Skip to main content

​
ReasonCX

ReasonCX is an autonomous analytics agent. Point it at your support tickets and order data and it produces an executive-level report — ranked issues, quantified impact, and recommendations with the evidence behind each one. The distinguishing feature is that it shows its reasoning. Rather than emitting a conclusion you have to take on faith, it works through the data in the open, questions its own inferences, and revises them before committing to findings.

​
What it analyses

Analysis is grounded strictly in the retrieved ticket and order context — it does not draw on outside knowledge to fill gaps.

​
Output shape

The agent emits two sections. <contemplator> — the reasoning process. Foundational observations, each insight questioned, uncertainties expressed and explored, interpretations revised, individual findings connected into patterns. This is the audit trail for the conclusions. <final_report> — the deliverable:
  1. Executive summary — critical issues, key insights, actionable recommendations
  2. Impact analysis — quantified customer effects, financial implications, resource use
  3. Issue prioritisation — ranked urgent issues with explicit criteria and ticket evidence
  4. Deep analysis — statistical patterns, anomalies, correlations, supporting examples
  5. Strategic recommendations — implementation plans, timelines, resources, success metrics
  6. Anticipated questions — cost, resource allocation, expected outcomes
Read <contemplator> when a finding surprises you. Because the agent shows where it was uncertain and what it revised, you can tell the difference between a conclusion the data strongly supports and one it reached tentatively.

​
Why the visible reasoning matters

An analytics agent that returns only conclusions is impossible to audit — you cannot tell whether a recommendation follows from the data or from a plausible-sounding leap. ReasonCX is built the other way round: the contemplation is part of the output, so a finding can be traced back to the tickets that produced it. That’s the same principle as verified decisions, applied to analysis rather than authorisation. There, a decision carries a cited proof. Here, a finding carries its derivation.

​
Architecture

The reasoning API is modular, with separate services for embeddings, vector storage, code execution, and model calls. Configuration is environment-driven — no credentials in code — and errors are handled so they don’t leak sensitive detail.