Decision Tracking
Decision Tracking Overview
Decision tracking reviews key judgments, metric changes, and action results after an agent's analysis. It turns a one-time conclusion into a recurring review task to check whether risks improve, actions work, and metrics return to target.
Features
In operations analysis, customer management, or sales reviews, an agent may identify an issue worth tracking, such as whether representatives with low visit success rates recover above the team baseline. Create a task to review it regularly.
A tracking task typically contains:
- The business question to review.
- Metrics, targets, and comparison rules.
- Review frequency and next review time.
- Source conversation, evidence chain, or report context.
- Review history with results, statuses, and reports.
- Evidence supporting creation and subsequent judgments.
Use Cases
- Observe whether a risk improves.
- Check action effectiveness regularly.
- Review business issues using fixed metric definitions.
- Preserve review history for retrospective analysis and audits.
- Extend an agent's conclusion into ongoing operational follow-up.
Create a Tracking Task
Create a task from a response or evidence view when a conclusion warrants follow-up, such as whether key-product sales recover, customer churn risk falls, or replenishment reduces inventory risk.
The system first drafts a task from the turn's evidence chain. The task is created only after confirmation. The draft includes the goal, baseline window, core metrics, improvement rules, recommended actions, and review frequency.
Before confirming, check that the question is clear, the baseline period is appropriate, metrics represent the goal, improvement rules match expectations, and the review frequency suits the issue.
Not every analysis suits tracking. A task needs a verifiable metric baseline, clear improvement rules, and value in continued observation.
Adjust the name, recommended actions, improvement rules, or frequency as needed. Saving preserves the original evidence snapshot and baseline for future comparison.
Do not change baseline definitions casually. An unclear baseline makes it difficult to distinguish genuine improvement from changes in definitions.
View the Task List
Open Decision Tracking. The page shows the task count and filters for all tasks or tasks due for review. If none exist, confirm a tracking candidate in an agent's final response or evidence view first.
| Column | Description |
|---|---|
| Task | Status, name, baseline description, and issue summary. |
| Review criteria | Key metrics, review question, current conclusion, and latest summary. |
| Schedule | Frequency and next review time. |
| Source | Return to the originating conversation and analysis context. |
Task status indicates progress. For example, an in-progress task continues review on its configured schedule.
View Task Details
Open a task to inspect its definition, reports, metric changes, evidence, and history.
| Area | Description |
|---|---|
| Left summary | Status, name, baseline, question, next review time, and evidence entry. |
| Review history | Initial report and subsequent reviews for period comparisons. |
| Task sidebar | Business objects, review and baseline periods, frequency, comparison definitions, and latest business date. |
| Review question | The recurring question. |
| Current conclusion | Judgment such as partially improved, not improved, or recovered. |
| Review report | Overview, metric cards, rule table, and recommended-action tracking. |
| Execution information | Status of the current review execution. |
Run a Manual Review
- Start a manual review in the history area to select a review window.
- Confirm the start and end business dates.
- Check the prompts to avoid using the baseline period as a new observation period, then start.
- Review execution status and the report in history. Available data may adjust the actual window, so check both requested and actual periods.
Manual reviews are useful after data refreshes, when automatic reviews fail or are delayed, after business actions, or before meetings requiring current results.
Completion adds a history record and updates the current conclusion and report.
Change the Next Review Time
In the summary, edit the next review time to adjust the schedule. Check data refresh frequency first. For weekly reviews, choose a time after a full business week to avoid incomplete-data judgments.
Understand Review Reports
Reports explain the basis of the judgment through an overview, metric cards, and rules.
Overview
The overview presents the core conclusion. For example, in the latest complete review week, visit success remains below baseline but visits within the standard distance are above it, resulting in partial improvement.
It normally includes the period, current key metrics, comparison with the creation baseline or previous review, conclusion, and explanation.
Metric Cards
Cards show current values, baseline differences, and business meaning. Sales-visit examples include visit success, completion, and within-standard-distance rates; target representative visit counts; and same-week team success and distance rates. Other tasks use metrics defined by their goals and criteria.
Use the cards to identify whether the issue improves and which metrics contribute.
Baselines and Improvement Rules
The rule table shows each creation baseline, current result, difference, judgment, and improvement rule. Check whether metrics recover above baseline, improve or worsen, whether judgments match business expectations, and whether rules still apply to the period.
Review Reports and Choose Next Actions
Open the latest report to compare current results against the original baseline, rules, actions, and review window. Check window validity, metric changes, and whether the conclusion indicates improvement, partial improvement, no improvement, or data not ready.
Unrefreshed data, invalid windows, or permission errors do not mean the business has failed to improve. Check data refresh, timing, connections, and permissions first.
Use reports to decide whether to keep observing, adjust actions, close the task, or run a manual review.
View Evidence
Use the evidence entry on the left of task details to trace creation and review judgments. Evidence usually includes the original candidate conclusion, baseline data and metrics, current review data and metrics, report context, and business rules.
If a result conflicts with business understanding, verify definitions, periods, and rules in the evidence first.
Delete a Tracking Task
Delete from the upper-left area of task details. Reviews stop after deletion.
Delete when the issue is closed, the creation criteria were incorrect and require a new task, or the task no longer fits the business goal. First determine whether history or reports must be retained.
Recommendations
- Track a clear question, not multiple unrelated issues.
- Use stable metrics such as success/completion rates, target-achievement counts, or anomaly counts.
- Match review frequency to data refresh cycles.
- Choose a representative baseline period.
- Read the overview, metric cards, and rule table together.
- Inspect evidence before changing a task or ontology after an unexpected result.