Understand how Qvasa automatically reads every support conversation, assigns meaningful categories and reason codes to deliver ready‑to‑report insights.
1. At a Glance
What the AI does | Why it matters |
Analyzes ticket data as it happens | No waiting! Reports always reflect the latest data. |
Decides which high‑level category a ticket belongs to | Quickly understand why customers are contacting you. |
Generates (or re‑uses) concise reason codes | Detects emerging bugs, confusing UX flows, or common agent tasks. |
Stores results as structured fields inside Qvasa | Slice & dice any dataset with the same filters you apply to status, channel, or assignee. |
TL;DR Qvasa combines two kinds of AI workflows, fixed categorization and generative, to turn raw conversations into actionable, report‑ready data the moment a ticket is closed (or any status you specify).
2. Key Concepts
2.1 Real‑Time Conversation Processing
When a new ticket event hits Qvasa, we stream the messages, condense them into a summary, and pass that summary to our analysis engine. Because we work on the fly, dashboards never lag behind the help desk. Workflows can be configured to run at different points in a tickets life cycle, or multiple times during a tickets life cycle, depending on the intended use for the data
2.2 Fixed Categorization Workflows
Think of these as multiple‑choice questions for your tickets.
Defined once, reused forever – You supply the list of categories (e.g., Billing, Bug, How‑to Question).
Context–aware – For each option you add a short “system explanation” that tells the AI exactly when to pick that option.
Because the options are fixed, the output field is perfectly consistent and ideal for high‑level reporting and trend lines.
2.3 Generative Workflows
Reportable open‑ended questions. These workflows ask the AI to respond in a bounded format you control:
Prompt question – e.g., “What is the product deficiency?”
Word length limits – e.g., 3–7 words keeps codes crisp.
Business context – Extra background improves accuracy.
Intelligent reuse – The AI first tries to match one of the codes already in your private database and if it can’t, the AI will create a brand-new code that describes the ticket.
This balance means you get fresh insights as they appear with code re-use for reporting.
2.4 Nested & Conditional Workflows
A generative workflow can run only when a certain fixed‑choice category is selected.
Example: Run the Product Deficiency generative workflow only if the top‑level category is product deficiency. This ensures we aren’t asking the AI to create generative codes for irrelevant conversations which minimizes noise in the database.
3. The “Product Contact” Out‑of‑the‑Box Workflow
To demonstrate the system, every new Qvasa account ships with this example two‑step workflow:
On ticket close, the AI picks one of these four categories:
Option | When we choose it | Follow‑up Workflow |
product_deficiency | A true defect, bug, or missing feature | Product Deficiency generative |
product_confusion | User needed an explanation/how‑to | Product Confusion generative |
agent_intervention | Only an agent‑side action solved it | Agent Intervention generative |
product_uncategorized | Doesn’t fit the above or unclear | (none) |
Generative Reason Extraction (runs conditionally)
Each follow‑up asks an open‑ended question—but with strict rules so the output stays dashboard‑friendly:
Product Deficiency → “The product deficiency must be 3–7 words … What is the product deficiency?”
Product Confusion → “… What is the product confusion?”
Agent Intervention → “… What is the agent intervention?”
Behind the scenes, Qvasa passes a curated list of known reasons to encourage reuse, plus the ticket summary, plus your business context. The generator then either selects an existing code or invents a new one that fits within the guidelines of the generative workflow.
3.1 Some Things You Can Report On
Filter any widget by Defect, Confusion, or Intervention.
See Top 10 new product deficiencies week over week
Compare average CSAT for tickets that required agent intervention
4. Building Your Own Workflows
Ready to move beyond the starter pack? Here’s the general recipe.
4.1 Design Clear Fixed Categories
Make them mutually exclusive – Overlaps hurt accuracy.
Add concise system explanations – A single sentence usually suffices.
Tip: Ask yourself, “Could two agents disagree on this category?” If yes, refine the wording.
4.2 Craft Generative Definitions
Field | What to include |
Name | Shown in the UI & reports. |
Prompt Question | Start with a directive (“What is the …?”) |
Business Process Context | Who is contacting you, why, product jargon, etc. |
Lower / Upper Word Length | 2–5 words for laser focus; 4–10 for richer phrasing. |
Keep the prompt tight, but load the context with anything that will disambiguate like SKU names, release phases, regional policy differences, etc.
4.3 Link Workflows Together
Within the definition of your fixed categorization AI workflow you can select generative workflow definitions to connect them to categories.
5. Reporting & Dashboards
Because analysis is done before data hits a widget:
Instant filtering – Everything is already tagged when you go to do your reporting
Baseline + drill‑down – Start broad (product_confusion) then drill into reasons ("Pricing page unclear").
Proactive alerts – Set an alert on New Product Deficiency codes > 5 tickets in 24 h.
Time‑sliced trends – See whether agent interventions drop as you automate workflows.
With Qvasa’s layered fixed and generative workflows, you’re no longer limited to surface‑level tags. Each ticket becomes a data‑rich insight that evolves as your business does. With this analysis always at your fingertips, Qvasa is powers smarter reports and better business decisions for its users.