In a digital world where every comment, review, and support ticket is a gold mine of information, a lot of companies find themselves overwhelmed. The volume of feedback is massive, and going through every single message by hand just to understand what customers are thinking is pretty much impossible. The result? Valuable insights get lost, chances to improve are missed, and the connection with your audience gets weaker.
What if there was a way to automate all of this? A way not just to collect, but actually understand the feeling behind every single word, quickly and at scale. The good news is this solution exists, and it’s available to every Tess user. In this article, we’ll dive into a practical and powerful use case: building an AI agent specialized in sentiment analysis. You'll learn how to build, from scratch, an AI capable of reading, classifying, and pulling insights from customer feedback, no matter what format it comes in.
To build an effective AI agent, everything starts with the prompt — the set of instructions that works as the brain of the whole thing. A well-crafted prompt makes sure the AI knows exactly what its job is, what it needs to do, and what’s expected in the end. For our sentiment analysis, we can use a prompt structure split into four key pillars:
"Role: Define who the AI is. In our case: "You are a highly specialized sentiment analyst, able to interpret and analyze texts in different formats, including customer comments, support tickets, emails, and content from files and spreadsheets."
Action: List the tasks the AI should perform. This includes reading the content, classifying the sentiment (e.g., very positive, positive, neutral, negative, very negative), identifying keywords, and providing a summary.
Context: Describe the environment where the AI is operating. Example: "You are working in a customer service environment, where different types of feedback are received daily. The data can come from several sources and formats (text, spreadsheets, PDFs)."
Expectation: Detail the expected response format. For example, for each item analyzed, the AI should give the sentiment classification, a justification, the main topics, and a summary."
A practical tip is to use a ready-made agent to help create the perfect prompt. Inside Tess, you can use public agents to refine your instructions, making sure your AI works well.
This first method is perfect for anyone dealing with feedback in multiple formats (spreadsheets, text documents, PDFs) and needs some flexibility. The idea is to create a conversational chat agent that can analyze different files with every interaction.
Step by Step:
Agent Creation: In Tess AI's AI Studio, start creating a new "Chat" agent. Drop in the structured prompt we set up earlier.
Starting the Interaction: When you open the chat with your new agent, it'll follow the prompt instructions and ask what format the material you want to analyze is in (text, file, spreadsheet).
Adding Data to the Knowledge Base: Let's say your comments are in a Google Sheets spreadsheet. You'll tell the agent that. Then, using the "Add Knowledge Base" feature, you'll provide the spreadsheet link and say which cell range should be read (e.g. Sheet1!A1:B41).
Running the Analysis: Once you upload the file, just give a simple command like: "All set. Do a sentiment analysis of the comments."
The AI will read the content of the spreadsheet in its knowledge base and provide a complete analysis, breaking down the sentiment of each comment, the keywords, and an overall summary.
This method is flexible since it lets you analyze an email PDF at one moment and an Instagram comments spreadsheet at another, all within the same chat.
Once the initial analysis is done, the true power of a chat agent shines. You can keep the conversation going to squeeze even more value from the data. For example, you could ask:
"Turn this analysis into a PDF report.": Using built-in tools, the agent can compile the whole analysis into a professional, formatted document that's ready to share.
"Create a pie chart showing the overall sentiment distribution.": By firing up advanced analysis tools, the agent can generate data visualizations like charts, making the insights a lot clearer and more impactful. This turns a list of comments into a visual representation of your customer's satisfaction.
What if your company puts all the feedback into one spreadsheet that's constantly updated? Uploading it again every time you want to analyze would be a pain. Here's where the second method comes in: set up a dedicated agent that's always hooked up to your data source.
Step by Step:
Refining the Prompt: First, tweak your prompt so it's more specific. Instead of saying "analyze texts in different formats," go for "analyze the customer comments spreadsheet." Drop any mention of other file types.
Creating an AI Step: In Tess AI, use the AI Steps feature. Create an "App Integration" step with the Google Sheets app.
Setting Up the Integration:
Pick the "Get Values" action, since we only want to read the data.
Paste the link to your Google Sheets spreadsheet.
Set the Data Range that holds the comments.
Connecting the Step to the Agent: Go back to your agent's prompt editing. At the end of the text, add a phrase like "The comments spreadsheet to be analyzed is as follows:" and then insert the AI Step you just created.
With this setup, the agent is already "born" knowing exactly which spreadsheet to analyze. The data source is permanently connected. Now, whenever you want an updated analysis, just open the agent and ask: "Generate today's sentiment analysis." No need for uploads or extra settings. The process becomes instant and totally automated.
AI-powered sentiment analysis isn't some futuristic tech reserved for giant companies anymore. With platforms like Tess AI, any business can build custom agents to turn a mess of customer feedback into clear, actionable insights.
Whether it's with a versatile chat agent for spot analyses or a specialized, automated agent for ongoing monitoring, the result is the same: a deep understanding of what your customers think and feel. That makes it easier to make quick, informed decisions, improve your products and services, and ultimately build a stronger, more positive relationship with your audience. Guesswork is over; the era of data-driven decisions is here.
