What a career centre can see in week one, and what it cannot

Week one of a deployment is when an institution is most likely to be sold a story. The screens are full, the numbers are new, and everything on the page looks like insight.

Most of it is not. Most of what a career centre sees in the first week is a measurement, not a finding, and the two lead to different decisions. A measurement tells you where you are standing. A finding tells you something has changed. Week one can give you the first one honestly. It cannot give you the second one at all.

Here is what each list actually contains.

What the first week genuinely gives you

All of it computed from what students did, with nobody on staff typing anything in.

You can see who has been measured and who has not. In week one that is usually the most useful screen on the site, because it is the only one that tells you where you really are rather than where the sample is.

You can see a readiness distribution for the cohort, and the same cut by department, across the eight capability dimensions.

The Career Readiness Score behind that distribution, which is out of 100, is a deterministic blend of nine measured components: skill assessment, experience and evidence, scenarios, career alignment, voice, academic record, written reflections, supporting documents, and engagement quality. Every one of them is computed in the backend rather than inferred by a model.

You can see a cohort heatmap and click a cell to get the names of the students behind the colour, because that heatmap is built from assessment responses and not from a model. You can see the gaps that are actually blocking readiness, set beside what employers are asking for, so a gap is priced against demand rather than against an internal ideal.

You can see students at risk and strong fit students as working lists rather than counts. You can see a placement pipeline whose early stages fill themselves from measured readiness. You can see where the course catalogue sits against what the market is hiring for, which is the one screen that points at something a Dean actually controls. And you can export the lot for a board pack or an accreditation file.

You can also see what employers are hiring for right now, by function, by sector and by market. That picture is captured continuously rather than sampled, so the history starts accumulating from your first day even though your first day has none of it to show yet.

That is a real map, and it is built from student behaviour rather than staff estimates. It is worth having in week one.

What it cannot tell you, and why each limit is there

It cannot tell you whether anything is getting better or worse. There is no history on day one. The trend line starts the day you start. Last year's cohort is not in the system and never will be, so there is nothing to compare this one to. Any product showing you movement in week one is inventing it.

It cannot show you a score that has moved, because nothing has moved yet. The score only rises on learning that has been verified by passing a check at the end of the week. Reading alone does not move it. The uplift from learning is capped, and it cannot carry a student across the top threshold from below, because crossing that line requires a fresh measurement rather than more activity. So the first read is a diagnostic. A delta needs a second measurement and the time in between.

It cannot tell you who got placed. The early pipeline stages derive from readiness automatically. Interviewing and placed do not, because those are events happening in rooms the system cannot see. Your staff move those two by hand, with the full move history kept and one click back. In week one those columns are empty, and that is correct rather than broken.

It cannot tell you anything about a student who has not been measured. An unassessed student sits in the lowest tier next to a student who was measured and scored badly. Those are two different situations and only one of them is a finding. The same logic runs through the score itself: a part of the assessment a student skipped counts as missing rather than being averaged away, so skipping it genuinely lowers the number. That is a design decision and not an oversight. It is named on the student's own report with a link to complete it, and completing it recalculates.

It cannot be confident about a thin attempt, and it says so. A confidence modifier scales the result so a rushed attempt does not produce a confidently wrong number. Week one, when a whole cohort is being pushed through at once, is exactly when attempts are thinnest.

It cannot tell you the departmental cuts are right until you have configured it. The departments list, cohort assignment, grading scale and programme length are institution settings, and the last of those drives how often students are reassessed. Until they are set, every departmental number is a number about the wrong groups.

It cannot tell you the curriculum caused the gap. Course drift puts your catalogue next to market demand. That is a comparison, not a cause, and it should never be presented as one.

It cannot tell you whether your intervention worked. That is the whole purpose of the system, and it is the one thing the first week structurally cannot give you. It needs a measurement, then an action, then a second measurement.

Why the second list is the reason to trust the first

A system that could show you a trend on day one would be inventing it. A system that could tell you who got placed without anyone telling it would be guessing. A score that could only ever go up would not be a measurement.

Three properties follow from taking the limits seriously, and they are the ones worth asking any vendor about.

The score is deterministic and reproducible. The same responses produce the same number, every time, and two people looking at the same student see the same thing.

The score can go down. If a student is reassessed and has not maintained what they had, the number reflects it. A metric that only moves in one direction is a marketing number.

The score explains its own movement. Every change is logged with a plain language reason derived from what actually changed in the underlying components, not written by a model after the fact.

The practical version

If you are evaluating anything in this category, the first week is the wrong time to judge it and the right time to set it up. Configure the departments and the cohorts properly, get the assessment coverage up, and treat the first read as your baseline rather than your result.

Then decide what you are going to change, change it, and measure again. The second number is where the value is. Everything before that is preparation, and a vendor who tells you otherwise is selling you week one as though it were week twenty.

If you would like to see this against your own cohort, write to sales@skilldrift.ai.