Student Retention and Support

Define the continuation outcome before predicting it

Terms such as *dropout*, *withdrawal* and *non-continuation* can describe different events. A student may formally withdraw, stop participating without completing the process, defer study, transfer programme or return after an approved break. If these outcomes are combined, the model learns from cases that require different responses.

Marketways begins by defining the student population, the continuation event and the period in which it must occur. The institution must also distinguish a student who requires support from a student who has made an informed decision to leave. This definition determines which records become examples, when a prediction can be made and what action the result can justify.

The warning must arrive while support can still help

Information available before enrolment can provide an early indication of risk. However, such information often describes broad background differences rather than a current support need. Attendance, assessment, learning-platform activity and service interactions can reveal a clearer change after study begins. By then, the institution has less time to respond.

The useful prediction point therefore depends on the intervention window. An alert issued after a decisive assessment failure may classify the outcome accurately but arrive too late to change it. Marketways tests several prediction dates and compares the additional accuracy gained by waiting with the opportunity lost by delaying support.

Reconstruct the learner’s sequence of events

Universities and training providers record relevant signals in separate systems. The student record contains enrolment and attainment. The learning platform records participation. Finance, accommodation and support teams maintain other histories. A current snapshot can conceal the order in which the warning signs appeared.

Marketways creates an as-of-date record that uses only information available at each historical decision point. This prevents later events from leaking into an earlier prediction. Sequence and journey analysis can then show whether disengagement followed an assessment result, a timetable conflict, a payment issue or an unsuccessful attempt to obtain help. The chronology gives an adviser a defensible starting point; it does not establish the student’s motive.

A risk score does not identify the cause

A prediction estimates which students resemble earlier students who did not continue. It does not prove why a particular student may leave. For example, low attendance may reflect declining commitment, employment hours, caring responsibilities, illness, inaccessible teaching or a recording error. Sending every student the same attendance reminder ignores these differences.

Marketways combines predictive analysis with student research, adviser knowledge and operational evidence to identify plausible explanations. Where the institution wants to know whether changing a service or policy improves retention, a causal design must test that intervention separately. This distinction prevents a correlated signal from becoming an unsupported diagnosis.

Support capacity belongs inside the decision rule

Every alert consumes adviser time and may create concern for the student. A threshold that identifies nearly every possible case can overwhelm the support team, delay contact with students who need urgent help and produce repeated false alarms. A threshold that is too restrictive can miss students who could benefit.

Marketways sets the decision threshold against the cost of each error, the urgency of the need and the institution’s capacity to respond. The analysis can reserve immediate review for higher-risk or safety-critical cases, direct other students to accessible support and monitor less certain changes for further evidence. The model and the operating process must be designed together.

Match the response to the likely barrier

Retention improves only if the institution can offer support that addresses the learner’s situation. Academic difficulty may require tutoring or a revised study plan. Financial pressure may require a conversation about funding or payment options. A timetable conflict may require a scheduling decision. Wellbeing or safeguarding concerns require qualified staff and established protocols.

The system should therefore present the evidence behind an alert, the relevant support routes and any uncertainty that remains. The adviser can then confirm the situation with the student and agree an appropriate next step. Prediction prioritises attention; professional judgement and student participation determine the response.

Design the complete support pathway

An alert has no value if it waits in a dashboard. The institution must assign responsibility for reviewing it, define how and when the student will be contacted, record the response, arrange the agreed support and check whether the issue was resolved. Each hand-off creates a chance for delay or loss.

Process analysis can identify where alerts remain unreviewed, students receive conflicting messages or referrals fail to reach the responsible team. An approved AI workflow may assemble relevant records, draft a clear outreach message, create a task and monitor the follow-up date. However, the workflow should refer ambiguous, sensitive or high-consequence cases to qualified staff rather than infer a personal circumstance from indirect data.

Student voice can explain what behavioural data cannot

Administrative records show what the institution observed. They may not show whether the student understood an instruction, trusted the available support or believed that asking for help would create a disadvantage. Interviews, surveys and adviser conversations can reveal barriers that transaction data cannot represent.

This evidence also tests the design of the outreach itself. A message intended as support may feel accusatory if it labels the student as likely to fail. Marketways uses qualitative and quantitative evidence together so that the intervention responds to the learner’s actual experience instead of the institution’s assumption.

Protect opportunity, autonomy and privacy

Retention analytics should offer assistance. It should not restrict module choice, funding, progression or another opportunity merely because a model predicts risk. Students also need a meaningful way to understand, question and correct the information used about them.

The institution should define which data are necessary, who may access them, which decisions remain with qualified staff and when consent or an additional safeguard is required. It should test errors and intervention rates across relevant student groups. These controls reduce the risk that incomplete records or proxy variables concentrate unwanted scrutiny on particular learners.

Evaluate the intervention as well as the prediction

Model accuracy measures whether the system separates higher-risk and lower-risk cases. It does not show whether the resulting support changes a student’s decision or academic outcome. A Dutch randomised field experiment demonstrated this difference: its early-warning system accurately identified risk, but counsellor access to the prediction did not reduce dropout or improve performance.

Marketways therefore evaluates the complete intervention. Depending on the setting, the institution may use a randomised rollout, a phased implementation, a matched comparison or another defensible design. The evaluation asks whether contacted students obtained suitable support, whether the intervention changed continuation or attainment and whether any improvement exceeded the result expected without it.

Measure whether the system resolves the problem

Retention is the final outcome, but it arrives too late to manage the pathway on its own. Earlier measures should show whether alerts were reviewed in time, students could be reached, support was accepted, referrals were completed and the identified barrier was resolved. Progression, completion, attainment and student experience show whether the response produced a durable educational benefit.

Marketways also separates prediction performance from service performance. A model may remain statistically accurate while support queues grow or students stop responding to repeated messages. Conversely, advisers may improve outcomes even when the model contributes little. Separate measures show management which part of the system requires attention.

Revalidate the system as conditions change

Enrolment patterns, programme structures, assessment rules and student behaviour change. A model trained on earlier cohorts may lose accuracy after a curriculum change, a new attendance system or a shift in the student population. Intervention effects can also change as students and staff become familiar with the process.

Marketways monitors data coverage, calibration, false alerts, missed cases, subgroup outcomes and support capacity over time. Material changes trigger a review of both the model and the intervention. Retention analytics remains dependable only when the institution treats it as a managed service rather than a model installed once.

How Marketways designs retention and support analytics

Marketways connects Machine Learning and Predictive Analytics with Survey and Psychometric Analysis, causal evaluation and Process and Workflow Systems. We define the continuation decision, reconstruct the learner timeline, validate the risk model, design the support pathway and measure whether the intervention improves outcomes.

The engagement may support an institution that needs to diagnose withdrawal, improve an existing early-warning system or design a controlled AI-assisted workflow. The result is not a list of students described as risky. It is a governed operating process that helps the institution offer suitable support in time and determine whether that support works.

Related guides and evidence

• Learner Experience and Service Redesign
• Training Need and Intervention Evaluation
• Education AI and Assessment Risk
• Timetable, Room and Teaching Capacity
• Student Retention and Responsible AI Analytics: case study
• Risk Detection
• Customer Satisfaction and Experience Research