Most social listening in agencies was, and is, a dashboard and a monthly slide of charts. This was built to be something else: a research program with a stated method, a stated sample and a stated margin of error, that a brand team could run every month for a year and trust the trend. I led Influence to Commerce and the social and digital practice for the region, and the Singapore lead role for this account sat with me. The program was approved and it ran.
Three questions, three methods
The brand wanted to understand the digital consumer in Japan, and the brief was reduced to three questions. What content moves and impacts them? Who influences them? What do buyers think about our products and our competitors' products?
Each question got its own method. Social listening answered the first. Influencer analysis and tracking answered the second. Analysis of consumer reviews on e-commerce sites, specifically Rakuten and Amazon, answered the third, for Pampers and for the competition. Mapping one method to one question sounds obvious. It is rarely done, because it forces the agency to say which tool answers which question and to be judged on that.
The deliverables, at three speeds
The program ran at three cadences. A deep dive into one year of conversations, the existing influencer ecosystem in Japan and a year of reviews, delivered once. A monthly report with the month's key insights, the updated influencer ecosystem and the month's reviews. And daily flashes during campaigns, with ad-hoc crisis monitoring reported hourly when a crisis was live.
The framework, with the durations attached
The method was the Social Insights Framework, in six phases, each with an estimate of how long it took: map the strategic output, two weeks; build the universe, one week; scope the extraction, one week; develop the segments, three weeks; structure the analysis, two weeks; define the report, one week. Ten weeks from proposal to first report.
The part I still consider the most important line in the deck is the sampling standard. The monthly analysis was defined as a maximum sample of mentions, cleansed and manually coded, sized to give 95 percent accuracy with a confidence interval of 3.5 percent for the estimated universe of conversation that month. The coding attributes, covering brand and product attributes, the customer journey and sentiment, were to be co-created with the client and agreed before any data was pulled. Quantitative analysis found which attributes were significant by volume, cross-tabulation found which attributes consumers liked and disliked, and qualitative analysis of the significant areas found the themes. A theme that recurred across attributes or brands was called a trend, and only then.
Choosing the tools for the market rather than the network
The listening tool was a Japanese one, chosen over the global platform the network used elsewhere for one reason. The two places where Japanese consumer conversation actually happened, the 2channel forum and Twitter, were sold to the global platform by the Japanese vendor, and the global platform did not carry all of the data. The local tool had complete access to both. A listening program that misses the country's largest forum is not listening to the country.
Reviews came from a platform that structured ratings and questions on the retail sites. Influencers were identified by a tool that used social graph analysis rather than follower counts, to find the people who connected communities rather than the people with the largest audiences.
The example outputs showed what the instrument would produce: mentions by brand against Merries, Moony and the category; mentions by platform and by media type; the gender, age, region and occupation of the people mentioning each brand; sentiment by brand; the other words that appeared alongside each brand; the accounts behind the mentions; and mentions plotted against television and news coverage over a month, so the brand could see which broadcasts moved the conversation.
What it took to start
The deck ended on two approvals: the cost, and the keyword list. Both were named as the decisions that had to be made before the program could start, and both took longer to settle than the ten weeks the framework allowed for everything else.
What I would do differently
I would have written the first month's report before the proposal, on public data, at a smaller sample, and shown it. A framework with durations and a confidence interval is persuasive to a research buyer. A page of findings about Japanese mothers that the brand team did not already know is persuasive to everyone else in the room, and it would have turned the keyword list from an approval into a refinement.
The instrument is on this page because the instrument is mine to describe. What it found about Japanese mothers belongs to the brand. The method is describable. The specifics are not mine to publish.