SASARAN TECHNOLOGY home

Reference architecture

Illustrative

Agentic triage and reply drafting for a B2B support queue

A reference architecture where an agent gathers context, proposes a resolution and waits for agent approval.

Illustrative scenario. Not a completed client engagement. It describes how we would approach the problem; no client, results or quotes are implied.

/ 01Problem

The situation

A software company’s support team spends the first part of every ticket gathering context: account plan, recent incidents, product version, similar tickets. Response times slip during releases.

/ 02Baseline

Measuring before building

Ticket timestamps are used to measure time to first response and handling time per category over a recent quarter.

/ 03AI architecture

Architecture snapshot

Trigger
  • New ticket webhook
  • Category and priority classification
Agent tools
  • Read account plan and usage
  • Read incident status page
  • Search resolved tickets and docs
  • Create internal note (write, approval required)
Reasoning
  • Plan, gather, propose loop with step limit
  • Structured output: summary, cause hypothesis, draft reply, sources
Human control
  • Agent edits and sends
  • Escalation for security, billing disputes and outages
Quality
  • Scenario suite incl. prompt-injection tickets
  • Draft acceptance tracking
  • Trace review
Illustrative reference architecture.

/ 04Delivery

How it would be delivered

  1. 1

    Task analysis

    Decide which steps need an agent and which can be a fixed workflow.

  2. 2

    Tooling and limits

    Build read-only tools first; add the single write tool behind approval.

  3. 3

    Shadow mode

    Agent drafts in the background on live tickets; drafts are compared with what humans sent.

/ 05Outcome measures

What would be measured

  • Draft acceptance and edit distance
  • Time to first response by category
  • Escalation precision for sensitive tickets
  • Cost per handled ticket

Limitations

  • No customer-facing autonomous replies in this design.
  • Value depends on the quality of historical ticket resolutions.

No results are claimed for this scenario. Targets would be set from the measured baseline in a real engagement.

/ 06Client quote

Illustrative scenarios do not carry client quotes.

Want this architecture for your workflow?

We will adapt it to your systems and constraints and give you a phased, costed plan.