10:57 — Jim: The human side of analytics depends on understanding the technical side.
10:58 — The mechanics are the stuff you need to do: collecting, cleaning, integrating, etc; the relational side will make sense of your data.
10:59 — Jim: We went wrong starting with the question, “What/How do I measure?” We start with influential topics or buzz or facts (and this gives us graphs, charts, and data), but those things cannot tell the difference between emphasis, sarcasm, idioms, age, etc.
11:01 — Jim: We start with measurement and we need to have governance. Who owns what bits of the data stream? We must own them to make sure they are trustworthy and have confidence in our data.
11:02 — Jim: Next comes data integration: we extract, transform, load, etc. This is measurement, and it's really tough with new data!
11:05 — Jim: The human side of social media data is analysis, which you conduct against goals. Goals need to be clearly defined (everyone knows what the terms mean), specific (amount and time), and supported (politically aligned with greater org).
11:06 — All of the goals need to relate to make more, spend less, and make people happy! These are the “Big Three Business Goals.”
11:07 — Jim: You must analyze if you're doing a good job in engagement. Measure what the brand people are doing and help them understand what the whole community is al about.
11:09 — Jim: In order to analyze well, here's what to do: First, look for the things that go bump in the night — what are the outliers? What are the giant spikes? Ask the question “Why?” Dig deep and make a discovery as to why!
11:10 — Jim: Second, segmentation. Everybody is different (averages are your biggest enemy; they hide the truth). Who are the leading edge? How do you lead people through? How are the acting differently and how can we treat them differently? Reason your way backwards.
11:11 — When doing segmentation, do not only use statistics. Be leery of cognitive bias because your brain will work against you! Also, correlation is not causation (e.g. when ice cream sales go up, drownings go up).
11:12 — Third, there is the display side. We want to display everything because we're able to take in so much visually. Periodic Table of Visualization Methods is a good tool to use.
11:13 — Jim: Data are individual bits of information. Each one is fascinating in itself, but we're asked to analyze an entire desert!
11:14 — Remember: “All models are wrong; some models are useful.” – George Box
11:15 — Jim: What makes a great analyst? A great analyst doesn't just collect data, a great analyst figures out what the numbers mean.
Q & A
Q: We start with the end in mind, but do you think that sometimes it's a trap because we are trying to prove that with which we started?
A: Jim: Figure out who you're trying to reach, what you're trying to accomplish, how it worked in other projects, etc. Try to figure out, too, what business objective you hope to achieve through the measurement. Ask this of the people who want to know what the data means.
Q: How do you tackle analytics in a Google anaytics obsessed company?
A: Jim: Two possible answers: 1) Education Google measures not just a subset of data but also measures it in a certain way; and 2) Find a different company?
Q: How do we sell this idea up?
A: Jim: By example. Find a small win and promote it! Tell them where the conversation went with a story.
Q: How can we determine brand perception through social?
A: Jim: There are bad ways to do it, and then it gets worse! First, share of voice. But first, you need sentiment in order to make that valuable. Sentiment is where we say, “We don't know the specific answer but I can tell you what to focus upon.” You must look directionally and when you find an alert, you go and look using actual human beings.
Q: How do we find the right talent? Someone with data capabilities but analytical thinking?
A: Jim: The data capability comes with experience (have you dealt with these tools, etc.?); and the innate skill for analytics is curiosity. You need an interview path that measures for curiosity. Give them a data set and ask them to derive something from it. They need to see the data and ask you 17 more questions about it!
Q: With paid media taking the lead, segmentation is easier, but for unpaid what are you talking about?
A: Jim: Sentiment divided by audience. It doesn't matter how you segment, really, but you're looking for how groups are different. You learn how to approach them uniquely with the goal being segmentation of 1:1.