What demand signals are and are not

What demand signals are and are not

Pathways surfaces two very different things that both get called “demand.” One is
labor-market data: what an occupation pays, whether it is growing, and how many
openings it produces each year. The other is enrollment demand: what prospective
students are actually looking for when they use Compass. They come from opposite
directions, they carry different guarantees, and confusing them leads to bad
decisions. This article explains what each one is, where it comes from, and, just as
importantly, what neither of them is. It does not cover how to enable tracking or how
to refresh the labor data; those are how-to tasks.

Two things wearing the same word

The word “demand” does double duty here, so it is worth separating the two meanings
before going further.

The first is labor-market demand: the outside world’s appetite for a given kind of
worker. This is public data about occupations, and the college does not own it; it
borrows it. The second is enrollment demand: the college’s own record of what
prospective students explored in Compass. The college owns this outright, because it is
generated on the college’s site and stored in the college’s database. The rest of this
article takes each in turn, because the honest thing to say about them is different in
each case.

The labor-market data: real, public, and mapped, not invented

The wage and outlook figures Pathways shows are pulled from three public sources, each
one the recognized authority for what it provides.

• Wages and employment come from the Bureau of Labor Statistics, through its public
  Occupational Employment and Wage data. This is where the median salary, the 25th and
  75th percentile figures, and the employment count come from.
• Job projections, meaning the percent change in employment and the average annual
  openings, come from the state and national projections published through
  Projections Central.
• Skills for an occupation come from O*NET, the U.S. Department of Labor’s
  occupational database.

None of these speak the language of academic programs. They speak in occupation codes,
the federal Standard Occupational Classification system. A community-college program,
meanwhile, is described by a Classification of Instructional Programs code. Pathways
bridges the two with a standard crosswalk, the same published CIP-to-SOC mapping that
institutional-research offices use, so that a nursing program can be tied to the
occupations a nursing graduate actually enters. This is why the connection between a
program and its wage figures is defensible: it runs through the same standard mapping a
registrar would recognize, not through a guess about which jobs “seem related.”

Two design choices are worth understanding because they shape what you see.

The first is geography with a fallback. Pathways asks for state-level figures first,
because a wage in one state is not a wage in another. When a state-level figure is not
published for a given occupation, and many smaller occupations have gaps at the state
level, Pathways falls back to the national figure and labels it as national rather than
pretending it is local. The projections data works the same way: state first, then
national. The reason for the fallback is that a national number a student can trust is
more useful than a blank, and the reason for the label is that a student deserves to
know which one they are looking at.

The second is caching. These figures do not change day to day, and the public APIs
are rate-limited, so Pathways stores what it fetches and reuses it. Wage data is held
per program and refreshed on a cycle measured in days; projections are held longer.
The practical consequence is that the numbers on a page are recent, not live to the
second, and a college controls how fresh they are by refreshing the data. This is a
deliberate trade of immediacy for reliability and speed.

The most important behavior, though, is what happens when there is nothing to show.
When a source has no figure for an occupation, Pathways does not fill the gap. It
records that no data is available and moves on, rather than substituting an average, an
estimate, or a plausible-looking number. A missing wage stays missing and is presented
as such. This is the same principle Compass follows when it declines to invent a
program, and it comes from the same conviction: an honest blank is worth more than a
confident fabrication. See
How Compass guides students for the finder side of
that same idea.

It is fair to push on the limits here. This data is borrowed, and borrowing has
consequences. The college does not set its cadence, cannot fill its gaps, and inherits
whatever lag the federal sources carry, which for wage and projection data can be a
year or more behind the present. A skeptical dean is right to treat these as
indicators of direction, not as promises about what a specific graduate will earn.
They are the best public evidence available, and they are still evidence about a
labor market, not a forecast about a person.

The first-party signal: owned, anonymous, and off until you turn it on

The second kind of demand runs the other way. Every time a student completes a Compass
search, Pathways can record the three choices they made: the interest, the goal, and
the pace. Aggregated over weeks, this becomes a picture of what prospective students in
your own service area are looking for, drawn from your own site rather than a vendor’s
national panel.

The reason this is valuable is precisely that it is first-party. A market-research
report tells a college what students nationally tend to want. This signal tells a
college what the people who came to its own site, on their own, chose to explore. That
is an enrollment signal the college owns, captured from its own audience, not a number
rented from a data provider and shared with every other client.

What makes it safe to collect is what it deliberately leaves out. The stored record
holds only the interest, the goal, the pace, a timestamp, and a session identifier.
The session identifier is a one-way hash, not a stored address. It is computed from a
site secret, the visitor’s network address, their browser, and the current date, run
through a cryptographic hash so that the original inputs cannot be recovered from it.
The network address itself is never written to the table; it exists only long enough to
compute the hash and to rate-limit abusive traffic. There is no name, no email, no
account, and no raw address in the record. Because the date is part of the hash, the
same visitor on a different day counts as a different session, which keeps the picture
honest as a measure of interest rather than a way to follow a person over time. This is
why the aggregate view can report a count of unique sessions without holding anything
that identifies a unique person.

This collection is off by default. Tracking does nothing until the college enables it,
and while it is off the tracking endpoint simply reports that it is disabled and stores
nothing. That default is a choice: the college opts in, rather than opting out, and a
college that never enables it never accumulates a single row. When it is on, the data
lives in a table in the college’s own database, is read back as simple counts grouped
by interest, goal, and pace over the last week, month, or quarter, and goes nowhere
else.

The candid limitation is that this signal is small and local by nature. It reflects
only the people who found the college’s site and chose to use Compass, which is a
narrower and more self-selected group than the general population. It will tell a
college a great deal about the intent of its actual web audience and very little about
students who never visited. Read alongside the labor-market data, though, the two
answer different questions well: the public data says which fields pay and are growing,
and the first-party signal says which of those fields the college’s own visitors are
drawn to. Neither is a forecast, and neither should be read as one.

What neither of these is

It is worth stating the boundary plainly, because the failure mode for both signals is
the same: treating an indicator as a guarantee. The labor-market data is not a promise
of a salary or a job; it is public evidence about an occupation, cached and clearly
labeled, with gaps left visible. The first-party signal is not a proprietary index sold
back to the college, and it is not surveillance of individual students; it is an
anonymous, aggregate read of what visitors explored, held only if the college chooses
to hold it. This article has stayed with what the two signals are and how far each one
can be trusted. Turning tracking on, refreshing the public data, and reading the
results in the dashboard are operational tasks, and the routes involved are documented
in the REST API reference.