## Dashboard: Realtime Adoption & Quality Trends ### Overview This image displays a dashboard titled "Realtime adoption & quality trends," presenting a comprehensive overview of agent performance, volume, quality, and operational metrics. The dashboard is organized into three main sections: "Volume & Adoption," "Quality & HITL," and "Performance & Operations," each containing multiple charts. A header section provides general information and filter controls for the data displayed. ### Components/Axes **Header Section (Top-Left to Top-Right):** * **Title:** "Realtime adoption & quality trends" * **Refresh Icon:** A circular arrow icon is present next to the title. * **Notification:** "Edwardgem, you have **3** work items awaiting your attention" with a right-pointing arrow. * **Filter Controls (Left to Right):** * **Start date:** Input field with value "08/01/2025" and a calendar icon. * **End date:** Input field with value "12/03/2025" and a calendar icon. * **Agent types:** Dropdown with value "5 selected". * **User:** Dropdown with value "All users". * **Granularity:** Dropdown with value "Weekly". **Section: Volume & Adoption (Top-Left Panel Group)** 1. **Chart Title:** "Total agents over time" (Line Chart) * **Y-axis Label:** Implicitly "Total agents" (based on title). * **Y-axis Scale:** 0 to 8000, with major grid lines and markers at 2000, 4000, 6000, 8000. * **X-axis Label:** Implicitly "Time". * **X-axis Markers:** 2025-08-11, 2025-09-15, 2025-10-20, 2025-12-01. * **Legend:** None (single data series). 2. **Chart Title:** "Volume by agent type" (Stacked Bar Chart) * **Y-axis Label:** Implicitly "Percentage". * **Y-axis Scale:** 0% to 100%, with major grid lines and markers at 25%, 50%, 75%, 100%. * **X-axis Label:** Implicitly "Time". * **X-axis Markers:** 2025-08-11, 2025-09-15, 2025-10-20, 2025-12-01. * **Legend (Bottom-Left):** * Light Blue: "Customer-Support" * Dark Blue: "Group-Email" * Yellow: "Invoice-Payment" * Light Grey: "Newsletter" * Pink: "Research" 3. **Chart Title:** "Top agent types" (Horizontal Bar Chart) * **Y-axis Label:** Implicitly "Agent types". * **Y-axis Categories:** Invoice-Payment, Customer-Support, Group-Email, Research, Newsletter (from top to bottom). * **X-axis Label:** Implicitly "Count". * **X-axis Scale:** 0 to 40000, with major grid lines and markers at 10000, 20000, 30000, 40000. * **Legend:** None (single color bars). **Section: Quality & HITL (Middle-Left Panel Group)** 1. **Chart Title:** "Finished vs aborted" (Area Chart) * **Y-axis Label:** Implicitly "Count". * **Y-axis Scale:** 0 to 8000, with major grid lines and markers at 2000, 4000, 6000, 8000. * **X-axis Label:** Implicitly "Time". * **X-axis Markers:** 2025-08-11, 2025-09-15, 2025-10-20, 2025-12-01. * **Legend (Bottom-Left):** * Green: "finished" * Pink: "aborted" 2. **Chart Title:** "Error distribution" (Donut Chart) * **Axes:** Not applicable. * **Legend (Bottom-Center):** * Light Blue: "none" * Dark Blue: "validation_error" * Yellow: "timeout" * Light Grey: "user_cancelled" * Pink: "system_error" 3. **Chart Title:** "HITL rate (%)" (Line Chart) * **Y-axis Label:** Implicitly "Percentage". * **Y-axis Scale:** 0% to 80%, with major grid lines and markers at 20%, 40%, 60%, 80%. * **X-axis Label:** Implicitly "Time". * **X-axis Markers:** 2025-08-11, 2025-09-15, 2025-10-20, 2025-12-01. * **Legend:** None (single data series). **Section: Performance & Operations (Bottom-Left Panel Group)** 1. **Chart Title:** "Average duration by agent type" (Vertical Bar Chart) * **Y-axis Label:** Implicitly "Duration". * **Y-axis Scale:** 0m to 333m, with major grid lines and markers at 0m, 83m, 167m, 250m, 333m. * **X-axis Label:** Implicitly "Agent type". * **X-axis Categories:** Customer-Support, Invoice-Payment, Research (from left to right). * **Legend:** None (single color bars). 2. **Chart Title:** "Queue wait trend" (Line Chart) * **Y-axis Label:** Implicitly "Time". * **Y-axis Scale:** 0s to 60s, with major grid lines and markers at 0s, 15s, 30s, 45s, 60s. * **X-axis Label:** Implicitly "Time". * **X-axis Markers:** 2025-08-11, 2025-09-15, 2025-10-20, 2025-12-01. * **Legend:** None (single data series). 3. **Chart Title:** "Concurrency heatmap" (Scatter Plot / Density Plot) * **Y-axis Label:** Implicitly "Concurrency level" or "Count". * **Y-axis Scale:** 0 to 23, with major grid lines and markers at 6, 12, 18 (implied), 23. * **X-axis Label:** Implicitly "Time". * **X-axis Markers:** 2025-10-05, 2025-10-20, 2025-11-03, 2025-11-17, 2025-12-04. * **Legend:** None (single color points). ### Detailed Analysis **Header Filters:** * Data is filtered for a period from August 1, 2025, to December 3, 2025. * Data pertains to "5 selected" agent types and "All users". * The granularity of the data is "Weekly". **Volume & Adoption Section:** 1. **Total agents over time:** * **Trend:** The number of total agents starts low, around 500-1000 in early August 2025. It then rapidly increases, peaking at approximately 6500 agents around mid-September 2025. Following this, there's a slight dip to about 5500 agents, then another rise to around 6000 agents by late September/early October. From mid-October, the number of agents gradually declines, dropping sharply from around 5500 in mid-November to approximately 1500-2000 by early December 2025. * **Key Data Points (Approximate):** * 2025-08-11: ~700 agents * 2025-09-15: ~6500 agents (peak 1) * 2025-09-25: ~5500 agents (trough) * 2025-10-05: ~6000 agents (peak 2) * 2025-12-01: ~1800 agents 2. **Volume by agent type:** * **Trend:** This stacked bar chart shows the weekly percentage distribution of work volume across five agent types. * **Customer-Support (Light Blue):** Consistently represents a significant portion, starting around 45-50% in August, decreasing slightly to 40% by mid-September, then increasing to about 50-55% by late October, and remaining around 50% through December. * **Group-Email (Dark Blue):** Starts around 10-15% in August, decreases to about 5% by mid-September, then fluctuates between 5-10% for the rest of the period. * **Invoice-Payment (Yellow):** Starts around 20-25% in August, increases to about 30-35% by mid-September, then gradually decreases to around 15-20% by December. * **Newsletter (Light Grey):** Remains a small, relatively stable proportion, generally below 5%, throughout the period. * **Research (Pink):** Starts very small, less than 5%, then increases to about 10-15% by mid-September, and remains around 10-15% for the rest of the period, showing a slight increase towards December. * **Overall:** The relative contribution of Invoice-Payment decreases over time, while Customer-Support and Research generally maintain or slightly increase their proportions. 3. **Top agent types:** * **Trend:** This horizontal bar chart ranks agent types by their total volume. * **Key Data Points (Approximate):** * **Invoice-Payment:** ~38,000 units * **Customer-Support:** ~18,000 units * **Group-Email:** ~8,000 units * **Research:** ~7,000 units * **Newsletter:** ~2,000 units * **Observation:** Invoice-Payment has by far the highest volume, more than double that of Customer-Support, which is the second highest. Newsletter has the lowest volume. **Quality & HITL Section:** 1. **Finished vs aborted:** * **Trend:** Both "finished" (green area) and "aborted" (pink area) tasks follow a similar trend to "Total agents over time." They start low, rise significantly, peak, and then decline. The "finished" tasks consistently outnumber "aborted" tasks. * **Key Data Points (Approximate):** * **Early August:** Both finished and aborted are low, finished ~500, aborted ~100. * **Mid-September (Peak):** Finished peaks around 6000-6200, aborted peaks around 1000-1200. * **Late October:** Finished around 5000-5500, aborted around 800-1000. * **Early December:** Finished drops to ~1500, aborted drops to ~200-300. * **Observation:** The ratio of finished to aborted tasks appears relatively stable, with finished tasks being roughly 5-6 times more numerous than aborted tasks throughout the period. 2. **Error distribution:** * **Trend:** This donut chart shows the distribution of different error types. * **Key Data Points (Approximate):** * **none (Light Blue):** Dominant, approximately 95-97% of all cases. * **validation_error (Dark Blue):** Very small slice, perhaps 1-2%. * **timeout (Yellow):** Very small slice, perhaps 0.5-1%. * **user_cancelled (Light Grey):** Very small slice, perhaps 0.5-1%. * **system_error (Pink):** Very small slice, perhaps 0.5-1%. * **Observation:** The vast majority of operations complete without any recorded error, indicating high overall process stability. 3. **HITL rate (%):** * **Trend:** The Human-In-The-Loop (HITL) rate starts high, around 78-80% in early August 2025. It remains stable until mid-September, then drops sharply to about 60% by late September. Another significant drop occurs in early October, bringing the rate down to approximately 40%. From mid-October onwards, the HITL rate stabilizes, fluctuating slightly between 40% and 45% until early December. * **Key Data Points (Approximate):** * 2025-08-11 to 2025-09-15: ~78-80% * 2025-09-20: ~60% * 2025-10-05: ~40% * 2025-10-05 to 2025-12-01: ~40-45% **Performance & Operations Section:** 1. **Average duration by agent type:** * **Trend:** This vertical bar chart compares the average duration for three specific agent types. * **Key Data Points (Approximate):** * **Customer-Support:** ~170m (minutes) * **Invoice-Payment:** ~300m (minutes) * **Research:** ~190m (minutes) * **Observation:** Invoice-Payment tasks have the longest average duration, significantly higher than Customer-Support and Research tasks. 2. **Queue wait trend:** * **Trend:** The average queue wait time fluctuates but remains relatively stable over the period. It starts around 50s in early August, dips slightly to 45s, then rises to around 55s in mid-September. It then fluctuates between 48s and 55s, ending around 50s in early December. * **Key Data Points (Approximate):** * 2025-08-11: ~50s * 2025-09-15: ~55s * 2025-10-20: ~50s * 2025-12-01: ~50s 3. **Concurrency heatmap:** * **Trend:** This chart displays the distribution of concurrent items over time. The X-axis spans from early October to early December 2025, and the Y-axis represents concurrency levels from 0 to 23. The chart is densely populated with light purple dots, indicating frequent occurrences of concurrency. * **Observation:** There is a consistent presence of concurrent items across the entire observed time frame. The density of points appears to be highest in the range of approximately 6 to 18 concurrent items, suggesting that these are the most common concurrency levels. There are fewer instances at very low (0-5) or very high (19-23) concurrency levels. The pattern appears somewhat uniform across the weeks shown, without clear peaks or troughs in overall concurrency density. ### Key Observations * **Agent Volume Volatility:** The "Total agents over time" chart shows significant fluctuations, with a sharp increase and decrease, suggesting project phases or seasonal changes in agent deployment or activity. * **Dominant Agent Type:** "Invoice-Payment" consistently represents the highest volume of work and also has the longest average duration per task, indicating it's a critical and time-consuming agent type. * **High Quality, Declining HITL:** The "Error distribution" shows very few errors, indicating high quality. However, the "HITL rate (%)" shows a significant drop from ~80% to ~40% over the period, suggesting a shift towards more automation or less human intervention, which could be a positive efficiency gain if quality is maintained. * **Stable Queue Times:** Despite fluctuations in agent volume, the "Queue wait trend" remains relatively stable, suggesting effective queue management or sufficient agent capacity for the observed wait times. * **Consistent Concurrency:** The "Concurrency heatmap" indicates a steady level of parallel processing, primarily within a mid-range, throughout the latter part of the period. ### Interpretation The dashboard provides a snapshot of an operational system, likely involving automated agents or processes, over a four-month period in late 2025. The "Volume & Adoption" section suggests a project or initiative that scaled up rapidly in August-September, reaching peak agent deployment, and then gradually scaled down towards December. The "Finished vs aborted" chart mirrors this trend, indicating that the overall workload (and completion rate) followed the agent deployment. The consistent ratio of finished to aborted tasks suggests that the underlying process efficiency or success rate remained stable even with fluctuating volume. The "Volume by agent type" and "Top agent types" charts highlight "Invoice-Payment" as the most significant workload contributor, both in terms of total volume and average duration. This suggests that optimizing "Invoice-Payment" processes could yield the most substantial improvements in overall system performance. "Customer-Support" and "Research" are also notable contributors. The "Quality & HITL" section presents an interesting dynamic. The "Error distribution" shows an overwhelmingly low error rate, implying robust and reliable processes. However, the "HITL rate (%)" experienced a sharp decline from nearly 80% to around 40%. This could be interpreted as a successful automation effort, where tasks previously requiring human intervention are now handled autonomously. If the quality (low error rate) is maintained despite reduced human involvement, this represents a significant efficiency gain. Conversely, if the HITL rate drop was unintended, it might indicate a problem with human agents not engaging when needed. Given the low error rate, the former interpretation (successful automation) is more likely. The "Performance & Operations" section provides insights into operational efficiency. The "Average duration by agent type" reinforces the importance of "Invoice-Payment" tasks, as they take the longest on average. This further supports focusing optimization efforts on this agent type. The "Queue wait trend" shows a relatively stable average wait time, suggesting that the system's capacity to handle incoming work is generally consistent, even with the observed fluctuations in total agent activity. The "Concurrency heatmap" indicates a sustained level of parallel processing, primarily in the mid-range of concurrency, which is a healthy sign of continuous operation rather than sporadic bursts. In summary, the data suggests a system that experienced a significant ramp-up and ramp-down in agent activity, with "Invoice-Payment" being a key workload driver. The system demonstrates high quality with very few errors, and a notable shift towards reduced human intervention (lower HITL rate), likely due to increased automation. Operational metrics like queue wait times and concurrency appear stable, indicating a well-managed system despite changes in overall volume. Further investigation might focus on the reasons behind the agent volume fluctuations and the specific drivers for the HITL rate reduction.