Introducing Social Signals and How We Went About Building an AI Entity

Reading time: 7 minutes

Written by

When businesses try to provide support on social media, not only do they get requests for help but also messages from prospects looking to try their product, people engaging with marketing campaigns, alongside compliments, insults and random messages. As a support agent, it can be difficult to go through all the tweets a business receives every day, filter out the messages asking for support and respond only to them.

We are happy to announce the launch of our first AI entity – Social Signals. We built Social Signals with the sole objective of helping businesses automatically figure out which tweets to prioritize and provide support. Social Signals goes through your entire feed, understands which tweets are asking for help and converts them to a ticket for an agent to answer, filtering out everything else. Social Signals learns continuously to improve itself and agents can teach it to recognize new types of tickets and requests anytime, making it better.  

This blogpost touches upon some interesting questions we had to answer when we first started building AI entities like Social Signals. The functions of our AI entities vary vastly and their capabilities are different as well. Any business looking to adopt AI for customer support will have to find a balance between what the AI is capable of and what the customer expects out of it. This blog talks about the guiding principles we adopted to bridge that gap between AI capabilities and customer expectations.

The Three Questions that helped build AI

Question 1: Where can AI help in a customer support environment?
Forrester says about 75% of AI entities currently used will underwhelm this year (in 2018) because the problems were not defined carefully enough. We wanted to be clear of the problem we could solve with Artificial Intelligence, and where exactly we could use AI in our support workflow. So we analyzed our own support structure to see where AI would be most effective.

At Freshdesk, we think of the support workflow in 4 segments. So we looked at how AI could potentially help in each of the segments. 

AI for Avoidance

When you call an Uber, in order to prevent just one ticket from you Uber uses AI to determine where your cab is, how long it will take for the cab to reach your place and pick you up, and what the ETA for your destination is.

Artificial Intelligence enables Uber to proactively support customers, and share information with them at the right moment. This helps Uber create a better, smoother customer experience with their product as well as reduce the tickets that come in to their support.

AI for Deflection

With a strong set of tutorial videos, knowledge base articles and self-help solutions, we found we could actually deflect upto 20% of the tickets coming into support to self-help articles so customers had answers faster. And one of the best ways to do deflect was with Google. 

With most customers searching for solutions on Google, if businesses made their support portals more SEO-friendly, they could use Google’s algorithms and AI abilities to deflect questions from their support portal directly to their own tutorial videos or guides.

Alternatively, chatbots can interact with customers coming into the portal, understand their problems and recommend the right solutions deflecting tickets, without agents having to enter the conversation at all.

AI for Efficiency

Some queries can neither be avoided nor deflected, but once they have been submitted as tickets in your helpdesk, AI can ensure that the ticket is assigned to the best agent available to respond to it. AI can recognize agent skills based on the tickets agents have solved earlier, look at incoming tickets and assign the same to them to ensure a fast and accurate resolution.

AI can also help cancel noise from social media, ensuring that agents in a helpdesk see only requests for help and interact only with customers reaching out for support. Any other interaction with people on social media can be easily weeded out by an AI entity like Social Signals.  

AI for Resolution

Whenever we speak of AI for customer support, everybody immediately assumes the AI entity is going to be used for resoution — where it talks to a customer, responds and resolves a ticket – all by itself. That is definitely a possibility.

However, AI like IBM Watson can provide more information to agents, who are trying to solve tickets. Watson can listen to conversations over the phone between agents and customers and pulls up the right information immediately for the agents based on keywords detected in what the customer says.

Once we had determined the four places where we could most effectively use an AI entity, and we had defined the problems AI for each segment would deal with, we had to determine whether our AI would work alongside our agents or replace them.

Question 2: Will AI replace our agents? Or will it assist them?
Most of the media reports nowadays talk about how the rise in automation and AI will take over human jobs. Bots will reportedly replace humans entirely. While this has already happened to some degree in the manufacturing sector and in the quality control / analysis space, we wanted to determine how effective AI would be for customer support – in improving efficiency as well as reducing cost.

The truth is AI, in its current condition, can be a massive asset to companies looking at economical ways to improve the quality of their support. AI can talk to customers and assist agents in gathering information. For example, Freshdesk’s experiments found that chat bots could respond upto a whopping 40% of the questions asked and was able to deflect anywhere from 8% to 25% of the questions based on the AI entity used and the quality of data provided to the AI to learn from. By gathering information from customers first, chat bots can reduce the number of exchanges between customers and agents, increase the resolution rates by as much as 15-20% leading to huge savings in time and money for businesses.

So we had to ensure that our AI entities would help support agents create a better experience for the customer by working alongside them.

Question 3: Is AI good enough for our support?
No AI entity has a 100% effectiveness rate at solving problems right out of the gate. When we speak to businesses regarding AI for customer support, they always ask us,

Is AI good enough for our support?

Businesses fear adopting AI for customer support because they worry the AI interactions will create a bad impression with the customer. But that AI entity will never improve unless businesses let the AI interact and learn from its mistakes.

A common maxim in the Machine Learning/Artificial Intelligence community is,

Good data beats good algorithm any day

So the discussion should not be only about whether AI is good enough for a business now. Businesses should also evaluate what they need to have ready in order to make use of the AI most effectively.

Yes, good AI entities will ensure decent 0-day performance with no training or learning. But by providing the right information to learn from, businesses can tailor an AI entity’s answers to be more accurate for their purpose. For example, a business might want to use a chatbot to deflect tickets. While the AI can deflect questions, businesses can increase effectiveness of the AI considerably by providing knowledge base articles for the AI to learn from, so the AI has better context when facing questions from the customer.

But how can a business determine what type of articles to write? Based on incoming queries, AI entities can also determine the type of articles that are needed and recommend the same for businesses to prioritize when publishing. Such an entity can increase the effectiveness of chatbots and is  one of the things we are actively working on right now.

How we are bridging the Gap

After assessing ourselves, our workflows and making sure our business was ready for AI, you might be wondering how we are bridging the gap. How are we using AI in customer support and still meeting customer expectations? We have a few principles that we adhere to every time we build a new AI entity or implement one:

  • AI does not stop a customer from contacting an agent
    If the customer wants to talk to an agent or is not satisfied with the AI’s answers, we make sure they are able to do so easily. At no point does AI deter the customer from receiving support.
  • AI provides extra value
    Be it Social Signals or any other entity, we make sure our AI provides extra value to the users. We pay close attention to the design of the entity so that it enhances the overall support experience for both the agents and customers. We ensure that even if the AI does not perform up to expectations, it does not impact the actual conversations between businesses and customers in any way.
  • Customer should always be aware of AI
    We make sure that whenever a customer is talking to an AI entity like a chatbot, they are very aware of it. Customers revise their expectations and are more open to rephrasing their questions if they know they are conversing with an AI. However, making an AI entity pretend it is human only frustrates customers faster leading to a poorer overall experience and satisfaction.

These are just some of our learnings from building AI for customer support. There are a lot more ways AI can influence the core support experience – the conversations between the agent and the customer – and we have some entities (like chatbots to deflect tickets) in the works. 

If you wish to try Social Signals out now, comment below saying “Count me in!” and we’ll get in touch with you shortly.  If you don’t want to try Social Signals but have used a different AI entity with your support before, let us know what the experience was like!

Subscribe for blog updates