A metric name is not enough
Teams may define resolved requests differently: last message versus confirmed action. Agree definition, unit, start/end events and exclusions before dashboards. Underspecified metrics look precise while generating arguments. Shared decisions require attention to meaning as much as presentation. Specify whether the unit being counted is a conversation, a request, a customer or an individual task. One conversation can contain several requests, and several conversations can concern the same unresolved case. Without a shared unit, an apparent change in volume may simply reflect how the team opened or grouped records during the period.
Questions connected to actions
Useful dashboards answer actionable questions: backlog locations, clarification-heavy categories and waiting steps. Identify decision owners. Separate daily monitoring from trend analysis and allow drill-down. Teams must distinguish operational changes from definition or capture changes.
Natural-language questions need context
“How are we doing this week?” needs process, period and comparison context. Conversational interfaces should clarify ambiguity, use approved definitions and expose periods and limitations. Fluency does not validate numbers. Separate causal hypotheses from observations: simultaneous growth in requests and waiting does not establish why. A response to a business question should make its interpretation visible. Show the process, period and comparison used, and ask for clarification when more than one definition could reasonably apply. This gives the person requesting the analysis a chance to correct the question before an attractive chart or explanation makes the chosen interpretation appear definitive.
Make incomplete data visible
Missing dates, unknown categories and stale states belong in quality assessment. Define incomplete-case treatment and correction owners. Definition changes need effective dates and historical recalculation disclosure. Honest comparison matters more than filling every dashboard space with numbers. Treat missing data as information about the measurement process itself. If many tasks have no completion time, the first decision may be to repair capture rather than compare team performance. Keep the affected records identifiable so their owners can investigate the source. When historical figures are revised, communicate the reason; otherwise users may assume that a changed chart reflects new business activity rather than a corrected calculation or dataset.
One definition card per metric
Document question, formula, source, frequency, owner and limitations. Validate with decision makers and data-entry teams, then test against known cases before full dashboards or natural-language interfaces.
Questions for your team before defining a metric
Before writing a formula, a short conversation between whoever uses the number and whoever produces it often prevents weeks of later debate. These questions help structure it:
- What specific decision would change if this number rose or fell consistently?
- What unit is being counted: conversation, person, case or action, and why that one rather than another?
- Which event marks the start and which the end, and what happens to items that never close?
- Who is responsible for keeping the source current and for flagging when its structure changes?
- Which other metric could this be confused with, and how is that avoided in its name and description?
Take it to your team.
- Agree each metric’s meaning.
- Distinguish observation from causal hypotheses.
