New Release: Open Ticket AI 1.4.1 — Explore the production-ready platform
Open Ticket AI

Ticket Automation Planner — Pick Your Best Path ​

Modernize ticket routing fast—no matter where you start. This planner helps you choose the right path based on your data reality: lots of labeled tickets, lots of unlabeled tickets, or almost no data. Each path ends in a concrete service package with clear deliverables and KPIs, so you can move from idea → pilot → production without guesswork.

Who this is for: IT/service teams on Znuny/OTRS/OTOBO (or similar) who want reliable queue/priority/tag predictions, either on-prem or via a hosted API.

What you’ll get: short decision flow, 4 actionable paths (A–D), add-ons (multilingual, extra attributes), gates/metrics to know when you’re ready, and a data-readiness checklist.

How to use this page

  • Start with the one-screen overview and answer three questions: Labeled? → Unlabeled? → Fast?
  • Click the box for Flow A/B/C/D to jump to its steps, deliverables, and KPIs.
  • Use the add-ons if you need multiple languages or more outputs (tags, assignee, first answer).
  • Keep the gates tight (per-class F1 + business KPIs) so pilots translate to production confidence.

Now continue with the overview diagram and the detailed flows below. Nice—here’s a fuller write-up you can drop under your diagrams. I kept it skimmable but added real guidance and thresholds so readers can confidently pick a flow.

Got it — I’ll keep your new short diagrams and add clear, concise explanatory text for each section so the article feels complete while still being easy to scan.

0) One-screen overview ​

How to use this overview: Start at the top, answer the questions, and follow the branch to your matching flow. Click a flow to see its details.

Flow A — Many labeled tickets ​

When to choose this:

  • You already have thousands of tickets with queue, priority, or tag labels.
  • You want a custom-trained model for maximum accuracy.

What happens in this flow:

  1. Audit/Tax — Check label quality, class balance, and naming.
  2. Train — Fine-tune the classification model with your data.
  3. Eval — Measure per-class precision/recall/F1.
  4. On-Prem — Deploy inside your own infrastructure.
  5. Pilot — Test in production with monitoring.
  6. Support — Iterate and retrain as needed.

Recommended package: Fine-Tune + On-Prem Install.

Flow B — Many unlabeled tickets ​

When to choose this:

  • You have large historical ticket archives but no labels.
  • You can allocate some human review time for quality checks.

What happens in this flow:

  1. Ingest — Collect tickets from your system.
  2. Auto-Label — Use LLM-assisted auto-labeling.
  3. QC — Spot-check & correct samples.
  4. OK? — Loop until quality meets threshold.
  5. Train — Fine-tune with the curated set.
  6. Eval / On-Prem / Support — Same as Flow A.

Recommended package: Auto-Label + Fine-Tune.

Flow C — Few or no tickets ​

When to choose this:

  • You're starting from scratch or have too few tickets to train on.
  • You want a cold-start solution to go live quickly.

What happens in this flow:

  1. Define Tax — Decide queues, priorities, tone.
  2. Synth Data — Generate realistic tickets (DE/EN).
  3. Baseline — Train initial model on synthetic data.
  4. Eval — Check performance before rollout.
  5. Pilot — Choose Hosted API for speed or On-Prem for control.
  6. Collect — Gather real tickets during pilot.
  7. Fine-Tune — Merge real + synthetic data.
  8. Prod/Support — Go live with ongoing iteration.

Recommended package: Synthetic Cold-Start.

Flow D — Quick start via Hosted API ​

When to choose this:

  • You need results immediately.
  • You want to try automation without training first.

What happens in this flow:

  1. Use API DE — Instant classification via hosted German model.
  2. Measure — Track routing, SLA, backlog impact.
  3. Tax OK? — If satisfied, scale usage; if not, go to Flow B or C for training.

Recommended package: Hosted API Pilot → Fine-Tune (optional).

Optional add-ons ​

Multilingual expansion ​

Add support for additional languages via multilingual auto-labeling or synthetic generation, then train and evaluate per locale.

Extra attributes ​

Predict more than queues/priorities — e.g., tags, assignee, or first answer time — by extending labeling and training a multi-task model.