AI Tasks: Enable and Manage
An AI Task is a purpose-specific analysis definition that reads a conversation transcript/thread and writes one or more outputs into Custom Fields.
In MiaRec, an AI Task typically includes:
- Prompt (instructions to the AI model, including output format)
- Attribute mapping (which output attribute populates which Custom Field)
- Optional filters (which conversations the task applies to)
- AI engine selection (LLM provider/model), depending on deployment
Menu path
Administration > Speech Analytics > AI Assistant > AI Tasks
In this view, tenant admins can usually:
- See tasks that are Enabled for the tenant
- Enable tasks from the Disabled tab
- Edit tasks to override Prompt, Filter, or both (where permitted)
- (Optionally) create tenant-specific tasks, if enabled in your deployment
What to check in an AI Task (before enabling)
When you open an AI Task, verify:
- Purpose – the task name/description matches the business outcome you want (e.g., “CSAT Scoring”)
- Outputs – which fields it writes to
- Output type – JSON is recommended for structured metrics
- Filters – whether it applies to the right subset of conversations (e.g., inbound calls only)
- Explanation behavior – whether it includes a reviewer-friendly explanation (recommended)
Mapping: output attributes → Custom Fields
The mapping table connects:
- ATTRIBUTE: the output key produced by the AI (e.g., csat, churn_risk)
- CUSTOM FIELD: where the value is stored for reporting/search/dashboards
One task may populate multiple fields, which is useful for: - grouping related outputs (e.g., “Churn Risk” + “Churn Risk Reason”) - producing both value and explanation (if explanations are stored in fields)
Filters: controlling eligibility
Tasks can include filters such as:
- channel (call vs chat vs ticket)
- direction (inbound vs outbound calls)
- duration threshold (e.g., > 15 seconds)
- other metadata-based filters (queues, tags, etc.)
Filters are important for: - relevance (avoid scoring irrelevant calls) - accuracy (avoid scoring too-short conversations) - cost control (reduce unnecessary processing)
Prompt: what you can change (tenant view)
Depending on your deployment, tenant admins may be able to:
- override the prompt to match your business definitions
- tighten output requirements (e.g., “JSON only”)
- update classification labels (dropdown values)
- enforce a scoring rubric
Best practice: keep prompts stable once you start tracking metrics over time. Prompt changes can shift results and affect trend analysis.
What tenant admins can and cannot change
Can override:
- Prompt (Task instructions and Task inputs)
- Filters (eligibility criteria)
Cannot override (managed by provider):
- Attribute mapping (output fields)
- Response schema
- AI engine selection
This ensures consistent data structure across tenants while allowing customization of analysis logic and filtering rules.
Figure: AI Task tenant view showing overridable settings.
