We recently gave a talk on Active Listening in digital media at the Digital Intelligence Congress 2014 held in Barcelona. The talk consisted of two parts: firstly, we discussed the possible goals that an organisation could reach by monitoring conversations in digital spaces and, secondly, we defined the process to be followed in order to set up a project of this kind.
So, let's summarise in 10 steps our methodology for carrying out an active listening project:
1. Objectives: as with any other project, in monitoring we must know beforehand in which subject or area we want to act. It is not a question of uploading performance graphs, but of being able to make decisions, so we need to define what possible routes we can take.
2. Decide which are the key indicators that will guide us in making decisions: a monitoring tool normally offers us a whole multitude of data, but we have to evaluate which are the key metrics for us (mentions, sentiment, trends, key voices, emerging topics, etc.) according to our area of interest.
3. Categorise key topics: it is not the same to say that we have a certain number of mentions as it is to know exactly which topics are being talked about when we are mentioned, and what these mentions tell us. At Zinkdo we always propose to categorise these topics according to the action areas of our company. This way, we can make categories that have to do with the price perception, with the advertising we do, with a specific message, with the company's managers, etc. It should be noted that a mention can touch on more than one key issue, so it should be reflected in as many as appropriate.
4. Semantic dictionaries: we have seen many good monitoring tools poorly configured because they do not include in each of the key topics all the ways in which different audiences talk about that topic. Formal terms, colloquial terms, misspellings, etc. need to be included.
5. Configuring tools: once we have done all the previous work, we can start up the tool by loading the configuration that until now we have only done on paper. Each tool is a different world and can offer different results based on the same data. It is important to know very clearly what kind of output in terms of metrics, classifications (languages, territories, etc.) and reports we need before choosing our tool.
6. Measurements: Once the data is loaded, the tool starts measuring. Depending on the tool selected, we may start seeing data sorted according to our choice from time zero or we may have to give the tool some time before displaying initial data.
7. Eliminate noise: it is easy for noise to appear when monitoring is activated. From Zinkdo's point of view, noise in monitoring is everything that does not contribute to decision-making. It could be that some of the terms entered have other meanings, that the name of one of our brands has a different meaning in another language, or that there are mentions of our product that are of no interest to the brand because of their generic use. Imagine we monitor the Kleenex brand. Would we be interested in every mention of our brand to make decisions? Surely not. Supermercados Día is not interested in the huge volume of mentions that occur with the word "día", but surely they don't want to be left alone with the entire chain of "supermercado día". Tous Jewellery will not be interested in the majority of mentions in French. All brands register noise and we must configure the tool's filters to eliminate it, without being so restrictive that we miss mentions that could be of interest to us.
In addition to correctly configuring the filters, we have to help the tool quite a bit with the manual debugging of data. Especially with metrics that are more sensitive to one-off interpretations, such as sentiment. Not even the best tool is capable of reliably classifying sentiment in an acceptable percentage of the mentions detected. Language is complex, and in social spaces even more so. In social media, each sender has the complicity of an audience that knows him or her and knows how he or she expresses himself or herself. A simple emoticon, a hashtag, a punctuation can change the meaning of the message. Not to mention the many words that can go from being the worst of insults to praise for the recipient. The only way to evaluate sentiment in an acceptable way is by hand, after establishing a consensus of what is positive, negative or neutral for us.
8. Analyse: once we have the data sorted and cleaned, we need to zoom out and look at the trends. Then zoom back in and look closely at profiles, going into more detail to see who is receiving the message. At this point we zoom out again to look at the overall picture, and conduct more searches with the tool to further refine our data. And so on, until we can draw conclusions that allow us to see beyond the numbers.
9. Define actions: we cannot stop at just being commentators and analysts of the consumer's voice in digital spaces. We have to invent possible actions to mitigate its negative effects and enhance the positive ones. It is about offering value, not generating data. At least at Zinkdo we see it that way 🙂
10. Optimise the tool: Some companies see monitoring simply as a task of setting up a tool and compiling regular reports. But the reality of companies, brands and the current environment is constantly changing. That is why we must include new terms, actions, key topics, meanings and continue to refine the tool on an ongoing basis. Imagine that for a commercial we hire a well-known character, we will have to include them in everything that has to do with our brand and classify them well, and again a lot of noise will be generated that we are not interested in and we will have to refine it... Active listening is a continuous and exciting process if it is done with love and with a view to learn and act on the information.
Each of the steps we’ve discussed is explained with concrete examples in the study that Zinkdo carried out on the Digital Reputation of City Brands. You can download it for free at the following address: http://bit.ly/zdciudad