both cases is to increase the ROI and future ROI of spending online – in my opinion an area that has, until recently, lacked the scrutiny it deserves in terms of delivering value on money spent.
There are technology companies looking to solve this problem, but it is far from a distinct market yet. Three broad approaches are emerging when it comes to this field, in the majority of cases centered on the developing field of machine learning: facial recognition, interesting for video and pictures, but irrelevant to text-based social media posts; voice recording analysis, most often focused in the contact center; and text-based analysis, of primary interest in this area for the time being. DigitalMR, the subject of a recent Ovum On the Radar report, has recently introduced these features as part of its core product set. This addition is representative of Ovum’s view that DigitalMR intends to provide marketing professionals with a higher level of online data accuracy than traditional social media tools.
There is, of course, a “but.” Refining noise to signal and detecting emotion are not easy things to do; in fact, many will tell tales of false positives and generally low accuracy (especially when it comes to emotion). The technology is, of course, constantly evolving, and Ovum has witnessed significant strides forward in the solutions that deliver these capabilities. In the past, I have strongly advised that this sort of capability is best delivered as-a-service, not least because mastering one popular social media site takes time and effort, and its popularity may not endure. Far better to allow a vendor that has made this its business to worry about what’s hot (or not) online. A final piece of advice when it comes to the points I’ve made about accuracy, and as with any service provision, demand to see recent examples – then talk with those references, bearing in mind that like any form of data quality, 100% may not be achievable or desirable: Did it achieve what the client wanted to do?
Tom M. Pringle, Head of Applications Research