I hope you enjoyed my presentation at Unbounce’s Call To Action Conference! Did you miss it live? Fear not! This page contains my presentation slides, notes, and links to all the resources mentioned.
Still have questions or need help analyzing your customers’ needs to improve your close rates? Get in touch!
What is sentiment analysis?
A field of study that attempts to determine the positive or negative emotions in a piece of text or speech. Scientists have been improving the methods since early research in the 1960’s, and the recent advances in processing speed, machine learning, and natural language processing have lead to more accessible tools that don’t require advanced programming knowledge to access.
Sentiment Analysis of a text results in two data points:
- Sentiment is the overall positive or negative emotional leaning of a text and is measured on a scale of -1.0 (negative) to 1.0 (positive) where 0.0 is neutral.
- Magnitude is the strength of a sentiment and is scored on a scale of 0.0 (low strength of emotion) to +infinity (high strength of emotion).
The sentiment and magnitude scores can be multiplied together to compare two texts. For example, a phone call that scores a 0.3 sentiment score and a 1.5 magnitude could be compared to a call with a 0.9 sentiment and 0.5 magnitude.
How can marketers use sentiment analysis?
Marketers that can understand their customers’ feelings before, during, and after a sale are able to improve the customer experience and reduce friction along the various paths to purchase. Instead of simply buying eyeballs and clicks, we can evolve to help close more sales.
In our example, we analyzed over 700 hours of phone calls (11,565 calls) generated by our digital marketing campaigns to learn what our clients’ prospective customers were feeling when they called to request more information and schedule sales consults.
How to get started
We have access to more data than ever before: social media posts, user generated content, phone call recordings, and customer service logs to name a few.
Hidden in this data are extremely powerful insights about the goals, expectations, pain points, frustrations, and attitudes that drive consideration and purchase decisions. We call this “the voice of the customer.”
Mining this data has traditionally been time consuming and difficult. A few hundred phone calls or social posts can easily run into dozens of hours or thousands of words. Sampling the data can be helpful but often feels like searching for a needle in a haystack. It is too much for humans to handle alone.
Enter the machines. Google, Amazon, and IBM are competing to own the cloud computing space and each has developed entire suites of machine learning and AI tools to help developers and analysts make sense of large data sets. You no longer need a PhD in computer science to access the same tools that scientists and mathematicians have used for decades.
Step 1: Compile your customer data
Gather up all of the data you can across your organization. My presentation focuses on phone call recordings and transcripts, but there are many other places to start looking:
- Phone call recordings and transcripts
- Customer service records
- User reviews
- Social media posts
- Other types of user generated content
Step 2: Convert all assets to text
Current sentiment analysis tools analyze the emotional leanings of text. Phone calls and other audio or video sources will have to be converted to text before they can be analyzed and compared. Luckily, there are several ways to do this. Some services allow for manual file uploads but it can be a tedious process with hundreds or thousands of files. If you have large data sets, have a programmer automate the upload and processing of these files through each service’s APIs.
At this point, you will have a choice of services that can process and analyze your data. Pick the one that’s best for you based on a combination of price and import/export options. Pricing is based on the number of records to be processed and the processing time.
- Amazon S3 (file storage) and Amazon Transcribe (speech to text transcription) are part of the AWS cloud. I found this to be an easy self-service platform for non-programmers to upload and process text or CSV files.
- Google Cloud (Google Cloud Storage, Google Cloud Speech to Text, and Google Cloud Natural Language)
- IBM’s Watson platform (speech to text and natural language understanding) can handle file storage and speech to text transcription, but do not offer as many self-service options for uploading files without accessing the API.
- Microsoft Azure’s speech to text and language understanding services are built on the same technology that drives Cortana and Skype.
Step 3: Determine the sentiment and magnitude of each record
No matter which platform you select, analyzing the sentiment of each record will result in a sentiment and magnitude score. Alone, these data posts can help you answer a lot of questions about your phone calls:
- Which callers are the most positive or negative?
- What percent of our calls are positive vs. negative?
- Are we generally creating more or less positive experiences over time?
Next, you can begin listening to individual call recordings or reviewing specific customer interactions to understand what happened this lead to a positive or negative experience. Identifying those needles in the proverbial haystack can help identify specific interactions that can become training opportunities.
Take it to the next level by combining the sentiment and magnitude scores with other data to learn even more about factors that contribute to a caller’s emotion:
- Call date/time: how does time of day or day of week influence caller sentiment?
- Number of calls and recency/frequency: if a caller calls multiple times, how does his/her emotion change as the number of calls increases?
- CRM integrations: do different customer segments have different experiences? How can you replicate successful interactions in one segment across others?
- Proximity: do callers nearby a place of business have a more positive experience than those farther away?
- Key words and phrases: which combinations of phrases yields higher or lower sentiment scores?
- Web analytics and campaign data: which traffic sources, campaigns, offers, or calls to action yield the most positive emotion?
- The data become even more useful as you interrogate it and layer in other sources of information. Look for anomalies and differences in scores between each dimension of your data. Digging into these interactions will lead to insights that were once locked up inside your organization!
Step 4: Act on the insights and become a better marketer
Once you identify which combinations of factors lead to happier customers, you can find ways to impact your customer experience and improve your chances of closing a sale.
For example, with insights generated by analyzing our 11,000+ phone calls, we are able to provide these strategic recommendations to our clients:
- Optimize your marketing mix and allocate more budget to channels that produce higher quality leads.
- Improve your website content by creating FAQ content to address common questions and sales materials that overcome common objections.
- Improve training and share examples of good calls and bad calls across your organization so that anybody that interacts with customers (sales teams, front desk/reception, and customer service) understand the factors that create happier customers.
- Minimize negative reviews by having managers or customer service proactively reach back out to callers that had a negative experience and offer a solution to fix their issues.
- Encourage positive reviews by soliciting your happiest customers with a follow up email asking for an online review.
By now, hopefully you can see the competitive advantage you’ll gain by listening to your customers and optimizing your sales funnel. Do these things you’ll be one of the select few digital marketers that has evolved beyond buying eyeballs and clicks into an indispensable marketing partner that helps close more sales.
Hope to see you next year at CTA Conference 2019!