Sentiment Engine Upgrade - FAQs

We're upgrading the Sentiment Engine to Claude Sonnet — a state-of-the-art large language model that delivers deeper accuracy and greater nuance than traditional NLP classifiers. The following answers address the questions we expect you'll have.

1. Why are you changing how overall sentiment is calculated?

We're upgrading our sentiment engine from AWS Comprehend to Claude Sonnet, a state-of-the-art large language model. AWS Comprehend uses a statistical classification approach that assigns sentiment based on keyword patterns and phrase structures. While reliable for straightforward feedback, it struggles with nuance — things like sarcasm, mixed emotions, or context-heavy responses. Claude Sonnet actually understands language in a much deeper way, allowing it to interpret what a customer truly means, not just what words they used.

2. Will I see any changes to my existing sentiment scores?

You may notice some differences, particularly on responses that contain nuanced or complex language. Feedback that previously returned a generic "neutral" score may now be more accurately classified as mildly positive or negative, better reflecting the customer's real experience. Scores on clear-cut positive or negative feedback are unlikely to change significantly. We'd recommend treating this as a recalibration towards greater accuracy rather than a change in your customers' behaviour.

3. Will I see improvements in accuracy?

Yes. Claude Sonnet has been trained on a vastly larger and more diverse dataset than a traditional NLP classifier. It understands context, tone, and intent in a way that rule-based and statistical models simply can't match. This means fewer misclassifications on ambiguous responses, better handling of longer or multi-topic feedback, and a more faithful representation of how customers actually feel.

4. Will sentiment score decisions be more decisive than AWS Comprehend?

Yes, and this is one of the more noticeable differences you may observe. AWS Comprehend has a tendency to return "neutral" as a default when it encounters ambiguity, mixed phrasing, or language it isn't confident classifying. This can result in a higher proportion of neutral scores than truly reflects your customers' sentiment.

Claude Sonnet is far more capable of reading between the lines. It will make a more considered judgement on ambiguous responses, meaning you're likely to see a shift away from neutral scores and a more decisive split between positive and negative classifications where that more accurately reflects the feedback. The result is a sentiment distribution that better mirrors real customer emotion — giving you cleaner, more actionable insight rather than a middle-ground score that's difficult to act on.

It's worth bearing this in mind when reviewing your sentiment trends post-migration. A reduction in neutral scores isn't a sign that your customers' experiences have changed — it's a sign that we're now capturing the true shape of how they feel with greater precision.

5. What are the key benefits of moving to Claude Sonnet?

The primary benefits are accuracy, nuance, and consistency. The model is far better at handling mixed sentiment — for example, a customer who says "the engineer was fantastic but the wait time was unacceptable" will now have that complexity reflected in their score rather than being flattened into a single middling result. It also handles industry-specific language and colloquialisms more effectively, which is particularly relevant for sectors like utilities and housing where customer language can be highly contextual.

6. Does this affect how sub-scores or driver analysis works?

Overall sentiment is the primary area of change at this stage. Any impact on downstream metrics or driver-level scores will be communicated in advance, and we'll provide clear guidance before any further changes are rolled out.

7. Will my historical data be re-scored?

Historical responses will not be automatically re-scored. Going forward, new responses will be processed through the upgraded engine. If you'd like to discuss retrospective analysis for specific time periods or campaigns, please speak to your Customer Success Manager.

8. Should I update my internal benchmarks or reporting thresholds?

We'd recommend reviewing your sentiment baselines following the upgrade, particularly if you use sentiment scores as KPIs or thresholds for alerts and escalations. Your CSM can work with you to understand what, if anything, needs recalibrating in your dashboards or workflows.

9. When is this change happening?

Monday, 2 March 2026. You'll receive advance notice and we'll be on hand to support you through the transition.

10. Will this impact topic sentiment scores?

No, This is just the beginning of our sentiment analysis improvements. We're currently working on enhancements that will help you:

• Identify topics and score topic sentiment within feedback more accurately.

• Surface insights that are more relevant to your business.

• Spot trends and issues faster.

11. Who can I contact if I have questions?

Your dedicated Customer Success Manager is your first point of contact for any questions about this change and what it means for your specific setup.

Need more help?

Your dedicated Customer Success Manager is your first point of contact for any questions about this change. Don't hesitate to reach out — we're here to support you through the transition.