Get predicted Engageability of messages with Intuition model

Updated 

Brands get thousands of messages daily on social media channels. Such high volumes of messages make it very difficult for these brands to manage their social conversations. It’s a nightmare for the community managers of such brands to go through such an enormous number of messages and based on certain brand guidelines find relevant messages on which they need to engage on. Due to this manual scanning of messages, it’s not possible to look at every message with perfect classification. Thus, a lot of messages which required a brand to engage in getting missed diluting the customer experience. 

To solve this problem, Sprinklr has developed an automation system which scans all incoming messages and classifies them into one of these two categories: 

  1. Engageable: Messages on which the brand should engage on. 

  2. Non-Engageable: Messages which does not require the brand to engage on. 

The Engagement Intuition model classifies on the basis of text content and is applicable across channels as well as on both owned and earned data. 

Guidelines of Engagement 

These guidelines define the overall approach to social engagement for brands which is followed by the Engagement Intuition model to predict engage ability of a message. 

  1. Based on message

  2. Based on the social profile

  3. Based on conversation history

Based on Message

Engagement Intuition model reviews the customer motivations and the content of the post to determine if it meets the criteria for responding as given below. 

Prediction Based on Customer Intention 

What to Engage

Message Type

Example(s)

Fans who are excited about the brand 

  • I am so excited about the new Phone?

  • When will this get launched in BB City? 

Fans who show love for the brand or its products or services 

  • oh God, I LOVE room service. And in any case, it can be the same sort of price as a hotel.

Expressing Interest or intent to buy a product/service (existing or future coming) 

  • I can’t wait for Air max launch date? 

  • looking to buy tab 4, can you help me locate a store? 

Individuals who need to be educated on brand policy, quality, or brand guidelines, etc. 

  • Can I smoke inside the room during my stay?

  • Is it safe to apply this lotion on my skin? 

Customers looking to learn more about initiatives put forth by the company or company information 

  • What vehicles will be affected in recall? 

  • How is this product more green now? 

Customers looking for help from customer service or support 

  • I’m getting this prompt on my TV that reads “are you still watching?”, and nothing I do will make it go away. Help? 

What Not to Engage
Message Type
Example(s)

Posts, shares, retweets by Trolls, haters, conspiracy theorists, or those looking to get attention 

  • All these companies are making fools of us. Burn in hell!

  • Seriously, what's up with the racist phone commercials? 

Messages meant to spam on social 

  • Earn free cash by subscribing here.

Prediction Based on Content of Post 

What to Engage
Message Type
Example(s)

Complaining or unhappy about a product/service 

  • I have not been receiving signals from past 2 days. please fix this.

Having queries on how to use/configure a product/service 

  • How should I connect to ABC live? 

  • How Should I update my office?

Praising a brand for its product/service experience or innovation/achievement of the brand 

  • I really love how MNO is easing the use of Technology in their products.

  • Z Excel is my best friend. I just love it.

Mentions of non-branded products/services that brand sells 

  • Just wondering how much it would cost to have my phone fixed or should I buy a new one

“I don’t like <brand> but I love their ___” 

  • I don’t like you but I fell in love with the new smoothie.

  • I usually don’t play XYZ but this new game is a bomb.”

“I only like <brand’s> ____” 

  • The only thing I like from PQR is their headset.

  • Perfumes are the only thing from MNO that I like.

Use of abbreviations (i.e., AF, LMAO, LMFOA, WTF) with positive Sentiment 

  • I just want a strawberry shake lmao 

  • This phone is good AF 

Use of Mild Swearing words 

  • This suit is hella good 

  • The experience is damn good 

Negative Emojis expressing concern or issue 

  • I wish this worked better for you 

  • I can really use some help here! 

Other non-competitor brand mentions that don’t violate direct competition guidelines 

  • Enjoying my McD on the couch with my Snuggie

  • All I want rn is Coke

Hyperbole in expressing desire or praise 

  • I’d kill for some coffee from ABC

Customer asking for products/services that are discontinued/limited-time 

  • I wish XYZ would bring back True color range in India!

Customer asking for products/services that are not available at their location 

  • What happened to the premium salon branch in Austin?

Posts related to customer requirements 

  • I am very particular about my health, we should have gluten-free options 

Posts concerning the difference in availability/prices of products or services 

  • I can’t find summer collection of shoes in New Jersey stores 

General complaints about broken products or services 

  • Coffee machines are always broken. 

  • I never get network in Texas 

Positive comments by customers in reaction to a shared article 

  • Amazing job! I don’t have think about paying in cash anymore 

Customer talking about small “freebies” received by the brand 

  • Today was my lucky day! I got an extra tall milkshake free 

  • Bonus Match passes, score! 

What Not to Engage
Message Type
Example(s)

Sharing News articles or updates about a brand

  • @MNO announces to end support for RED 10 from November. www.newswire.com/MNO-ends… 

  • ABCD one Limited edition coming this month. www.gamingtechworld.com/ABCD-” 

Brand or Product mention where the context is generic

  • Going to @WittyA’s house to grab some beer and check his ACB.

  • I have PQR 2016 crack. Anybody wants this.

Unsolicited ideas from customers via owned channels 

  • you would make $$$$ if you sold chips & drinks for $2!

  • You should sell convertibles 

Suggestions from Fans regarding brand’s Marketing or Management which will not affect the fans directly. 

  • DEF’s should bring in a new CEO who truly understands the Food industry” 

  • “@RED365 should be a free product and not sold.

Use of abbreviations (i.e., AF, LMAO, LMFOA, WTF) with Negative Sentiment 

  • WTF why in the world?

  • wtf is wrong with you people?

Negative emotions expressing hate or disgust 

  • These MNO is totally ****ed up! 

If someone comments on customer post with extremely horrific statement or offensive slurs 

  • NA

Strong swearing or sexually explicit language that is clearly written out 

(configurable by rule engine) 

  • Post with words like: b****, w****, etc, and N-word or any other racial slur, fag 

Self-censored profanity 

(configurable by rule engine) 

  • Abbreviations (e.g. “af”, “wtf”), 

Hangover/Drunk References 

(configurable by rule engine) 

  • I’m drunk and want to buy your new phone.

Reference to “food porn”, “car porn”, “fashion porn”, “real-estate porn” etc. 

(configurable by rule engine) 

  • Want to see more Food Porn?

Reference to illegal activity like drug use, underage drinking 

(configurable by rule engine) 

  • Damn, I love that dress when I stop smoking pot lmao 

Reference to brand being unhealthy, irresponsible, non-compliant etc. 

(configurable by rule engine) 

  • I’m really tempted to break my diet and get myself some junk food

  • I don’t care what toxic chemicals it has, I like the smell 

Other Competitor brand mentions 

(configurable by rule engine) 

  • NA

Posts that have no merit beyond their own personal agenda or desires 

  • Can someone buy me a new watch?

Posts that reference a program that hasn’t been officially rolled out, is no longer available, or isn’t part of the brand 

(configurable by rule engine) 

  • Can someone tell me about the Christmas offers?

Questions regarding private company financial data (e.g. sales figures) 

(configurable by rule engine) 

  • How much do you earn from selling books?

Posts that use the word “addicted” or “fix” in context with products/services fo brand 

(configurable by rule engine) 

  • I am totally addicted to this new game?

Posts asking for or mentioning health and dietary issues. 

(configurable by rule engine)

  • I want to try this out but I have lung cancer.

Straight retweets, where customer does not add a comment to the conversation. 

(configurable by rule engine)

  • NA

Messages in non-relevant languages 

(configurable by rule engine)

  • All non-English posts 

Based on the Social Profile

Classification can be enhanced by configuring profile conditions in the rule engine so that posts from only genuine profiles are marked as engageable. 

What to Engage
Message Type
Example(s)

Handles of genuine individuals (profiles not in exclusion, Do Not Engage, Spam or Blocked profile lists) 

  • When will this get launched in Europe?

Verified accounts by social network 

NA

Social Influencers 

(configurable by rule engine, using profile lists) 

  • Profiles with greater than 10K followers 

Mild curse words and self-censored swearing in profile bio/description 

(configurable by rule engine)

  • Damn beautiful & hell-raiser

What Not to Engage
Message Type
Example(s)

Parody accounts and fraudulent handles (using exclusion, Do Not Engage, Spam or Blocked profile lists) 

NA

Bio containing strong cursing, sexual references, drug references, violence or racial slurs should not be responded to* 

(configurable by rule engine)

  • Badass b****

Commercial, non-individual or media accounts (using exclusion, Do Not Engage, Spam or Blocked profile lists) 

Handles of businesses, school accounts, furry or fetishized personas, super-fan, media, industry-focused or other entities 

Brand employees on social media (using Employee profile lists) 

NA

Based on Conversation History

In order to maintain the context of the conversation, the rule engine can be configured to mark messages of relevant conversations as engageable. 

Rule 1: Messages in active conversations or have open cases are classified as engageable.

Rule 2: Messages older than a specified period of time e..g 4 days are classified as non-engageable.

FAQs

Yes, this solution is a combination of AI and Rule engine. The configuration is done based on the engagement guidelines of the brand. For e.g. “messages coming from verified profiles should be engageable” can be configured in rule engine while guideline like “User is complaining about the product” is done through AI Classification Engine.

We can claim 80%+ accuracy overall. However, based on our tagging, we have seen that the human teams also perform around 85% accuracy strictly based on engagement guidelines. Thus, the results will remain similar but at a faster speed without any manual intervention. 

Yes, the user can give feedback by changing the custom property (using macros for bulk). This feedback will be registered and the model will be trained further based on the feedback. 

The automation system takes only text part of the message into account. Photos/Videos/Links are not used to classify a message.If the content in photos/videos is also sent to the automation model, the engagement recommendation will be done only on the basis of text part of the message which includes photo/video captions as well.