What partner institutions actually changed.
Each of these was measured with matched comparison against a comparable cohort, not by comparing this year to last. Institutions are described generically where our agreements require it.
Three deployments, three problems.
Three dashboards, no agreement on definitions
Our first deliverable was a data dictionary signed by the registrar, the deans and the IR office. Only after that did the analytics start producing numbers anyone acted on. Persistence rose 4.1 points over two terms against matched comparison.
Students blocked by full sections, not by ability
Bottleneck analysis found four courses explaining a third of all delay, and capacity shortfall mattered more than failure rate. They added sections and moved one annual course to both terms. On-time completion in affected cohorts rose 7.8 points.
400 students per advisor and no way to triage
The queue replaced a weekly spreadsheet exercise. A time-and-motion study before and after found advisors spent 31% less time working out who to contact, and duplicate outreach fell from 14% to under 3%.
Effect sizes and confidence intervals for all three are documented in the corresponding efficacy reports, which each institution received in full.
Two deployments where we found no effect.
Publishing only successes would make this page useless. Both institutions gave permission to describe what happened, because they thought other institutions should know.
No measurable change in year one
A 3,400-student institution deployed the full platform and saw no persistence effect after two terms. The efficacy study found the cause: outreach closure was recorded for 41% of items, so most interventions were happening informally and could not be attributed. The dashboards were used; the workflow was not.
What changed: we now flag closure rates below 70% as a warning in month two rather than discovering it at the first efficacy study.
Effect present, but not from us
A college saw persistence rise 3.2 points in the year after deployment. Matched comparison attributed almost all of it to a fee-restructuring decision made independently in the same term. We reported that the platform’s contribution was not distinguishable from zero.
Why it matters: a vendor comparing before to after would have claimed that 3.2 points. Matched comparison is the difference between a number and a result.
Talk to an institution like yours.
We will introduce you to a partner institution of comparable size and structure, including one where the first year did not go smoothly.