ReasonCX

An analytics agent that reasons over your support tickets and orders to produce an executive report — with its thinking shown.

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

Source	Data
Gorgias	Support ticket history
Shopify	Order data
Vector store	Both, embedded for retrieval

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

curl --request POST "https://api.stateset.com/v1/reason/reports" \
  --header "Authorization: Bearer $STATESET_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
    "window": { "from": "2026-08-01", "to": "2026-08-28" },
    "sources": ["gorgias", "shopify"],
    "question": "What is driving the rise in return requests this month?"
  }'

Because the reasoning is part of the output, responses are long and are streamed rather than returned whole. Read <contemplator> as it arrives and keep <final_report>:

const res = await fetch("https://api.stateset.com/v1/reason/reports", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.STATESET_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    window: { from: "2026-08-01", to: "2026-08-28" },
    sources: ["gorgias", "shopify"],
    question: "What is driving the rise in return requests this month?",
  }),
});

for await (const chunk of res.body) process.stdout.write(new TextDecoder().decode(chunk));

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.

Next steps

Verified decisions

The same auditability principle, applied to actions rather than analysis.

ResponseCX platform

The platform ReasonCX reports on.

RAG quickstart

Building the vector store it retrieves from.

Gorgias quickstart

Getting ticket history into that store.