VIPThink CRM Recording Analysis

Problem

Sales and support teams generate many call recordings daily, but reviewing every call, filtering valid connections, judging customer intent, and scoring talk scripts still relies heavily on manual work. Recordings are often buried in CRM student details, so sampling coverage is low and quality-review standards can drift.

How It Works

This skill targets VIPThink CRM recording workflows. It first retrieves call recordings through browser automation, then runs speech transcription and AI quality analysis.

• CRM retrieval: locate the target team and filter, open the student detail, switch to the phone tab, and select valid calls where isValid === '是' and duration is above 30 seconds.
• ASR transcription: use paraformer-v2 for Mandarin and Cantonese, with speaker diarization and disfluency removal.
• AI quality review: use qwen-plus to produce structured output covering script score, customer intent, issues, and improvement suggestions.
• Batch reporting: process multiple recordings in batches and export Markdown or Excel reports for team review.

Before execution, it asks for team scope, filter type, single or batch mode, and output format, which helps prevent accidental large analysis runs.

Limitations

This is better suited for recording sampling, talk-script review, and call quality checks than automatic CRM data editing. Watch for CRM selector drift, expiring OSS recording URLs, and API cost during batch analysis. Cantonese recognition needs yue in language_hints. Confirmation checkpoints must not be skipped automatically.