Build CX intelligence on an AI-native architecture

General-purpose models are trained to do a bit of everything. Lyra AI, our proprietary stack, is built for one job: turning customer feedback into reliable, structured insight. Our models are fine-tuned on each client's own feedback data, so the analysis reflects your customers' actual language and your business's taxonomy, not a generic view of the world. This purpose-built approach unlocks things general-purpose models can't do out of the box: consistent tracking of themes and sentiment over time, proactive surfacing of emerging issues, clustering of feedback at scale, and enterprise-grade customisation to your taxonomy and workflows. Your data is never used to train models for anyone else, and PII is excluded from training entirely.

Chattermill's core analysis is built on Aspect-Based Sentiment Analysis, performed by LLMs fine-tuned specifically for classification. These models identify the aspects customers mention and the sentiment attached to each, working entirely from what was actually said. Because they classify rather than generate text, there's nothing to hallucinate. Where our LLMs do generate text, in summaries and agentic workflows for example, every output is grounded in real customer feedback and traceable back to the underlying responses. We benchmark all models against human-annotated ground truth, monitor performance continuously, and keep expert analysts in the loop via a dedicated MLOps team.

Yes. You can connect Chattermill to any agent you already use through the Chattermill MCP, giving your agents access to verified customer intelligence that's reliable, consistent, and backed by quantified results.

An agent analyzing customer feedback is only as good as the intelligence it reasons on top of. Without structure and validation beneath it, the agent is left to improvise. So we built the agentic Customer Experience Intelligence architecture to give agents exactly that foundation.

The result is a multi-layer system, where data ingestion, the insights engine, analytical features, AI agents, agentic intelligence, and domain expertise are connected by design. Each layer feeds into the next, producing outputs that are deterministic, auditable, and reliable at enterprise scale.

An agent sitting on top of this stack is not improvising — it is reasoning on top of intelligence that has already been structured, validated, and made analytically meaningful at every level beneath it.

Chattermill automatically redacts PII from customer feedback at the point of ingestion — before it's stored or processed.

This includes common formats like email addresses, phone numbers, credit card numbers, and social security numbers, along with custom formats defined by your organization.

Once redacted, the information is permanently replaced and can't be recovered.