Stealth melt is bigger than summer melt: what the data says about silent applicants
Stealth melt — applicants who quietly fade mid-cycle without formally declining — is larger than summer melt and mostly invisible. Here is what the data shows.
Everyone in enrollment management knows about summer melt. It is the easy number to measure, the most-studied pattern in admissions operations research, and the topic of a reasonable percentage of NACAC conference sessions every year. And yet, for most institutions, summer melt is the smaller of the two losses occurring in the admissions funnel.
The larger loss is stealth melt. It happens earlier. It is harder to see. It is harder to measure. And — partly for those reasons — it is where the more interesting intervention opportunities exist.
This post is an attempt to make stealth melt visible, to sketch what the research shows about it, and to connect it to the operational reality that makes it so hard to catch.
Defining stealth melt
Summer melt has a clean definition: the gap between a confirmed enrollment commitment and actual matriculation. It is binary (did the student show up or not), late (occurs between May and September), and visible (you know a deposited student by name).
Stealth melt is messier. I use the term to describe applicants who quietly disengage during the admissions cycle without ever formally withdrawing their application or explicitly declining an offer. They submit their application, maybe complete a step or two of the follow-up, and then fade. They do not email. They do not call. They do not formally withdraw. They simply stop responding.
Concretely, stealth melt includes students who:
- Apply and never complete the application
- Submit the application but never submit supporting documents
- Receive an admission decision and never respond to it
- Are admitted and open correspondence but never engage with admitted-student content
- Deposit but then disengage between deposit and the first tangible matriculation step (housing, orientation registration)
All of these are melting in slow motion. None of them show up in a melt rate that is measured only as “deposited and did not matriculate.” They evaporate earlier.
Why stealth melt is bigger than summer melt
No institutional research office I know of has published a definitive, cross-institution estimate of stealth melt because the definition is not standardized. But the component pieces are well-measured, and when you sum them, the picture is striking.
- Application completion rate: Across U.S. four-year institutions, somewhere between 20% and 40% of started applications are never completed. This is the earliest and largest form of stealth melt, and most admissions teams treat it as a top-of-funnel marketing metric rather than a melt problem.
- Admission-decision response rate: Research from NACAC and from multiple institutional studies suggests that between 10% and 25% of admitted students never formally respond to the admission offer — neither deposit nor decline. At selective institutions the number is smaller; at broad-access institutions it approaches 30%.
- Document-follow-through rate: Of students who reach the post-admission phase but have outstanding documents (FAFSA verification, transcripts, immunizations), a meaningful fraction — often 15% to 20% — never complete the documents and quietly exit the funnel.
- Deposit-to-first-tangible-step fade: Even before the traditional summer-melt window, a subset of deposited students never advance to housing, orientation, or registration. This cohort is sometimes counted within summer melt and sometimes escapes measurement entirely.
If you compose these categories honestly, the total number of students who quietly disengage from any given institution’s admissions funnel in a given cycle is routinely two to four times the size of traditional summer melt. For most institutions, stealth melt is the biggest single loss event in the entire enrollment cycle, and most institutions are not explicitly measuring it.
Why stealth melt is harder to catch
Summer melt is easy to catch operationally because you have a manageable list (deposited students) and a short window (May to August). A counselor with a thousand deposits can at least triage weekly.
Stealth melt is operationally harder in three ways:
The list is huge. Institutions receive anywhere from one thousand to thirty thousand applications per cycle. Trying to monitor all of them for engagement decay is a task no human team can perform manually. The cohort of students at risk of stealth melting is the entire applicant pool, until proven otherwise.
The signals are subtle. Summer melt signals are often concrete (no housing application, no orientation registration). Stealth melt signals are behavioral (declining email opens, longer gaps between portal logins, no event engagement). They are the kind of signal that a human cannot scan across thousands of records.
The window is long and continuous. Stealth melt can begin the day a student submits an application and continue for months. There is no single inflection point where you can say “now is when we catch it.” Monitoring has to be continuous.
These are exactly the characteristics that make stealth melt a good AI problem and a bad human-process problem. A model can scan thousands of applicants daily, read the behavioral signals across months, and flag engagement decay in real time. A counselor cannot.
What the research shows about interventions
Research on stealth-melt-style interventions is thinner than the research on summer melt, partly because the phenomenon is not as neatly measured. But what does exist is consistent.
Proactive low-friction outreach at early stealth signals. Castleman and Page, whose work on summer-melt interventions is the most cited in the field, have also studied lighter-weight interventions earlier in the cycle. Proactive text or email outreach triggered by specific engagement-decay signals (no portal activity for 14 days, no application progress for 30 days, unopened welcome sequence) consistently reduces the subsequent stealth-melt rate by three to eight percentage points.
Document-completion concierge. Institutions that offer a human or chat-based walkthrough of the document checklist at the point where a student appears to stall reduce document-related stealth melt substantially. The intervention is simply identifying the student, reaching out, and offering to walk through the outstanding items. First-generation students benefit disproportionately.
Family and guardian inclusion early. Stealth melt is often a family-side event: the family is stalling, questioning, or reconsidering, and the student has gone quiet because the decision has not been made at home. Early and proactive family engagement — especially through channels the family uses (text, not email) — reduces stealth melt.
Re-engagement content sequences. For students who have stopped opening email, sending the same email again in a different channel at a different time of day recovers a meaningful fraction. This is the kind of operational detail that most institutions do not bother with because it is tedious — and which AI-driven pipelines can handle at scale.
How engagement analysis surfaces stealth melt
The core insight is that stealth melt is the tail end of an engagement decline curve. By the time a student fully disengages, they have been drifting for weeks or months. The pattern is in the data. The institution simply could not read it in real time.
CeliaConnect’s Engagement analysis (see Engagement, Readiness, Yield) is the surface where this shows up. Engagement is scored daily on every student in the funnel, and the trajectory matters as much as the score. A student whose Engagement has dropped from 78 to 34 over the last three weeks is a stealth-melt candidate, even if the absolute score is still acceptable and even if the student has not formally declined anything.
The writeback lands in Slate as a declining-Engagement flag with the specific signal that drove the decline — “portal logins halved, email open rate dropped from 64% to 12%, last counselor reply was one word and sent at 2 AM.” A counselor looking at the student’s record sees the signal beside the student’s existing profile. The intervention is a proactive, personalized reach-out at the moment the decline becomes readable — which is often weeks before the student would have formally dropped out of the funnel.
Critically, this is the kind of monitoring that no counselor can do manually across a thousand-plus applicant territory. The AI reads the pattern. The human owns the conversation.
A note on measurement
If you want to start measuring stealth melt at your institution, a reasonable first approximation is:
Stealth melt (proxy) = (Applications started + admitted-students-no-response + deposited-to-no-matriculation) / (Total applications started)
Measured for a complete cycle. This will overstate the true stealth-melt number because some students in each category had legitimate reasons to withdraw. But the order of magnitude will be honest, and it will almost certainly surprise you.
From there, break it down by funnel stage. You will usually find that one or two stages contribute disproportionately. That is where the highest-leverage intervention lives.
The point
Summer melt gets the attention because it is concentrated in time, well-defined, and historically measured. Stealth melt is larger, more diffuse, and historically ignored. As institutions move from human-capacity-bound monitoring to AI-driven continuous monitoring, stealth melt becomes the more interesting opportunity — simply because the tools now exist to catch it.
CeliaConnect was built with both problems in mind. The Engagement analysis surfaces stealth-melt signals daily across every student in the funnel; the Yield analysis folds in deposit-stage signals for the traditional summer-melt window; the writebacks land in Slate where the counselor already works. The AI does the pattern-reading, continuously, for every student. The human does the relationship.
That division of labor is the one that actually produces fewer melted students.
Tagged
