Brand Decisions are Grand Decisions.
So, I often ask folks in the text mining industry, particularly the linguists and applied mathematicians that specialize in language detection, “have you ever read 100,000 customer service records, complaints, etc.?”. I have.
Of course not. These folks, who are toiling in their applied sciences, must attack the challenges of computational linguistics with an expansive and intentionally non-specialized approach to text harvesting. Likewise, the well developed BI community has a certain orientation towards data visualization and raw statistics. Neither of these camps has the requisite immersion in customer relations and the language of ‘consumer perceptions’ that occurs after a product is released into the hands of the public.
Don’t get me started on the ersatz ‘brand speak’ of the current early market entrants specializing in brand monitoring via sentiment analysis – I have made my opinions known, and those with far more august credentials have slapped me down good’o. This is all as it should be in the open media.
The brand intermediaries (multi-line dealers and retailers) that are the victims of branding decisions and the blunders of the brand owners are the fat part of a potentially untapped market being completely ignored by the brand monitoring leaders, both old school, as well as new age text harvesters. This is predictable, as the technology adoption curve has shown this myopia will occur across numerous innovation categories – the telephone (never envisioned to become a residential device), the computer (“might be useful for up to five of our largest corporations”, Thomas J. Watson., the PC, (what do people want a computer in their home for?).
Whereas the old school brand equity and monitoring agencies have made a conscious and informed choice for the moment, not to take the mid-market into their plans for brand monitoring, the new entrants are just skating by the opportunity due to arrogance, hubris, and let’s say, a lack of vision. Plus, it’s harder to serve the mid-market with an affordable decision support service, than it is to deliver the current sentiment scoring clap-trap.
Enough already, my position is known! What am I going to do about it as an independent analyst? Why, Dear Watson, I’m going to analyze, and then find a friendly technology leader to hire me as a consultant to implement these systems. Let us now break down the dynamics of Customer Perceptions and Interaction Outcomes in a semi-formal manner that is somewhat long for the blog format, however, I will condense the specifics to a few pages, and add a link at the bottom to a full white paper. Read On Babes:
Customers have a priori perceptions of brand that are more crucial to intermediaries than to brand owners, but these perceptions are of import to the entire commerce value chain.
Brand owners have to make decisions that impact their semi-durable destiny, whereas the retailers and distributors of high end durable goods are recipients of the fruits of these decisions, but in a less permanent way. The decisions to imbue a brand, a product, with a message is a decision impacting billions or at least hundreds of millions in the long term for the brand owner; brand intermediaries, spanning a much wider size distribution, are impacted severely in the mid-term economic horizon.
Product Perceptions (caps intentional) are an intangible capital asset. This is probably one of the reasons we are witness to such a fascinating bifurcation in the consumer electronics business of the later 20th century and the contemporary period. One the one hand, Sony can stumble multiple times and badly at that, while remaining a viable brand in the consumer’s collective perceptions, while one of the many new age electronics purveyors, such as Creative, Apple, or Dell, can make a bad product call and, theoretically, capsize a company in one season. Ah, you disagree? I posit that there was rampant speculation within Apple that, should the iPhone fail in any number of spectacular or not so spectacular ways, it could, nay would, have serious long term repercussions for the viability of the brand (and future efforts at competing in the mobile space).Yet it does seem that the likes of Nokia, Motorola, Sony, Ericsson, etc., have and will continue to make mobile devices with horrible usability and severely compromised product perceptions, that go on their merry way with a whistle and nod.
That is to say that Nokia and Motorola go on with a whistle and a nod; their poor distributors and the multi-line retailers take the brunt of the horrible decisions these brand owners perpetrate. And the persistent impression of these accumulated blunders, despite the ostensible success of the brand, create these a priori product perceptions that affect future buying behavior. Thus the need to steer brand monitoring away from brand owners and toward brand intermediaries. It’s gonna happen anyway.
Brand Decisions are Grand Decisions. Tools are needed to collect the resulting impact of these results on the perceptions of the consumers. This will not be done with the the horribly flawed sentiment metrics currently in vogue.
Interaction Outcomes are the total impressions formed by a consumer’s intercourse with the pre-sales, post sales, customer service, and warranty / technical services touch point continuum. If you are breathing, I don’t have to explain to you that a perfectly good brand or product can be mortally wounded by bad interaction outcomes.
It useful to, unfortunately, again make an example of the mobile telecom services monopolies, as they continue to amaze me, that as a group, they somehow manage to provide uniformly bad customer service, thus providing uniformly awful interaction outcomes. The upshot of this truly fascinating study of incompetence by an entire industry was essential to my research, for now I had access to perceptions and impressions of interactions outcomes shaded by incremental negatives. Can you beat that? An entirely bad industry – monopolistic, disparaging, punitive (cancellation fees rather than competing on merits), anti-competitive (refuse to allow VOIP transit). Ye Gads….
Where else can you get such purely wonderful, negative perceptions of crappy mobile devices that are poorly supported, years behind EU and Asian markets in terms of features and capabilities, and yet still maintain a semi-permanent, angry, and disillusioned clientèle? Ladies and gentlemen, I ask you?
Therefore, Hats off to AT&T, Verizon, Sprint, whoever – for the Golem continues to live.
Now then, where was I: We have here in the wireless trade a uniformly negative community impression from which to gather interaction outcomes, specifically. The product perception picture arising from the mobile device industry is somewhat tainted, as the pricing for most mobile devices is skewed by a witches brew of specialized incentives. The public does, however, retains a remarkable gullibility and immaturity over breakout products, as the iPhone has demonstrated a uniquely American trait, that childlike quality that seems to say, ‘if i can only have that doll, car, toy, all will be right by me’. The model of ‘living in a dream ‘, is at the bedrock layer of the science of product perceptions and interaction outcomes.
Science….! Please, the ego here.
The detection of these perception and outcomes comes from recognizing families of phrases that are commonly found in customer service complaints that are gathered from warranty reports and inbound contact call center and support logs. It’s a lot of reading and doing so may damage one’s ability to read for pleasure for many months after. We shall extend this phrasal mining and detection to the public corpus, as is being done now with sentiment analysis.
I have found the following linguistic phrasal patterns that lend themselves to symbolic substitution (more on creating symbol tables for sentence dynamics detection in a later article):
1) Generic Outcomes, free from declaratory clauses
“It didn’t work out”
“”She was not able to help me”
“It was still non-functional”
We see many generic outcomes in call center data mining in the consumer and professional electronics industry. Outcomes of a generic nature can be good or bad (from the consumer’s perspective) or expected / unexpected (from the servicing dealer’s perspective).
Declarations are non-generic statements about an interaction within the customer service continuum. They are often about what someone observed or has experienced:
“My wife felt belittled by the staff”
“I saw the technician hammering on the alternator”
“I arrived at the appointed time, but they would not take the car”
“I saw my car, it was a mess”
“When I took the computer home, it had the same problem”
Declarations are, therefore, non generic statements of observations and experiences.
Desires are the pure statements about what was hoped for, not what is sought, which would be redress.
“I wanted it to be ready by x time”
“I had hoped that installation would have gone better”
“I really wanted a solution to this ongoing problem”.
Desires are one class of expressions we find that often border on redress without the specificity and stridency of “outcome seeking language”.
4) Redress is the “big daddy” when it comes to targeting change within a services continuum, but is just really another non-generic declaratory phrase. When scoring, calculating stochastic values to the dataset, or extracting reports, Redress detection can be one of the routing parameters for action. For our purposes of mining multiple corpora for such perceptions and interaction phraseology, redress gets no special attention, other than it’s calculated values over time, and its integration with other phrasal tokens for the purpose of assigning value to the overall interactions.
“I want my money back”
“I demand satisfaction”
“We demand that this be addressed”
One of the problems with detecting redress is the labyrinthine way it can sometimes be expressed by those who are trying to be polite. Still, the engines made to cut and detect these phrases will get better, as will the cottage industry of modular ontologies for phrasal dynamics. Once we see an earlier dynamic, it is fairly easy to mark the tree for token inclusions that will, ultimately, be statistically sound enough for providing the kind of output we need.
Pronouncements are composed of decisions and advisories. They are the sign-off statements of a disgusted customer; we have all been there. As you can see by now, all of these phrasal markers express certain classes of perceptions and the results of interactions.
“I will never go there again”
“take m advice, ….”
“I thought x, now I believe Y”
“as a result of x, I am now Y”
A Good Start
There is actually very little new in this type of text classification, except perhaps the notion of re-targeting the marketing and product definition away from brand owners and sentiment, and towards brand intermediaries and the significantly more complex task of creating a hosted decision support tools for these mid-market companies. Such a tool will use the above mentioned classification to score the corpus on a brand and product basis, taking into account the program incentives that are always being offered to these multi-line dealers.
The resulting integration will aid these retailers and distributors by providing visualizations of the stochastic values of the scored corpus, and relating program incentive offers to this scored value.