Understanding Dimensions

Any pice of textual data contains many different dimensions of meaning. We use AI to extract information from the raw text, offering a powerful solution for structuring data at scale and making comparisons.

Sentiment is a Question

You are probably familiar with sentiment analysis used to categorize unstructured text according to positive, negative and neutral tonality.

Sentiment analysis may be thought of as posing a simple, straightforward research question about the tone of textual data. The resulting value of positive, negative or neutral is the answer to that question.

❓
Sentiment asks a question:
Question: What is the tone of the conversation?
Answer: positive, negative, neutral

Executed across thousands of conversations a sentiment analysis reveals the overall proportion of answers—which is how the metric is often presented:

e.g., 27% of conversations with customer support in March were characterized by a negative tone.

Dimensions Answer New Questions

Dimensions similarly provide answers to business questions. Today we can go beyond basic sentiment to ask any question of a document. With the advent of LLMs we may classify data using prompts instead of relying on machine learning algorithms.

❓
Outcome dimension asks a question
Question: What was the outcome of the conversation?
Answer: completed, escalated, abandoned

Executed across thousands of support conversations the overall proportion of escalations becomes clear

e.g., 19% of live chat Support conversations were escalated to a manager in June

Big Idea:

🚧
The Breakthrough:
While sentiment analysis tells us the tone of customers engagements, dimensions offer unlimited potential to extract any meaningful information from text about what happened, why it happened, and what to do about it.

Powerful Question Types

Transform your data with these specialized approaches:

Annotation-Based Questions

Products: What products were mentioned in the conversation?

Purpose: Identify specific items, services, or features discussed during interactions.

Real Impact: Discover which products generate the most support requests, revealing quality issues before they escalate.

Scoring-Based Questions

Predicted Rating: How would the customer rate their interaction on a scale of 1-10?

Purpose: Quantify satisfaction levels without explicit customer ratings.

Real Impact: Predict customer satisfaction in real-time, enabling proactive intervention before customers churn.

Summarization-Based Questions

Predicted Feedback: How would the customer describe the conversation?

Purpose: Generate customer perspective summaries for quality assessment.

Real Impact: Understand your service from the customer's viewpoint, identifying gaps between internal metrics and customer experience.

Recommendation-Based Questions

Predicted Solution: What could be done to resolve the issue going forward?

Purpose: Identify actionable steps for issue resolution and prevention.

Real Impact: Build a knowledge base of proven solutions that continuously learns from successful resolutions.

Answer Open-Ended Questions

A key feature of dimensions is the ability to pose open-ended questions.

Beyond manual tagging, traditional text analytics rely on users building complex keyword searches, training machine learning models or using out-of-the box segmentations—and with every approach categories are determined in advance.

In contrast, Dimension Lab's may dynamically label and categorize data based on the context and content of conversations and documents. This method is more adaptable and responsive to emerging themes and topics within the data.

👉 Reason - What is the reason the customer reached out to support?
👉 Category - What is the overall theme of the conversation?

Answer clustering to create levels of analysis

Some Dimensions employ answer clustering to create a hierarchy of analysis. In addition to Reasons \ Categories, other dimensions employ answer clustering as well.

👉 Predicted Solution - What is the potential resolution to the customer issue?
👉 Solution Cluster - What are the themes resolutions?