social monitoring and network analysis

The Age of Predictive Social Media Monitoring

In the next iteration of social media monitoring solutions, conversation data from discrete topic groups will be run through a “cycles scanner”, which shows dominant cycles related to specific topics. Insights professionals will be able to accurately predict future behavior based upon recurring patterns discovered using the “cycles scanner”. The way the cycles scanner works is simple: it takes data and uses a variety of well known algorithms, along with spectral analysis, to determine recurring patterns. Insights professionals will see before their eyes on a graph the ebb and flow of conversation relative to periods in the past, as well as a predictive signal (in the form of a wave pattern) going out into the future. This is important because insights professionals will have a higher threshold of confidence when advising brands on when to execute on specific manufacturing volumes, distribution foci, and campaign content. Insights professionals will also be able to blend audience intelligence with this enhanced predictive ability to reveal to brands who will buy next and where/when. We are entering an age of true prediction. 

Social Listening 2019

SUMMARY OF SOCIAL LISTENING – 2019:

Social listening gives us clues related to our customer, our competitor and our marketplace based on conversation analysis at scale. These clues lead to insights that inform laser specific strategy for ALL silos in the enterprise. In addition, we can build a customer base from conversation analysis by enriching handles of those in the conversation. In addition, we can see who is influencing people talking about our chosen themes/topics.

Sentiment analysis and Emotions analysis gives us insight into what causes Joy, Anticipation, Fear, Disgust, Anger within conversations where our chosen themes and topics are being discussed. This is important because we can design better marketing campaigns, bring efficiency to our overall budget based on what’s working/not working in the marketplace, we can spot trends in the marketplace and anticipate where to focus our resources as a brand.

When we see WHO is talking about themes/topics important to our brand, WHERE these conversations are occurring, WHAT is driving awareness of our themes/topics, HOW customers are arriving to our channels AND to our competitors channels, then we make more intelligent decisions for our brand, for each silo in our enterprise.

Social listening is a process by which insights are derived from truly massive quantities of social data (online conversations, documents, and profiles). We distill these huge amounts of data into digestible insights for business stakeholders, accompanied by detailed datasets of prospects, and individuals/entities who influence these prospects. Software solutions combined with human analysts are our chief means for achieving this work

Our net deliverables are PDFs, spreadsheets and in-person meetings where we deliver insights & recommendations related to our research. These deliverables are important because brand leadership has a set of insights/action steps related to discovered individuals (our initial data lake of prospects). We also meet with stakeholders in various business units to discuss the findings, participate in action teams who are executing on initiatives supported by the insights, adjust process and rinse-repeat as needed, honing in on specific additional items desired. This refinement process is where we really drill into the “2nd concentric ring”* and find the exact targets worth acquiring. This is also where we find out what’s working and what’s not working for a specific unit/team.

*The “2nd concentric ring” is everyone who is following a specific influencer. Segmenting and defining the demographics/psychographics/personalities of every single person following an influencer gives us a better idea about whether this influencer is a good pick for our organization. We also find out a lot more about our ideal consumer when studying the “2nd concentric ring”. This is important because when we see every single person who has chosen to follow and engage with an influencer, we gain insight into the culture, buying choices, and online habits of our ideal consumer.

………..

THE NITTY GRITTY PROCESS 2019 (STEP BY STEP):

1. Define the questions/problem to be answered. This is done by sitting down with the client and interviewing leadership, then specific stakeholders, then those who work with specific stakeholders. It is best to do this in small groups or with individuals so as to get honest and truthful information. This is important because we want to start our study with a very deep set of insights on the organization and those who will be using our research. When I interview stakeholders and their teams on an individuals basis, I learn more than when I interview a group together. When I interview a group together, all of the politics and internal issues prevent individuals from sharing fully what is needed.

2. Define the scope/parameters – dates/topics/desired outcomes. After interviewing everyone, I have insight into how the organization will be most helped by the research work. Very few internal stakeholders and employees in the enterprise have the “big picture” view. Usually each person is interested in his/her own agenda or in pleasing specific senior stakeholders. When we know what is at the heart of the organization’s psyche, it’s heartbeat as it were, then we can deliver a scope of work that truly meets the brand’s need vs. individual stakeholders’ needs. This is important because we want our work to feed the brand, to nourish it’s life. A brand can breathe when fed useful insights, and it can die if it is shielded from useful insights. Truth about the marketplace, the consumer, the competitor, and, importantly, the inner body of stakeholders, employees, vendors, contractors and non-human drivers is vital.

3. Define the audience for the report(s) – who is this for/what is purpose of study/why are we doing this study. Each insights report we deliver will nourish a specific person, unit, division, region. When we know the true need, as stated in the last point, we can speak to the audience who has this need…we can speak to the heart of the organization itself. Knowing one’s audience affects one’s voice, one’s tone, one’s approach. This is important because we want our insights to be digestible, used, passed around. We want our research to truly affect change in the organization, change for the good of the brand.

4. Gather existing pre-study materials from the client and study these. Ask client questions about submitted materials. Gather more materials if needed/available.

5. Create boolean queries for conversation data aggregation in a conversation analysis tool.

6. Refine these boolean queries for more precise conversation data aggregation.

7. Download the raw mentions from the conversation analysis tool.

8. Download other relevant sheets from the conversation analysis tool, including Topics, Most Mentioned Authors, Most Used Hashtags, Leading Authors (in terms of Inf), Leading Sites (blog), Leading Sites (forum), Leading Twitter Authors, Leading Blog Authors, Leading Forum Authors, targeted Mentions downloads (using Rules and sub-queries within the search field in the Mentions tab.

9. Organize conversation data downloads, cleaning up columns, deleting un-needed columns/rows, filtering for site type and putting into separate sheets/tabs in Excel.

10. Enriching the Twitter handles/Instagram handles with additional info about those authors using audience intelligence solutions.

11. If needed, further enriching these titles with add’l social handles using APIs that give us PII (phone, email, address, etc.).

12. Analysis of conversation snippets for insights (junior analysts do this, filling in coding columns and insights columns).

13. Analysis of leading sites.

14. Analysis of leading authors.

15. Upload specific sets of conversation snippets into a tool using the LDA algorithm (Latent Dirichlet Algorithm) for Topic Modeling and Emotions Analysis.

16. Look over analysts’ hand coding work and develop macro insights based upon this work.

17. Look over emotions analysis and topic modeling for add’l insights.

18. Look over the types of people talking (from audience intelligence tools) and add add’l insights.

19. Conduct in-person focus groups, where needed/if required by client (using ideal candidates found from the conversation/audience data analysis).

20. Create final reports with insights, recommendations, appendix and, where needed, the working Excel sheets. Add charts and graphs to Appendix of report.

21. Deliver Insights report to client.

22. Deliver an Excel sheet of the enriched author handles, along with addl charts containing psychographic insights, influencer insights, related to these authors.

23. Deliver Excel sheets from audience intelligence dashboards (includes offline insights from sources like Acxiom and Experian).

24. Go over recommendations one by one with the client.

25. Ask client if there are add’l questions they have. Identify opportunities to collaborate on future research together.

Audience Intelligence Formula – Late 2017

SUMMARY: The formula for best-practice audience intelligence work is dynamic, due to rapidly advancing technologies and practices. We also have to take into consideration federal and nation-state regulations which are constantly in flux. As of December 2017, this is a formula for producing a fully enriched audience based on the beginning point of email, social handle, or a conversation snippet from social media, blogs, forums or comment threads on news sites. I might add that the acceleration of AI (Artificial Intelligence) and ML (Machine Learning) is rapidly condensing the steps below. APIs (Application Programming Interface) are being woven together to produce highly sophisticated machinery that blends outputs from various data sources.

 

THE FORMULA (LATE DECEMBER 2017):

STEP 1: Interview the client to determine goals and deliverables. This is important because we want to understand the client’s business and his specific goals related to the research. This is the time to see eye-to-eye with the client and really get to know his/her business from the inside-out.

 

STEP 2: Perform tests using the software below to discover opportunities related to the client’s desires/needs. Most social monitoring/aud intel solutions provide a way to quickly preview “the universe” related to a specific research request. This step is important as part of putting together the proposal for the client.

 

STEP 3: Design the audience intelligence study. Gain approval from the client, sign contracts, receive funding from the client. This is important because the client will want to look over the exact type of deliverable he/she will be receiving, as well as hearing a bit about your approach in putting together the final “data lake” of prospects and, of course, how the insights will be presented.

 

STEP 4: HOW DO PEOPLE DESCRIBE THEMSELVES IN SOCIAL BIOS: We gather a set of social handles where our keywords are in the bio. This from Audiense.com, Affinio, Brandwatch and PeoplePattern primarily, but can also be from using Data-Miner.io in LinkedIn, along with many other sources herein un-named. This is important because self-description by individuals is rather verbose these days..and this is helpful to us in finding our targets. And, where self-description is not verbose, there are clues through company/press/3rd-party descriptions of specific individuals who work for/play at/eat at/drink at/shop at/you-name-it at the locations/venues/places we are studying.

 

STEP 5: WHAT DO PEOPLE TALK ABOUT IN SOCIAL MEDIA/BLOGS/FORUMS: We gather a set of conversation snippets related to our keywords from Brandwatch, Sysomos, Crimson Hexagon and/or Meltwater. This is important because what someone says about an activity/product/service/location/you-name-it contributes towards our understanding of the consumer/customer/competitor’s consumer/customer. In addition, these conversation snippets from social media, forums, blogs, comment threads on news sites form a body of data that we can segment into specific topic groups. These specific topic groups can then be used to form a point of view/set of insights on the target we are studying.

 

STEP 6: SEGMENTING STEP BY MACHINE – TOPIC MODELING APPLIED TO BIOS & CONVERSATION SNIPPETS: We use Converseon’s Conversus tool to perform topic modeling and separate bios and conversation snippets into discreet topics. This is important because this speeds up the analysis of the bios and separates out the bios that matter to us the most. No solution in the world is more accurate and complete in segmenting bios and conversation snippets into discrete topic groups than Converseon’s Conversus. This solution is pure magic and the analysts at Converseon who are using Conversus are second-to-none in their expertise at building out superior insights based upon the use of their in-house solution. It is during this stage where an analyst begins to really gain deep insight into a sets of bios and sets of conversation snippets.

 

STEP 7: SEGMENTING STEP BY ANALYST:  Now, these analysts wade into the output from Converseon’s Conversus and identify the Topic where our targets are present (and any other discovered target – unknown unknowns). This vital step by humans helps us know which Topic groups in Conversus are populated by our target. This is important work that can be recursive, whereby the analyst segments the data using Conversus, reads through the results, and then segments again to refine even more deeply. I might add that this particular step is where the machine will eventually outstrip the analyst. That outstripping of the analyst will take some time to come along, though. For now, on this step, the human continues to be the last mile.

 

STEP 8: RE-STITCHING AFTER SEGMENTING STEPS: Reunify the bios from the “right” Topics in Conversus with the handles in the original source data sheet. This is important because we want to have the correct bio next the correct social handle. We also stitch handles and bios next to discovered conversation snippets at this stage. We find that working in Jupyter notebooks using Python is one of the handiest ways for our teams to work together efficiently on this step.

 

STEP 9: PEOLE PATTERN AUDIENCE INSIGHTS STEP: Upload Twitter AND/OR Instagram handles into PeoplePattern for deeper enrichment of Interests, Location, Age, Persona. This is important because we find out more about each person and we also delineate between Individuals and Organizations. In addition, we move a step closer through PeoplePattern to verifying the “real people”. Finally, we gain insight into the Persona types, Interest groups and lots of other useful info.

 

STEP 10: FULL CONTACT STEP: Use the Full Contact API to append add’l social handles. The value in this particular step is discovering a full name related to the handle AND additional social handles. We also gain bios from various social media handles, thus bulking out our story about an individual.

 

STEP 11: PIPL STEP: We use the PIPL API to gather Email, Phone, Address, other social handles, age, and many other bits of info on individuals. This is important because we will gain additional important information about the individuals that verifies they are “real people”.

 

STEP 12: CRYSTAL STEP: We use the Crystal Knows API to enrich the profiles with DISC personality type, personality overview, messaging guidance, selling guidance. This is important because then we are guiding our client on how best to market, advertise and sell to this individual. We can then group individuals by DISC type, if desired.

 

STEP 13: CLEAN UP ROWS FOR COMPLETION: Stitch together the results from the various APIs and then filter for complete rows. Again, we use Jupyter notebooks and Python for this work, as the teamwork and efficiency is vastly improved. This stage is important because we want every row to have every cell filled with correct & complete data.

 

STEP 14: EXACT DATA STEP: If we want to add an extra step for verification at this point, we run the Exact Data enrichment on the emails/names discovered. This is important because this extra step adds validity to the claim that our audience is full of “real people”.

 

STEP 15: SPOTRIGHT STEP: Upload the social handles into Spotright to gain enrichment of offline Axciom data, such as buying/purchasing styles, net worth, income level, political/religious affiliations, housing info, household complexion info, brand preferences of a specific group uploaded and much much more.

 

STEP 16: INSIGHTS, METHODOLOGIES & RECOMMENDATIONS PDF: Create 3-5 page Summary PDF with insights related to the research. This is important because brand leadership will now have a set of insights about the discovered individuals (our initial data lake of prospects) and, importantly, we can recommend further research steps for successive work together.

 

Who is the actual human we are targeting?

FINDING REAL PEOPLE:
We begin by identifying every single person who is online. We do this by proving which social handles are connected to real people and which are simply junk accounts. The way to prove if a social handle is connected to a real person is to append extra data next to a social handle, such as other social handles, emails, addresses, phone numbers, the social handles of family members, job titles/bios from social sites (LinkedIn, Facebook, Twitter, Instagram, Vimeo). This is important because we want to analyze real human beings and their behaviors. The reason we want to study real people is to make accurate predictions about human behavior and the outcomes of behavior in different contexts. When we can predict behavior more accurately, then we are able to influence individuals, which is one of the ultimate end games of marketing, advertising and pubic relations.

SEGMENTING PEOPLE BY PERSONA/INTERESTS:
When we have a healthy set of columns with this kind of information, then we can move on to using tools that study all of the content posted by an individual, as well as the way the individual describes himself/herself in bios on various sites. We can also study posts about individuals, some of which may include video footage, images, interviews, and sites that denote achievement. This set of software studies all of this content about an individual and classifies him/her as a specific persona type, along with the interests this individual focuses on. This is important because we want to know more about the type of person each individual is and what influences his/her behavior.

DISCOVERING BRAND PREFERENCES/PURCHASING-BUYING STYLES:
The next set of software we use derives insight from offline data, such as the data from credit bureaus, credit card companies, direct marketing companies, catalogue marketing companies, club membership research, background checks, etc. When we blend offline data with online data, we are able to demonstrate with more confidence the brand preferences, purchasing-buying styles, and many other classifiers related to an individual. This is important because online conversations, posts, and self-description in bios do not always give us enough to deeply understand the behavior of individuals. The blending of offline and online data results in a more complete portrait of the individual human being.

WHAT DO WE DO WITH ALL OF THIS INFORMATION:
In short, we are able to more precisely influence what an individual human being will do in the future. For a brand, this long-term influence is very important as this will be how loyalty and sales are ensured. There are, of course, deeper goals for other organizations, such as governments, religious groups and media groups. These groups are often interested in wholesale culture change, particularly in enemy/competitive territory. The action of changing another culture is a top long term priority of groups that have been around far longer than the Unilevers and P&Gs of the world.

A Formula for Better Insights on Consumers & Their Emotions

TheSocializers

THE PROBLEM & QUESTION: The Head of Insight at a major CPG brand is trying to figure out why brand equity is falling. A CEO is wondering why sales are dropping. A Senior Analyst is assigned by the Head of Insight with answering these questions.

THE FORMULA FOR ANSWERING THE QUESTION: The Senior Analyst decides to work with social data to discover the reason for falling brand equity. She comes up with the following formula:

A focus group of individual consumers (with a full dossier on each one) + these consumers’ emotions (at the individual level) about a brand or category topic (machine classifies millions of conversation snippets by emotion) + macro insights on each “Emotion Group”, based on offline purchase data AND all known interests of individuals in that group.

THE RESULT: When Heads of Insight/Brand Managers/C-Suite execs can zoom from macro insights about consumers INTO individual level dossiers making up an “Emotion Group” (a group of people expressing a specific emotion about a product/service), they will gain a clearer understanding of the intent and actions of these consumers. They will see when a consumer says, “I hate this product because…” or “I love this product because…”. They will see what characteristics entire groups of consumers who love or hate a specific product share.

THE IDEAL FINAL SCENARIO: Head of Insight to C-Suite execs, “I can see groups of consumers who are excited/disgusted with our product, the statement each consumer made about our product, and dossier-level insight into each individual making up these groups. I can then compare these groups with groups of consumers who are excited/disgusted with our competitor’s product in a nice neat single screen. Now I know why our brand equity is falling and I can make evidence-backed recommendations about our next move.”