Knowing how your customers truly feel when they interact with your business is vital, but how can you collect this data?

In our previous articles, we have been discussing the benefits of using data analysis tools that use NLP (Natural Language Processing) and ML (Machine Learning) technology to ‘read’ and flag key information from the conversations you have with customers.  We stated that there are three vital components to get truly beneficial, valuable insights that can improve your growth and operations.  The first two components we discussed were Topic Identification (which you can read here) and Behaviour Identification (our last article, here).  Now we are on to the third and final component, Sentiment Analysis.

Sentiment analysis involves using ML to identify and catalogue communication based on the tones conveyed within.  They will rate the sentiment identified as either positive, negative or neutral with an overall score.  There are also tools on the market that can rate and tag the emotions identified, such as joy and sadness.  In our research here at Synetec, we reviewed IBM Watson Natural Language Understanding (NLU), Amazon Comprehend and Google’s Natural Language API.  As with all data analysis tools, the sentiment ratings across each one varied and we determined that a combination of systems and techniques would provide the best accuracy and most meaningful insight.  It is important to note that these tools can be trained, learning to associate certain meanings to specific keywords.  This is highly beneficial, because the language used by customers for one business may vary wildly from the language used with another business, industry, or country / culture and it is important to keep your data analysis in context.

Another important consideration is that sentiment analysis can involve reading and rating an entire conversation, but the result is likely to be less accurate or useful, since you cannot pinpoint what sentiment is related to which topic.  Therefore, when you want to identify how your customers are feeling, you need to pinpoint the specific communication that contains relevant topics and behaviours, then use these more specific pieces of communication for sentiment analysis.  For example, below is a string of conversation that we used in our research.

Sales person: “I would like one of our service people to stop by so that you can take advantage of our free inspection and cleaning. Is Wednesday afternoon at 2:00 p.m. a convenient time for you?”

Customer: “You know, I really don’t want to spend time or money on this now.“

Sales person: “I completely understand. <COMPANY NAME>is known for delivering efficient and affordable solutions for heating.“

Customer: “My burner seems to be working just fine.”

Sales person: “<CUSTOMER NAME> do you remember how cold it was last winter? …With our free annual inspections, you never have to worry about breakdowns during those fierce cold spells.”

Customer: “Okay, well, I suppose I could see you Wednesday.”

As a team, we processed the communication first manually, meaning we read it and decided what sentiment was present in the conversation.  We quickly discovered we all had differing opinions on the answer!  Because it is a long conversation, with a variety of answers and agreement for a follow-up call, depending on our personal viewpoints we chose neutral or positive.  Unfortunately, this does not tell us whether the customer was happy that they had been contacted, or whether they simply agreed to a follow-up call because they did not want to continue a conversation at that current moment.  The same applies to the tools, context is key.  Using the sentiment analysis tools in our research, we identified that the customer’s overall sentiment was negative, the first response is negative, but the final response was positive / neutral.  Below you can see the final ratings the tools gave us, which we scored out of 3 depending on their result.

As you can see, there are many factors involved in producing sentiment analysis results that are accurate and relevant to your business activities.  We would recommend that you use a combination of sentiment analysis tools and test them with pre-processed conversations to determine what works best for your business.  Decide on acceptable benchmarks for accuracy and actively train your systems so they learn what words are most important to your business.  For the most relevant results, use sentiment analysis in combination with topic and behaviour identification so you can identify exactly what conversations you should be analysing.  Flagging the different topics and behaviours your customers are exhibiting – in relation to specific business activities – ensures that the text analysed is small, so you can get the most valuable information without being lost in a tidal wave of data.

The data collected from daily business activities is incredibly valuable, the key lies in extracting the right data and then actively using the information where it is most impactful for your business

If you are interested in sentiment analysis or would like further advice on how machine learning and data analysis can be used in your business, please contact us.



Synetec is an Agile solutions provider with expertise in diverse development technologies, such as Angular, the .Net Framework, SQL Server and other cloud friendly data stores. We are certified and have successfully delivered projects across different cloud technology stacks such as Microsoft Azure and AWS, delivering integration and development solutions since 2000.

We work with a number of the UK’s most respected financial institutions to deliver a range of innovative solutions. We have expertise in working with both established businesses as well as start-ups and extreme growth businesses.

Leave a Reply