Build a training dashboard around the decisions its readers need to make. Show participation, learning, application and relevant results with clear definitions, coverage and dates. Keep each measure at its proper level and make the supporting evidence accessible to authorized reviewers. A dashboard should help someone choose a next action; a set of favorable numbers does not by itself prove that training worked.
This lesson is for learning and program teams preparing a recurring review. You will sketch a dashboard, define its cards and write one action note. The number of cards depends on the work, not a rule that every program needs exactly four metrics.
- Name the audience and recurring decisions.
- Select a small set of measures with different jobs.
- Define each measure’s population, timing and calculation.
- Show coverage and missing information beside results.
- Connect detail and sources without exposing unnecessary personal data.
- Review the view, record actions and preserve issued snapshots.
Who is the dashboard for?
A delivery coordinator may need overdue tasks and access problems. A program lead may need learning patterns and areas for revision. A leadership team may need organizational results and unresolved evidence gaps. These views can share governed evidence while presenting different detail.
Write one sentence for each reader: “At the monthly review, the program lead decides which delivery issue to investigate.” If a card does not help a defined decision or explain its context, consider moving it to the detail view.
Which measures belong together?
Keep delivery and outcome measures distinct. Completion can help identify delivery coverage or a required task; it simply should not stand in for learning. A satisfaction score can inform experience without establishing skill use. Choose appropriate evidence for each question.
The Kirkpatrick framework distinguishes reaction, learning, behavior and results. Those categories can help organize evaluation evidence, but they are not a requirement to display exactly one number in each category.
| Dashboard question | Possible measure | Context shown alongside |
|---|---|---|
| Who received the intended delivery? | Attendance or completion | Eligible population, due status and delivery definition. |
| What learning evidence do we have? | Assessment criterion or matched change | Instrument, matched population and missing observations. |
| Is the skill being used? | Reviewed application evidence | Opportunity, source, follow-up period and coverage. |
| What result needs attention? | Relevant team or program outcome | Baseline, unit and other possible influences. |
| What should happen next? | Agreed review action | Owner, due date and delivery status. |
What does a useful card contain?
Use a plain title, value, denominator or unit, period, coverage note and route to the definition or evidence. State whether the value is current, incomplete or frozen for an issued report. Avoid a large percentage with its reporting base hidden in a tooltip.
In a fictional cohort of 40 learners, 30 have a comparable pair of assessments and 21 of those meet the defined improvement criterion. A card can say “21 of 30 matched learners improved; 30 of 40 have a usable pair.” It should not say “70% of learners improved” without explaining that matched base.
Another card might show 14 of 20 follow-up respondents reporting skill use. That is a different population and method. Do not imply that those 14 are necessarily among the 21 who improved unless the authorized joined records establish the overlap.
How do you keep different records from becoming duplicate results?
Keep learner, enrollment, assessment attempt and organizational results distinct. Aggregate repeated events under a documented rule before building a person-level card. A team productivity result belongs to a team and period, even when several trained people belong to that team.
For multi-site delivery, permit different local collection forms and agree the small common measures used in the dashboard. Record the mappings and limitations in the data dictionary. Leave incompatible measures separate rather than forcing a total that has no consistent meaning.
How should missing data appear?
Distinguish not yet due, missing, invalid and not applicable where useful. A follow-up that is not yet due should not appear as a zero application rate. A low response rate should remain visible beside the reported result.
In the fictional cohort, suppose 25 learners are due for follow-up and 20 respond. Show 20 of 25 coverage. The remaining 15 learners in the 40-person cohort are not yet due; they do not belong in that due-follow-up denominator. Retain the rule so later refreshes can be interpreted.
When does a trend deserve interpretation?
Check whether the measure, population, timing and coverage remain comparable. Several points make a sequence, but they do not automatically establish a meaningful trend. A different cohort mix or changed assessment can explain a movement.
Annotate a definition change or a new delivery route. If source evidence supports a valid restatement, retain original and restated values with the rule. Otherwise mark a break. Do not silently redraw previous cohorts using the newest definition.
How does the dashboard lead to action?
Pair a finding with a specific review question and owner. “Coverage fell in the evening group; the coordinator will check invitation delivery and access routes” is actionable without assuming why people did not respond. Record what was learned at the next review.
Keep an issued board or funder report separate from the changing live view. A later response can legitimately change the current total. Preserve the data cutoff and approved snapshot so the team can explain why last month’s report differs.
Where does Sopact fit?
You can prototype the cards with a small table and manual calculations. Sopact’s connected workflow can keep recurring surveys, documents and qualitative evidence available with the relevant history and definitions, reducing repeated assembly for each cohort.
Verify refresh behavior, calculations, filters and permissions in the actual configuration. Check a card back to its source records and review rule. AI can help summarize findings; it does not verify a formula merely by presenting it confidently or make a dashboard a causal evaluation.
Practice: build three cards and one action
Use the fictional figures above to sketch a matched-learning card, a due-follow-up coverage card and a reported-application card. Write each denominator and limitation. Add an action note and a source cutoff. Ask a colleague which conclusions the view supports and which it does not.
Frequently asked questions
Should completion be removed?
No, if it informs a delivery or operational decision. Label it accurately and keep it separate from learning and outcomes. The problem is using completion as proof of effect, not displaying a useful participation measure.
Does every dashboard need a 90-day application rate?
No. Choose follow-up timing around the opportunity to use the skill and the decision. Show the due population and evidence source. A fixed date is useful only when it fits the program.
Can one dashboard serve everyone?
It can share common definitions, but readers may need different views and access. A leadership summary should not expose individual learning histories merely because a coordinator needs them for support.
What should we save before sharing a report?
Retain the approved output, data cutoff, definitions, filters and calculation rules needed to reproduce it. A live dashboard may update later; the issued report needs an explainable historical basis.