At RedChirp, our mission is to make communication better for businesses and their customers.

RedChirp makes it easy for highly regulated businesses to text with their customers in all kinds of ways including one-on one, in bulk, through automations, drips and more. But our mission doesn’t end once a text message is sent; RedChirp is dedicated to helping our customers create the most delightful and effective conversations possible! Not all texts get the same results and we want to help our customers learn how to text in the ways that will earn them the most sales, create goodwill, and build strong and genuine relationships.

The opportunity

At RedChirp, we have long held suspicions and shared best practices about simple changes businesses can make in their texting to delight their customers more consistently. But these suspicions were nothing more than that – ideas we had based on anecdotal evidence. We did not have a way to know that we were right.

Advances in Artificial Intelligence (AI)

News about the stunning advances in the field of artificial intelligence is everywhere right now. This gave us an idea for an experiment: could we harness recent breakthroughs in AI to create evidence-based best practices for businesses to follow to better text their customers?

What we did

We began by thinking about the data that we have available and crafting a more precise problem statement, starting with our very popular webchat feature. This feature lets consumers on a business’s website request someone text them about a specific thing they need help with. And consumers use it a LOT! Over the past twelve months, just shy of 200,000 webchat requests have been submitted. These requests resulted in about 65,000 text conversations that included a significant amount of content; we defined a “significant amount of content” as more than five exchanges between the business and the consumer.

From this extensive dataset available for analysis, we crafted the following problem statement: What specific features of a business’s response to an incoming web-to-text request favorably influence the resulting conversation…and by how much?

Our approach

Amazon has a product, Amazon Comprehend, that is something of an industry standard for sentiment analysis (both targeted and untargeted). We took our data set of 64,621 conversations, removed all personally identifiable information, and used sentiment analysis to create a quantitative measure of sentiment in each of these conversations following a webchat request. Amazon Comprehend provides four scores for each conversation — Mixed, Positive, Neutral, and Negative — which we blended to create a single sentiment score. We used this to rate the sentiment of each conversation between 0% and 100%, with 0% being the worst conversation you can imagine and 100% being the best conversation possible.

We used this sentiment data to tweak the hyperparameters of an AI model set up to generate an ideal answer to a question posed by a consumer; that is, the answer most likely to increase the sentiment of the resulting conversation.

By looking at the features of the idealized responses versus the actual responses, we were able to identify patterns and then provide specific guidelines you can follow when crafting your own webchat responses to set yourself up for the best conversation possible.

What we learned

In priority order, we learned you should do the following when first replying to these web-to-text requests:

• Use the first name of the customer in your response.
  
  
• Introduce yourself using your full name and the organization you work for.
  
  
• Mirror the language used by the customer.
  
  
• Restate their question so the consumer knows that you have understood them.
  
  

Points 1 and 2 had an amazingly large impact on the sentiment of the ensuing text conversation. Simply doing those first two things is correlated with a significant improvement in the sentiment of the conversation 89% of the time.

Although this change is easy to make in theory, we found that fewer than half of actual conversations followed this best practice in real life.

Examples

Let’s take a look at these suggestions in action with a real-world example. To protect the innocent, we will assume that the customer’s name is Jane Doe and the employee’s name is Mary Watts with Barrel Creek Vineyards. Also, where the customer has made grammatical mistakes, we will not correct them.

Jane Doe: I got an e-mail notice of an upcoming shipment but I can’t find any such order on my account.  Need to either cancel or alter

Some common (but not ideal) responses to the above might read as follows:

• I can help you out with that
  
  
• Let me look into that for you
  
  
• Sorry to hear that, give me a minute to check into that for you
  
  

Our AI model, trained based on actual conversations, suggests the following five idealized responses:

• Hi Jane, this is Mary Watts from Barrel Creek Vineyards. I’m sorry to hear about the confusion with your shipment order. Let me look into this for you and get back to you as soon as possible.
  
  
• Mary from Barrel Creek Vineyards here. I apologize for the inconvenience, Jane. Can you please provide me with your account details so I can locate your shipment order and make the necessary changes?
  
  
• Hello Jane, it’s Mary with Barrel Creek Vineyards. Thank you for bringing this to my attention. I have canceled your upcoming shipment order, and you should receive a confirmation email shortly.
  
  
• Hi Jane, Mary from Barrel Creek Vineyards again. Just to follow up, I have located your shipment order and made the requested changes. You should see the updated information reflected on your account now. Please let me know if you have further questions or concerns.
  
  
• Jane, this is Mary Watts from Barrel Creek Vineyards. I apologize for any inconvenience caused by the confusion with your shipment order. We appreciate your loyalty as a club member, and we strive to provide the best service possible to our valued customers.

Wait…aren’t those suggestions just common sense?

The best practices themselves are certainly nothing new. For example, take the suggestion most highly-correlated with better sentiment — that you use the first name of the customer in your response. As early as 1936, Dale Carnegie wrote in How to Win Friends and Influence People that:

“A person’s name is to him or her the sweetest and most important sound in any language.”

So, certainly, this advice is nothing new. That does not, however, mean that our employees — or ourselves — consistently implement this best practice. In our analysis of a subset of the contact requests, this simple suggestion to use the first name of the customer in the initial response was followed less than 40% of the time.

This led us to a startling revelation:

When replying to a customer, it’s much easier to be influenced by examples than it is to remember rules.

Working on this experiment taught us just how difficult it can be to write a natural-sounding reply that implements the four best practice suggestions our research identified. It’s hard to hold the rules in your head and remember to use them consistently while in the “heat of battle”.

We found, however, that it is far easier to respond properly when we were looking at AI-generated idealized responses. Somehow, having some ideal responses in front of you makes replying much easier!

So…

Introducing AI-generated contact responses in RedChirp!

Now when you reply to a webchat request, RedChirp will automatically display 5 AI-generated idealized responses. Compose your own unique response with the benefit of these helpful examples in front of you, select any AI-generated response to send as it is, or select an AI-generated response as a starting point that you edit before sending.

It’s never been easier to introduce your team to working with practical and helpful AI, and to improve the performance of your webchat requests without any extra training or effort.

It’s helping already!

As soon as this improvement went live, we started hearing from chirpers about how much it helps save time and delight their customers, like this text from Adam Gurzenski, E-Commerce & Inside Sales Manager with Peju Province Winery:

Learn more about RedChirp

Have a few questions? Complete this form to start texting with us!

Ready to see for yourself? Schedule a demo: https://calendly.com/jennie-gilbert/redchirp-demo

texting sms text marketing sms marketing AI artificial intelligence webchat dtc