Every capability, and what each one doesn’t do.

The model is fitted on your students, your programmes and your academic calendar. We do not ship a sector-average model, because our transfer study found one loses roughly 19 points of precision when moved between institutions.

Every refit is validated against a simple attendance-and-assessment baseline on held-out data. If the fitted model does not beat that baseline we tell you and do not deploy it.

On-time completion is modelled at programme level over a multi-year horizon; course risk is modelled per enrolment within a term. Separating them matters because the intervention differs: one is an advising conversation, the other is often a timetable problem.

Course risk feeds the gateway-course bottleneck analysis, which is where we have seen the largest measured effects.

Every model refit produces a fairness report: flag rate, precision and recall broken out by first-generation status, gender, programme, campus and entry route. It is part of the refit output, not an audit you have to commission.

We built this because our own first-generation model under-flagged first-generation students by 11 points and we did not notice for two terms. The reporting exists so that cannot happen quietly again.

Each term we assess the initiatives you nominate using matched comparison: students who received the intervention are matched to similar students who did not, on the covariates that predict the outcome. We report effect sizes with confidence intervals.

This is the part institutions find uncomfortable and valuable in equal measure. In our 2026 study across 220 initiatives, fewer than half showed a measurable effect.

Queues are ranked by predicted impact rather than raw risk: a student who is very likely to leave and very unlikely to respond ranks below one where an intervention plausibly changes the outcome. Each entry carries the contributing factors so the first call is informed.

Every item has an assignee, a channel and a due date. Unclaimed items escalate to a supervisor rather than expiring silently, which is the failure mode we saw most often before we added it.

Advising, faculty and student services write into the same record, scoped by role. Before this, the most common complaint we heard was a student contacted three times in one week by people who did not know about each other.

Notes are permissioned by category. A wellbeing note is not visible to a faculty member who only needs academic context.

Every outreach closes with a recorded action and observed outcome. This is the least glamorous feature in the platform and the one that determines whether next term’s efficacy study is worth anything.

Institutions that skip closure discipline get dashboards that look fine and efficacy studies that cannot support a conclusion. We are direct about that trade at implementation.

We build and maintain the connectors. Integration is our deliverable rather than a specification we hand to your IT team, which is the distinction that determines whether a project finishes.

There is a mandatory reconciliation phase in which your existing reporting is treated as the source of truth. Where our numbers disagree with yours, we fix ours or explain why yours is wrong — before anything is built on top.

Our first deliverable is usually not a model. It is a dictionary defining retention, persistence, completion, enrolled, active and withdrawn — agreed and signed by the registrar, the deans and the IR office.

Definitions are versioned. When one changes, every downstream report notes which version it used, so a year-on-year comparison cannot silently compare two different questions.

Default residency is India, in line with the Digital Personal Data Protection Act, 2023. Dedicated tenants and on-premises deployment are available where institutional policy requires it.

There is no cross-institution pooling. Your data is not combined with another institution’s for benchmarking, and it is not used to train models that serve anyone else.

Students see their own progress and next actions. Faculty see cohort analytics for their courses. Administrators see institution-wide KPIs. The permission model is enforced server-side rather than by hiding UI.

Credentials are issued by your institution, not by us. We never create an account for a student directly.

SAML and OIDC against your identity provider, so account lifecycle follows your existing joiner-mover-leaver process. SCIM provisioning is available on System engagements.

Role mapping is driven by attributes your IdP already asserts, which avoids a second directory that drifts out of date.