What You Can Learn from This Article
Through this guide, you will understand:
How to pinpoint escalation triggers by reviewing chat transcripts
How to diagnose significant shifts in your chatbot's self-service success rate
How to refine your knowledge base once escalation drivers are identified
How to Analyze Chat Records
Below is a practical approach to reviewing chat logs and boosting your bot's independent resolution performance:
How to Obtain Chat Records
You will need to export the logs where customers were transferred to live agents. In the Instadesk admin panel, these records can be accessed via the escalation reporting section.
Pick a specific day or a chosen time window, then randomly sample 200–300 conversations for detailed assessment. During the initial two weeks, perform this review 2–3 times per week.
How to Label and Categorize Escalation Records
Tag each conversation and assign a reason for the handoff to human staff. This helps drive improvements in self-service rates.
Tagging is done along two axes: operational and business. The business axis can be tailored to your specific industry. See the breakdown below:
Operational axis: You can classify each chat as Direct Human Request, Human-Assisted Resolution, or Bot-Only Resolution. For Human-Assisted, further subdivide into Guidance Needed, Product Query, Unclear Intent, No Apparent Reason, or Invalid. For Bot-Only, note if the answer was Subpar, Unknown, or Misguided.
Business axis: For an e-commerce example, tag as pre-sale, during-sale, or post-sale, with optional finer granularity.
Why use the operational axis? It tells you what kind of fix is needed.
Direct Human Request means the user asked for an agent right away, giving the bot no chance — this isn't solvable by bot tuning.
For Human-Assisted cases, examine the guidance and product aspects to feed back to business teams.
For Bot-Only cases, you can take operational action. A Subpar answer can be rewritten; an Unknown can be added as a new FAQ; a Misguided response can be improved by adding more similar phrasing or adjusting match rules.
Why use the business axis? It reveals which types of issues tend to go to humans most often, helping business units adjust their processes accordingly.
How to Diagnose Large Fluctuations in Self-Service Rate
A 5% swing is normal, and patterns vary by cycle. A more reliable comparison is year-over-year or period-over-period across similar timeframes. For instance, in e-commerce, weekends, weekdays, and promotion days behave differently — avoid lumping them together.
When you see a notable drop or spike, export two sets of escalation logs (same sample size) from the dates before and after the change, then apply the same labeling and analysis method described above to uncover the root cause.
