This feature compares repeated work videos of a target person (junior staff, newcomers, etc.) against the standard (finalized video + element definitions) established through STEP 5 (Selection → Extraction → Classification → Improvement → Verification) by viewing and aligning them with AI, and visualizes cycle time comparison, delay per element, and AI coaching.
This makes it instantly clear who is taking longer or showing variation in which elements, clarifying priorities for training and on-the-job training (OJT).
This article is STEP 6 of the Work Analysis series. For the full overview, see Work Analysis overview and getting started.
Preparation: finalize the standard first
Before running a comparison, you must complete STEP 1–5 (Selection through Verification) and finalize the standard video and element definitions. The comparison step cannot run if the standard has not been established.
Add a comparison (select target person and video)
- In the left sidebar, open "Compare" (STEP 6)
- Click [Add] in the top right
- In the dialog that appears, set the following:
- Target person: Select from team members (records whose work video this is)
- Work video: Select the target person's repeated work video from the asset library
- Click [Start] to begin AI video analysis (you will be notified by email when complete)
Comparison list and details
Added comparisons appear in the target person list. You can view the target person, video, analysis date, and status; clicking a row opens the detail view (comparison layout).
- While analyzing, "Analyzing" is shown; if partially failed, "Partially completed" appears; if failed, "Failed" is displayed
- Completed comparisons show no status display (only states requiring attention are highlighted)
Comparison layout (side-by-side standard and target person)
In the detail view, the standard video is on the left and the target person's video on the right, with playback position synchronized between both. In the comparison timeline at the bottom, the target person's delays are visualized with red shading.
Cycle time (stacked bar chart)
- Left edge = standard cycle time, followed by each cycle of the target person
- Each bar is displayed as the total of stacked elements = cycle time
- Standard (blue dashed line) and target person average (orange dashed line) reference lines are shown
- Clicking a cycle switches to a detailed comparison for that cycle
Delay per element
The difference when the standard is set to zero is displayed as a line chart for each element.
- Selected cycle: solid line (orange) + red shading
- Average: dotted line (light)
Variance per element (when 2 or more cycles)
For each element, you can compare the target person's shortest – longest range, average, selected cycle, and standard on a single screen. You can instantly see whether the work is consistent or shows high variation.
AI coaching (improvement suggestions)
From the [Improvement suggestion] tab on the right panel, AI can generate personalized coaching for this target person.
- Click [Generate suggestion] in the top right
- AI proposes element-level improvement points based on differences between the standard and the target person
- Each suggestion shows priority (high / medium / low), the target element, point, and advice
If you are not satisfied, you can retry with [Regenerate suggestion].
Change target person / adjust cycle
- You can also change the target person later from the top of the detail view (if you need to correct whose data this is)
- From [Cycle adjustment] in the top right, you can fine-tune if AI's automatic segmentation has shifted the target person's cycle boundaries
- If you are not satisfied with the analysis results, you can retry with [Re-analyze]
Manually adjust target person measurements
In [Cycle adjustment], you can perform the following operations in addition to fine-tuning boundaries:
- Split elements / merge with next / return to not detected
- Add measurements to elements that were not detected (add them from the gray chips)
- Undo / redo
- Warnings for unusually long or short elements
Previously, when AI missed elements, there was nothing to do but leave that element blank. Now you can add measurements by specifying the interval while watching the video.
The "Match with the reference" bar shows which reference elements exist in that cycle and which are missing. This is information that cannot be shown on a time-axis Gantt chart, so check it together.
[Image: Target person cycle adjustment (adding measurements to undetected elements)]
Which standard was this comparison against?
Comparison results are displayed based on the standard that AI used for alignment. Even if you later add, remove, or reorder elements on the standard side, the meaning of the results already shown does not change.
If there is a discrepancy with the current standard, the detail view shows that fact and an entry point for re-analysis. When you want to review with the latest standard, re-analyze from there.
Handling failures and partial failures
- Failed: AI observation and error details are displayed. After checking the cause, run [Re-analyze]
- Partially completed: If there are failed segments, you can retry just that part with [Re-analyze failed chunks only] (this does not consume AI counts)
Tips for effective use
- Refine the standard before comparing: If the standard's element definitions are sloppy, the comparison results will also be sloppy. The more carefully you create the model in STEP 1–5, the higher the accuracy of the comparison.
- Prepare multiple cycle videos: Having 3–5 or more cycles of the target person's repeated video per analysis yields more accurate variation and average figures.
- Use it as OJT material: Concrete statements like "You are delayed by ◯ seconds in element ◯◯; performing it this way will speed it up" dramatically change the nature of conversations on the floor.
- Re-run comparisons periodically: As a measure of training effectiveness, re-running the same comparison monthly or quarterly makes improvement visible.