Predicted scores are part of a suite of advanced analytics known as Dimensions. They are used to quantify customer experience. These scores are derived from the raw text of conversational data to provide scores without the need for surveys. As a result, predicted scores may serves as a proxy for direct customer feedback, enabling continuous monitoring across a wider range of customer channels.

Predicted Rating

The "Predicted Rating" in Dimension Labs is part of a suite of analytics that are provided out-of-the-box (i.e., standard dimensions). It is a score that can be applied nearly universally across an omni-channel dataset without the need for any configuration.

It assigns a rating from 1-10 that a customer might give based on the analysis of their conversation. It's used for gauging customer experience.

👉 Examples: A conversation with where the customer’s issue was resolved might be assigned a high predicted rating, indicating a good customer service experience.

Understanding the Predicted Rating Dimension

The Predicted Rating analyzes the content and context of conversations using AI models. The system is able to predict a rating that reflects the customer’s feeling about the conversation. This rating may be compared with scores like NPS.

Calculation of Predicted Rating

The Predicted Rating is calculated at the session-level, analyzing the full text of a conversation to predict how a users would rate their experience on a scale, from 1 to 10, with 10 being the highest. The overall Predicted Rating Score is then averaged across all sessions.

💡 Prompt: An estimation of the success in handling the user's requests, rated on a scale of 1 to 10 (with 10 being the best).

Implications of Predicted Rating

The Predicted Rating offers valuable, turnkey insights into customer experience. It helps in understanding how customers perceive their interactions and the quality of service they receive. This can be especially useful in scenarios where explicit customer feedback (like survey responses) is limited or unavailable.

Using Predicted Rating for Improvement

This score can be used to identify areas of concern ranging from agent performance to customer frustration with various products and services. The Predicted Rating serves as a proxy for direct customer feedback, offering a continuous stream of insights derived from everyday customer interactions.

Predicted CSAT

The "Predicted Rating" in Dimension Labs is part of a suite of analytics known as custom dimensions.

It assigns a satisfaction score between -2 and +2 (Very/Somewhat Dissatisfied, Neither, Somewhat/Very Satisfied) that a customer might give based on the analysis of their conversation.

👉 Examples: A conversation with where the customer was provided with prompt service might be assigned a high Predicted CSAT rating.

Understanding the Predicted Rating Dimension:

Predicted CSAT is calculated by assigning a satisfaction rating to every interaction or piece of content, regardless of its source. Because the score does not rely on customer surveys it may provide a holistic view of satisfaction that extend VoC analytics across more data sources.

Calculation of Predicted Rating

The Predicted CSAT is calculated at the session-level, analyzing the full text of a conversation to predict how a users would rate their satisfaction: very satisfied (+2), somewhat satisfied (+1), neither satisfied or dissatisfied (0), somewhat dissatisfied (-1), very dissatisfied (-2). The overall Predicted CSAT Score is then averaged across all sessions.

💡 Prompt: An estimation of the user's satisfaction, rated on a scale of -2 to +2 (with +2 being very satisfied, -2 being very dissatisfied, and 0 being neither satisfied or dissatisfied).
Lense: Session

Implications of Predicted Rating

The Predicted CSAT score provides businesses with insights that would traditionally require extensive survey research, but without the associated limitations like low response rates. This metric is particularly valuable in scenarios where direct customer feedback is limited or challenging to obtain.

Using Predicted Rating for Improvement

This score is instrumental in identifying areas for customer experience enhancement. By analyzing reasons and categories derived from customer feedback, businesses can pinpoint specific aspects of their service that impact customer satisfaction.

Transactional NPS

NPS is traditionally a relationship metric — it measures how a customer feels about your brand overall. Transactional NPS narrows that lens to the individual interaction level: how did this specific experience move the needle on that customer's likelihood to recommend you?

This is an important distinction. Predicted Satisfaction tells you whether a customer walked away from an interaction happy or unhappy. Transactional NPS is asking a different question — did this interaction create or erode brand advocacy? A customer can have a satisfactory interaction that still wouldn't move them to recommend you, and a single bad experience can turn a long-time promoter into a detractor.

A few things to keep in mind:

This is a directional indicator, not a true NPS score. It's inferred from the interaction rather than collected via survey, so treat it as a signal for where to focus rather than a reportable number.
It's most useful for identifying risk. Interactions that score in the detractor range (0–6) are the ones most likely to damage overall brand perception — those are your priority for deeper investigation.
Null values are intentional. Some interactions just don't carry enough signal to indicate brand impact one way or the other. Those are filtered out rather than forced into a score.

How to think about the scoring

The model isn't just picking a number — it's reading against specific behavioral anchors for each NPS bucket. Detractor-range scores (0–6) are driven by signals like complaint language, unresolved issues, and expressed frustration. Passive scores (7–8) reflect a neutral tone — the customer is satisfied but not enthusiastic, with no strong emotional signal in either direction. Promoter-range scores (9–10) indicate delight, exceeded expectations, and language that suggests the customer would actively recommend you.

This calibration is what separates it from a general sentiment score. It's specifically tuned to the question "would this person recommend us after this experience?" rather than just "was this person happy?"

Where it gets actionable

On its own, transactional NPS tells you the what — this interaction was likely brand-damaging or brand-building. The real value comes when you cross-tabulate it against other dimensions in the dataset. Layering in issue type, product, channel, agent, or resolution status lets you move from "we have a detractor problem" to "we have a detractor problem driven by billing disputes in chat that go unresolved after multiple contacts." That's the level of specificity that lets you take targeted operational action rather than reacting to an aggregate score.

It also gives you coverage that traditional survey-based NPS can't. Average survey response rates for CX measurement can run as low as 7%, and research shows that promoters are significantly more likely to respond than detractors — meaning the customers you most need to hear from are the ones least likely to tell you.¹ ² This scores every interaction in your dataset, eliminating those coverage gaps.

The prompt behind the score

For reference, here's the instruction that drives this dimension. We've included two versions — the first optimized for precision using research-backed behavioral anchors, the second optimized for simplicity and speed.

Version 1 — Research-anchored

Predict the Net Promoter Score the customer would give based on this customer interaction. This measures brand advocacy impact, not just satisfaction. Score 0-10 where: 0-6 = Detractor (concrete complaints naming specific failures, broken promises or unmet commitments, repeat contact about unresolved issues, explicit statements of unwillingness to recommend, language indicating active consideration of alternatives), 7-8 = Passive (functional and transactional tone, issue resolved but no emotional engagement, hedging or conditional language such as "it's fine" or "not bad," absence of both frustration and enthusiasm), 9-10 = Promoter (emotional and relational language, references to specific people or features that exceeded expectations, expressed enthusiasm or gratitude, language indicating willingness to recommend or refer others). Output null if the interaction is too short or transactional to carry meaningful advocacy signal.

Version 2 — Streamlined

Predict the Net Promoter Score the customer would give based on this customer interaction. Focus on whether this experience would make the customer more or less likely to recommend the brand — not just whether they were satisfied. Score 0-10 where: 0-6 = Detractor (frustrated, unresolved issues, broken commitments, would not recommend), 7-8 = Passive (satisfied but indifferent, no strong emotion in either direction, functional tone), 9-10 = Promoter (delighted, exceeded expectations, expressed enthusiasm, would actively recommend). Output null if insufficient signal to make a prediction.

The key difference from the original prompt is the explicit framing around brand advocacy rather than satisfaction, and the use of more specific behavioral anchors — particularly for passives, where research shows their language patterns actually align more closely with detractors than promoters despite the moderate score.³ ⁴

Validation

The transactional NPS dimension is grounded in the Net Promoter framework developed by Fred Reichheld at Bain & Company, first published in the Harvard Business Review in 2003.⁵ The behavioral anchors in the scoring prompt are drawn from Bain's research on the economics of NPS — which found detractors account for over 80% of negative word-of-mouth while promoters generate nearly seven times as many positive referrals — and from academic research establishing distinct language patterns across NPS segments.⁶ ³ ⁴

The predicted approach addresses well-documented limitations of survey-based measurement. Bain's own analysis demonstrates that non-response bias can dramatically distort results, showing scenarios where a company's apparent NPS of +50 masks a true NPS of −22 once non-responder behavior is accounted for.² Predictive NPS models — endorsed by both Bain and McKinsey as the next evolution of CX measurement — achieve overall accuracy above 80% for classifying customers into the correct NPS segment.¹ ⁷

Predicted scores are directional. They should be used to identify patterns, prioritize action, and surface risk — not to replace formal NPS reporting. Cross-tabulating this dimension against others in the dataset produces more reliable insight than reading any single score in isolation.⁸

NPS Validation Sources

McKinsey & Company, "Prediction: The Future of CX" — https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/prediction-the-future-of-cx
Bain & Company, "Net Promoter: Creating a Reliable Metric" — https://www.bain.com/insights/creating-a-reliable-metric-loyalty-insights/
Adams et al. (2024), "From Detractors to Promoters: A Comparative Analysis of Patient Experience Drivers Across NPS Subcategories," International Journal of Nursing Practice — https://pubmed.ncbi.nlm.nih.gov/37648254/
Cendyn, "What Makes a Guest a Promoter, Neutral, or Detractor?" — https://www.cendyn.com/blog/promoter-neutral-or-detractor/
Reichheld, F. (2003), "The One Number You Need to Grow," Harvard Business Review — https://hbr.org/2003/12/the-one-number-you-need-to-grow
Bain & Company, "The Economics of Loyalty" — https://www.bain.com/insights/the-economics-of-loyalty/
Bain & Company, "Sidestepping Survey Fatigue: Predictive NPS" — https://www.bain.com/insights/sidestepping-survey-fatigue-predictive-nps-gets-to-the-heart-of-customers-attitudes/
McKinsey & Company, "From Touchpoints to Journeys" — https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/from-touchpoints-to-journeys-seeing-the-world-as-customers-do