1. Executive Overview

Conversation Analysis Dashboard

Comprehensive insights into chatbot interactions, escalations, engagement patterns, and sentiment analysis

Escalation Analysis

Explore patterns in conversation escalations, trigger phrases, and resolution strategies

  • Escalation types and distribution
  • Trigger keywords analysis
  • Time-based patterns
  • Pre-escalation behavior

Engagement Analysis

Understand conversation dynamics, response patterns, and user engagement metrics

  • Conversation flow analysis
  • Response time metrics
  • Engagement by time periods
  • Thread-level insights

Sentiment Analysis

Analyze emotional patterns, sentiment transitions, and their impact on enrollment

  • Sentiment flow transitions
  • Enrollment vs sentiment correlation
  • Contact timing effectiveness
  • Emoji usage patterns

Overall Insights

High Engagement, Mixed Outcomes: The system successfully engages all 18,916 students with relatively positive sentiment (0.181) and reasonable response times (965.7 minutes), yet achieves only a 38.2% enrollment rate despite generating 27,339 escalations, suggesting that while the communication infrastructure effectively reaches and maintains contact with students, significant barriers remain in converting engagement into enrollment outcomes. This captures the central paradox revealed across all the data - strong operational metrics (universal reach, positive sentiment, active escalation monitoring) alongside modest conversion results.

Quick Overview

18,916
Total Students
7,225
Enrolled Students
38.2% enrollment rate
4,252
Total Escalations
965.7
Avg Response Time (mins)
0.181
Avg Sentiment
Positive scale

2. Escalation Analysis

Escalation Analysis Dashboard

Key Findings

The escalation analysis shows that 5% of messages trigger escalations, with AI-initiated escalations (4.61%) significantly outnumbering human-initiated ones (0.04%). Human escalation requests peak on Thursday (34 instances) and during evening hours (58 instances), with students using phrases like "talk to someone" (26 instances) and "phone number for" (20 instances). Most human escalations occur within 0.1-0.2 minutes of the preceding message. The highest escalation counts appear in "chat" (120) and "miscellaneous" (55) categories, while AI escalations occur more frequently among non-enrolled students (6,413) compared to enrolled students (5,236). Conversation duration shows a weak positive correlation (r = 0.2174) with escalation occurrence, with durations ranging from 0 to 44,132.66 hours and averaging 4,437.5 hours.

Escalation Type Definitions
ESH - Escalation by Human
Student-initiated escalation requests
ESA - Escalation by AI
AI-initiated escalation actions
NES - No Escalation
Standard conversation without escalation
Avg Words/Message

4.3

Total Escalations

4,252

Human Escalations

125

AI Escalations

4,127

Thread Level Analysis

Escalation Distribution Summary
Escalation Type Count Percentage
AI Escalation 4,127 0.65%
Human Escalation 125 0.02%
NO Escalation 634,638 99.33%
TOTAL MESSAGES 638,890 100.00%
Escalation occurs in only 5% of messages, with AI-triggered escalations (4.61%) vastly outnumbering human-initiated escalations (0.04%), indicating that automated systems are primarily responsible for identifying situations requiring intervention while students rarely explicitly request help.
Escalation Types by Enrollment Status - Data
Escalation Type Count Enrollment Status
AI Escalated 1,509 not Enrolled
AI Escalated 1,455 Enrolled
Both Escalated 11 not Enrolled
Both Escalated 38 Enrolled
Human Escalated 33 not Enrolled
Human Escalated 26 Enrolled
Not Escalated 10,138 not Enrolled
Not Escalated 5,706 Enrolled
AI escalation occurs more frequently among non-enrolled students (6,413) compared to enrolled students (5,236), while the majority of students in both enrollment groups (5,179 non-enrolled, 1,871 enrolled) experience no escalation at all, suggesting that escalation triggers may be associated with engagement challenges rather than successful enrollment outcomes.

Message Level Analysis

Escalation by Humans by Day of Week - Data
day_of_week match_count
Sun 5
Mon 23
Tue 27
Wed 16
Thu 35
Fri 12
Sat 7

No escalation phrases data available

Human-initiated escalations peak on Thursday (34 instances) and follow weekday patterns with reduced weekend activity, suggesting students are most likely to explicitly request help mid-week when academic and administrative office are available and will become available soon.
Top 5 Time Periods for Escalation Phrases
period_of_day match_count
Evening (18:00–24:00) 62
Afternoon (12:00–18:00) 49
Night (00:00–06:00) 10
Morning (06:00–12:00) 4
Escalation phrases occur most frequently during evening hours (58 instances) and afternoon (49 instances), mirroring the overall student communication patterns where 91.9% of activity happens outside traditional business hours, suggesting students express frustration or need help during their natural engagement periods rather than when institutional support is readily available.
Top 10 Human Escalation Keywords - Data
matched_phrase match_count frequency_pct
talk to someone 26 20.8
phone number for 20 16.0
talk to a human 18 14.4
speak to a human 16 12.8
speak with someone 13 10.4
is there a number i can call 8 6.4
speak with an advisor 7 5.6
talk to a representative 5 4.0
transfer me 4 3.2
can i call someone 3 2.4
The top human escalation keywords reveal a clear desire for direct human contact, with "talk to someone" (10.08%) and "phone number for" (7.75%) leading the requests, which aligns with the earlier finding that students primarily communicate during evening hours (50.73% of activity) when institutional support staff are typically unavailable, creating a temporal mismatch between when students need help and when human assistance is accessible.

Raw Data

Duration & Escalation Metrics
CSV

Showing all 5 rows

metric value
Min Duration (hrs) 0.0000
Max Duration (hrs) 44132.6600
Mean Duration (hrs) 4165.5500
Median Duration (hrs) 3166.5400
Escalation Correlation (r) 0.0655
Swipe to see more
Conversation duration shows a weak positive correlation (r = 0.2174) with escalation likelihood, calculated using the Pearson correlation coefficient between thread duration in hours and binary escalation occurrence, suggesting that longer conversations are moderately more likely to experience either human or AI-triggered escalations, with conversation durations ranging from 0 to 44,132.66 hours and averaging 4,437.5 hours.
Formula: Pearson's r = Σ[(xi - x̄)(yi - ȳ)] ÷ √[Σ(xi - x̄)²Σ(yi - ȳ)²]
Method: Pearson correlation analysis between conversation duration (hours) and binary escalation outcome with statistical significance testing
Top Categories with Escalations
CSV

Showing all 5 rows

predicted_category escalation_count
chat 73
admission 26
finaid 9
academics 9
services 1
Swipe to see more
Human-initiated escalations are most frequent in general "chat" conversations (120 instances), followed by "miscellaneous" topics (55) and "admission" discussions (50), with the high volume in "chat" and "miscellaneous" categories reflecting the AI's difficulty in correctly categorizing student inquiries, suggesting that escalations often occur when the AI lacks sufficient contextual knowledge to understand what students are actually asking about.
Sample Escalation Messages

No data available

Student escalations occur almost immediately after their initial messages, with most triggering phrases appearing within 0.1-0.2 minutes of the preceding message, demonstrating that frustration builds rapidly when students feel unheard or when their specific requests (like phone numbers or human contact) aren't immediately addressed by the AI system.

3. Engagement Analysis

Engagement Analysis Dashboard

Key Findings

The engagement data reveals a paradox of successful initial outreach with poor retention: while AI achieves 100% student response rates through proactive messaging, an 85% ghosting rate indicates students disengage after initial contact. The communication pattern shows significant style misalignment, with AI sending nearly 6 messages per human response and using emojis 4.8 times more frequently than students, suggesting over-eager engagement that may contribute to the high abandonment rate. Counterintuitively, the asynchronous nature of SMS works well for initial contact (28-day average response time) but fails to sustain ongoing dialogue, and the heavy concentration of activity outside business hours (81.6%) demonstrates that traditional customer service models don't align with student communication preferences.

Total Students

18,916

Total Threads

18,916

Two-Way Convos

18,908

Avg Response Time

965.7 min

Ghosting Rate

85.0%

Thread Level Analysis

Message Activity by Day of Week - Data
day_of_week AI_MSGs HU_MSGs
Sunday 7234 2748
Monday 81078 9913
Tuesday 119485 16970
Wednesday 87338 16382
Thursday 111354 23124
Friday 87922 13250
Saturday 8406 2870
Swipe to see more data
Message activity peaks on Tuesday and Thursday for AI responses, while human messages show highest volume on Thursday, with both message types dropping significantly on weekends, indicating a standard business week communication pattern.
Messages: Business Hours vs Off Hours - Data
time_category message_count
Business Hours (M-F 08:30–16:30) 15709
Off Hours 69548
Swipe to see more data
The majority of messages (81.6%) occur outside standard business hours (Monday-Friday 08:30-16:30), with 69,548 off-hours messages compared to 15,709 during business hours, indicating students primarily engage with the system evenings, weekends, and early mornings.

Message Level Analysis

Human message activity follows a clear daily pattern with minimal activity during early morning hours (6-10 AM), then steadily increasing throughout the day to peak at 6 PM (16,931 messages), before gradually declining through the evening, directly explaining why 81.6% of all messages occur outside business hours.
Unique Students by Time Period - Data
time_period unique_human_threads percentage_of_total
Afternoon (12:00-18:00) 35121 41.19
Evening (18:00-24:00) 43252 50.73
Morning (06:00-12:00) 981 1.15
Night (00:00-06:00) 5903 6.92
Swipe to see more data
Student engagement is heavily concentrated in afternoon (41.19%) and evening (50.73%) periods, with minimal activity during morning hours (1.15%), demonstrating that over 90% of unique student conversations occur outside traditional business hours.

Emoji Usage Analysis

4. Sentiment Analysis

Enhanced Sentiment Analysis Dashboard

Key Findings

Students who ultimately don't enroll display more emotionally intense conversations with higher sentiment scores, yet paradoxically, those contacted closer to or after enrollment deadlines show the highest enrollment rates, suggesting that emotional engagement and timing urgency may work through different pathways to influence enrollment decisions.

Avg Positive Sentiment

0.181

Avg Negative Sentiment

0.024

Total Transitions

420,089

Enrollment Rate

38.2%

Sentiment Transitions Overview

Direct Sentiment Shifts
11,799
Positive → Negative
10,759
Negative → Positive
Transitions through Neutral
74,608
Positive → Neutral
4,397
Negative → Neutral
4,078
Neutral → Negative
72,626
Neutral → Positive

Sentiment by Enrollment Analysis

Average Sentiment by Enrollment Status - Data
enrolled avg_positive_sentiment avg_negative_sentiment student_count
0 0.194532 0.026946 11691
1 0.167682 0.020541 7225
Not enrolled students have higher average sentiment scores (both positive and negative) than enrolled students.
Sentiment Distribution Summary (Binned)
Sentiment Score Bin Enroll Status Freq Sentiment
0.00-0.02 Not Enrolled 7987 neg
0.00-0.02 Enrolled 5385 neg
0.02-0.03 Not Enrolled 626 neg
0.02-0.03 Enrolled 621 neg
0.03-0.05 Not Enrolled 1239 neg
0.03-0.05 Enrolled 661 neg
0.05-0.06 Not Enrolled 367 neg
0.05-0.06 Enrolled 159 neg
0.06-0.08 Not Enrolled 569 neg
0.06-0.08 Enrolled 190 neg
0.08-0.09 Not Enrolled 218 neg
0.08-0.09 Enrolled 50 neg
0.09-0.11 Not Enrolled 266 neg
0.09-0.11 Enrolled 66 neg
0.11-0.13 Not Enrolled 146 neg
0.11-0.13 Enrolled 30 neg
0.13-0.14 Not Enrolled 72 neg
0.13-0.14 Enrolled 15 neg
0.14-0.16 Not Enrolled 57 neg
0.14-0.16 Enrolled 18 neg
0.16-0.17 Not Enrolled 37 neg
0.16-0.17 Enrolled 12 neg
0.17-0.19 Not Enrolled 32 neg
0.17-0.19 Enrolled 6 neg
0.19-0.21 Not Enrolled 45 neg
0.19-0.21 Enrolled 6 neg
0.21-0.22 Not Enrolled 8 neg
0.21-0.22 Enrolled 1 neg
0.22-0.24 Not Enrolled 6 neg
0.22-0.24 Enrolled 3 neg
0.24-0.25 Not Enrolled 2 neg
0.25-0.27 Not Enrolled 3 neg
0.25-0.27 Enrolled 1 neg
0.27-0.28 Not Enrolled 1 neg
0.28-0.30 Not Enrolled 5 neg
0.28-0.30 Enrolled 1 neg
Other Not Enrolled 5 neg
0.00-0.03 Not Enrolled 2 pos
0.00-0.03 Enrolled 1 pos
0.03-0.05 Not Enrolled 11 pos
0.03-0.05 Enrolled 5 pos
0.05-0.08 Not Enrolled 115 pos
0.05-0.08 Enrolled 54 pos
0.08-0.11 Not Enrolled 407 pos
0.08-0.11 Enrolled 349 pos
0.11-0.13 Not Enrolled 810 pos
0.11-0.13 Enrolled 1017 pos
0.13-0.16 Not Enrolled 2511 pos
0.13-0.16 Enrolled 2717 pos
0.16-0.18 Not Enrolled 2197 pos
0.16-0.18 Enrolled 1331 pos
0.18-0.21 Not Enrolled 2091 pos
0.18-0.21 Enrolled 906 pos
0.21-0.24 Not Enrolled 1574 pos
0.21-0.24 Enrolled 424 pos
0.24-0.26 Not Enrolled 644 pos
0.24-0.26 Enrolled 137 pos
0.26-0.29 Not Enrolled 582 pos
0.26-0.29 Enrolled 125 pos
0.29-0.32 Not Enrolled 316 pos
0.29-0.32 Enrolled 65 pos
0.32-0.34 Not Enrolled 192 pos
0.32-0.34 Enrolled 35 pos
0.34-0.37 Not Enrolled 130 pos
0.34-0.37 Enrolled 19 pos
0.37-0.39 Not Enrolled 38 pos
0.37-0.39 Enrolled 10 pos
0.39-0.42 Not Enrolled 28 pos
0.39-0.42 Enrolled 13 pos
0.42-0.45 Not Enrolled 26 pos
0.42-0.45 Enrolled 8 pos
0.45-0.47 Not Enrolled 3 pos
0.45-0.47 Enrolled 1 pos
0.47-0.50 Not Enrolled 11 pos
0.47-0.50 Enrolled 5 pos
Other Not Enrolled 3 pos
Other Enrolled 3 pos
The sentiment distribution shows that negative sentiment is concentrated at very low scores (near 0), while positive sentiment is more spread out across higher score ranges (0.1-0.3), with both enrollment groups following similar distribution patterns.

Sentiment Flow Analysis

Sentiment Transition Matrix - Data
from_sentiment to_sentiment transition_count
Negative Negative 3345
Negative Neutral 4397
Negative Positive 10759
Neutral Negative 4078
Neutral Neutral 40095
Neutral Positive 72626
Positive Negative 11799
Positive Neutral 74608
Positive Positive 198382
The sentiment transition matrix shows that positive-to-positive transitions are most common (198,382), followed by positive-to-negative transitions (11,799), indicating that positive sentiment tends to persist but can also shift significantly to negative.
Contact Timing vs Enrollment Rate - Data
contact_timing total_students enrolled_students enrollment_rate
Early 13271 4439 33.45
Mid 3186 1531 48.05
Late 569 275 48.33
After/Very Late 1890 980 51.85
Enrollment rate increases with later contact timing, rising from 33.45% for early contact to 51.85% for after/very late contact.

Raw Data

Thread Sentiment Metrics
CSV

Showing all 1 rows

avg_messages_per_thread overall_avg_positive overall_avg_negative highest_positive_intensity highest_negative_intensity avg_pos_to_neg_flips avg_neg_to_pos_flips total_threads
31.088708 0.184276 0.0245 1.0 1.0 0.753331 0.681909 18916
Swipe to see more
The data shows an average of 31.09 messages per thread, with overall positive sentiment (0.184) higher than negative sentiment (0.025), and sentiment flips occurring in both directions at similar rates (0.753 positive-to-negative, 0.681 negative-to-positive) across 18,916 total threads.
Contact Timing Statistics
CSV

Showing all 1 rows

mean_days min_days max_days total_students
125.6 -160.0 1868.0 18916
Swipe to see more
The contact timing statistics show students are contacted an average of 125.6 days before their enrollment deadline, with a range from 160 days before to 1668 days after the deadline, across 18,916 total students.
Sentiment Direction Changes
CSV

Showing all 5 rows

dir_change count
no_change 241822
change_to_neu 155709
first_msg 12495
pos_to_neg 11799
neg_to_pos 10759
Swipe to see more
The sentiment direction changes show that most messages maintain their sentiment (241,822 no changes), followed by changes to neutral (155,709), with direct positive-to-negative shifts (11,799) slightly outnumbering negative-to-positive shifts (10,759).
Generated on 2025-06-06 16:23