To spot participants who may need support mid-program, review current participation, comparable baseline measures and the person’s own account of what is happening. Use agreed rules to flag records for staff review, then confirm the situation, agree an action and check the result. A flag identifies a reason to follow up; it does not establish that someone will leave or explain why they are struggling.
By Sopact Academy · Updated September 12, 2026. Participant records and numerical examples below are fictional.
Turn a check-in into a support decision
Bring: the baseline questions from the intake lesson, a current attendance or participation record, and your team’s support procedures.
Leave with: a short check-in, written review rules and an action list that shows who will follow up and when. Use the fictional training example below, or adapt the method to a fellowship, coaching or service program.
A four-step mid-program review
- Collect a small set of signals linked to a decision.
- Write review rules and identify what they cannot establish.
- Check each flagged record against its source and context.
- Assign support, follow up and evaluate whether the process helps.
The linked exit-measurement reading has a separate purpose. This one has a different job: helping the team decide what to ask and do while the program is running. Avoid turning every check-in into a full evaluation survey.
What can an early-warning signal tell you?
A signal can identify something worth checking: missed sessions, an unresolved barrier, an explicit request for help or a change in a comparable measure. Its meaning depends on context. A schedule conflict, an inaccessible venue, an incorrect attendance entry and a decision to leave for a suitable job call for different responses.
The What Works Clearinghouse guide for secondary schools recommends monitoring student progress and intervening when attendance, behavior or academic problems appear. The guide rates this monitoring recommendation as supported by minimal evidence. Its scope is secondary schools; it does not validate a particular cutoff for adult training or a Sopact prediction model. Use it as a reason to build a review-and-support process, then test the indicators in your own setting.
Do not describe every departure as a preventable failure. Some people leave for reasons the program cannot resolve, and some outcomes are positive. Record the person’s explanation where available rather than treating an AI-generated theme as their confirmed reason.
Step 1 — Choose signals the team can act on
Start with participation, one relevant progress measure and a question about current support needs. These are useful categories, not a universal three-question limit. The information should be proportionate to the program and suitable for the people completing it.
- Participation: sessions attended out of sessions expected during a defined period. Distinguish cancelled sessions, agreed absences and missing records.
- Progress: repeat a baseline question or assessment when it measures something relevant at this stage. Keep the wording and scale comparable.
- Participant context: ask what is making participation difficult and whether the person would like a conversation. Allow a private route for sensitive matters.
Choose timing around an opportunity to respond. A check-in after a difficult module may be more useful than an arbitrary calendar midpoint. Explain who will review the answers and when; the form should not imply immediate monitoring if the team reviews it weekly.
A check-in design prompt
Use only the program plan and intake questions supplied below: [paste them]. Propose a short check-in for [program stage]. For each question, name the decision it supports, the baseline item it pairs with, the response format and the staff owner. Keep wording and scales comparable where change is being assessed. Identify missing information and sensitive questions needing review. Do not invent a validated instrument, a response-rate claim or a dropout threshold.
If a baseline is absent, the current answer can still identify a support request. Label the missing comparison rather than inventing a direction of change. Likewise, a small movement on a self-rating may reflect ordinary variation rather than a meaningful decline.
Step 2 — Write rules for review, not labels for people
Agree what puts a record on a staff review list, who checks it and what happens next. Use plain statuses such as routine follow-up, review needed and urgent staff review under the organization’s existing protocol. Keep insufficient data as a separate state; missing answers do not mean that everything is fine.
Rules can use explicit requests for help, repeated missed participation, unresolved previous actions or combinations of signals. Test proposed numerical cutoffs against historical records and staff experience before relying on them. A threshold that creates more alerts than the team can review may delay help rather than improve it.
| Evidence | Review state | Required check |
|---|---|---|
| Participant asks for help with an unresolved learning task | Review needed | Confirm the difficulty, preferred support and responsible staff member |
| Attendance record is incomplete and the check-in is missing | Insufficient data | Check records and use the agreed contact procedure; do not infer a reason |
| Several concerns persist after an agreed support action | Further staff review | Check whether support was delivered and whether a different response is appropriate |
| A response raises an immediate concern covered by an existing protocol | Protocol-based escalation | Follow the organization’s procedure and named responsibilities; do not wait for an AI score |
Document the rule version and the reason for each flag. Allow staff to override a suggested status with a recorded explanation. A rule applied consistently can still be unsuitable: check whether it disproportionately flags people because of inaccessible provision, work schedules, language or incomplete records.
Step 3 — Read the signals together and confirm the context
The original record matters more than the label. Before contacting someone, check the observation date, participant and enrollment match, completeness of the baseline, and whether the narrative refers to a current situation. Look for negation and resolved problems; “transport is no longer a problem” should not be treated as an unresolved transport barrier.
| Signal | Fictional record | What staff can conclude |
|---|---|---|
| Attendance | 96% of expected sessions recorded as attended | Participation is high in this period; that does not rule out a learning difficulty |
| Confidence | 4.3 at intake; 4.0 at check-in on the same 1–10 scale | A small recorded difference, not by itself a reliable prediction or proof of decline |
| Engagement self-rating | 2 out of 5 | Ask what the rating means to this person |
| Participant’s words | “I don’t get the blueprint symbols yet and it’s stressing me out.” | A specific learning concern to discuss, rather than assuming low motivation |
| Proposed action | Offer an instructor conversation about the module | Confirm the person wants it, record the owner and schedule a check-back |
The useful finding is the stated learning difficulty. High attendance should not dismiss it, and the 0.3-point confidence difference should not exaggerate it. One tutoring session may help, but the record does not establish that it will resolve the issue or prevent departure.
In a configured Sopact workflow, new responses can be connected to earlier observations and analyzed with agreed instructions. Require source references for suggested themes and separate extracted text from interpretation. Confirm integrations, notifications and access arrangements during implementation; an automated alert is not evidence that a staff member has read or acted on it.
Step 4 — Work from an action list and close the loop
A useful caseload list shows the evidence, current owner, next action, due date and status. Rank by the team’s agreed review priority, not by an unvalidated probability of dropout. Include records that lack sufficient data and overdue actions alongside newly flagged responses.
| Participant | Evidence to check | Next action | Follow-up record |
|---|---|---|---|
| A | Low recorded attendance; asks for help with subnetting | Coach confirms attendance accuracy and offers instructor support | Owner, contact attempt, agreed support and review date |
| B | Reports difficulty keeping up with reading | Discuss the material and available learning support | Participant preference and whether the support was received |
| C | Earlier transport concern has no recorded resolution | Check whether the situation remains current and what action occurred | Current barrier status; do not assume an old note is still accurate |
| D | High attendance but a specific module difficulty | Offer a focused instructor conversation | Check understanding afterward and retain the participant’s account |
Use distinct statuses for suggested, assigned, contacted, agreed, delivered and reviewed. Marking an action “closed” should explain what was resolved or why no further action is planned. If a participant declines support, record that accurately without converting it into a failure label.
Look for program-level patterns as well. Several accounts of the same scheduling barrier justify investigating the timetable, not automatically changing it. Check who is affected, possible alternatives and whether the proposed change creates new barriers for others.
Test false alarms, missed concerns and workload
Before expanding the workflow, compare flagged records with staff-reviewed examples. Include a sample of unflagged records so you can find concerns the rules missed. Check whether the proposed reasons are supported by the source, whether the suggested priority is useful and whether the team can respond within the stated time.
Fictional pilot: a rule flags 12 of 40 records. Staff review finds that eight need a support conversation, two refer to resolved issues and two contain incorrect attendance data. In five sampled unflagged records, one explicit request for help was missed. This exposes specific improvements; it does not establish the rule’s overall accuracy or recall because most unflagged records were not reviewed.
Track the number of alerts, time to review, actions actually delivered and the person’s subsequent account. A lower flag count could reflect resolved barriers, missing submissions or a changed rule. Explain which before claiming that the program improved.
Exercise: run a review meeting using five records
- Choose five fictional or appropriately de-identified records, including one with missing data and one with high attendance but a stated concern.
- Apply the written rules independently with a colleague.
- Compare the flags and source evidence; document disagreements.
- Write a next action, owner and review date for each record that needs follow-up.
- Identify one rule or collection question to improve before a larger pilot.
Ready to continue? The team can explain why a record was flagged, distinguish a suggestion from an agreed action and show whether support happened. Keep that history for the exit lesson; do not treat the support list itself as evidence of program impact.
Watch the training-record workflow
4 minutes 13 seconds · Sopact demonstration using synthetic training records.
▶ Play: Training Program Data
Frequently asked questions
What is an at-risk participant in this lesson?
It means a participant whose current information gives the team a reason to review support needs. It is a working review category, not a diagnosis or a prediction that the person will leave. Describe the evidence and uncertainty rather than attaching a permanent label.
Is attendance enough to identify support needs?
Attendance is useful, but its meaning depends on the program and the accuracy of the record. High attendance does not rule out a learning concern, while low attendance may reflect an agreed absence or a recording error. Combine it with appropriate context and a conversation.
How large a confidence drop should trigger a flag?
There is no universal cutoff supplied by this lesson. Consider the instrument, normal variation, program stage and other evidence. A self-rating change may justify a question without justifying an urgent label. Test any proposed threshold before relying on it.
What should happen when a participant does not answer?
Keep nonresponse distinct from a positive response. Check whether the invitation reached them, whether another completion route is needed and what the agreed contact procedure allows. Do not infer the person’s motive or support needs from silence alone.
Can AI automatically decide the intervention?
AI can help organize records and suggest source-linked themes in a configured workflow. Staff confirm the situation, consider available support and agree the response with the participant. Review uncertain classifications and follow established procedures for urgent concerns.
How do we know whether the early-warning process helps?
Check source accuracy, missed concerns, unnecessary alerts, response time and whether support was delivered. Then evaluate relevant participant outcomes with an appropriate design. Neither a higher alert count nor a lower dropout count alone establishes that the workflow caused an improvement.
Continue from support decisions to outcome measurement
Related practice: measure change at exit. Carry forward the baseline, dated check-ins and record of support. For the wider workflow, explore Case Management.