In one sentence

The institution knows its org chart. The student knows their problem. Discovery fails in the space between those two maps.

What Project Ardi showed

Project Ardi began with course planning, but students regularly brought more context than a schedule tool could resolve. The pilot showed the value of a conversational front door and the danger of leaving handoffs implicit: none of the 22 distress-related conversations identified in the study produced an advisor referral. The lesson is not to make a chatbot handle more. It is to make recognition and routing real.

Why a directory is not a discovery system

A directory describes what each office does. Discovery begins earlier, when the student may not know what kind of problem they have. Terms such as incomplete, satisfactory academic progress, degree exception, supplemental instruction, and basic-needs support carry institutional meaning that a first-year or transfer student may not possess.

Fragmentation compounds the problem. The relevant answer can live across a policy page, an appointment system, a departmental site, a course catalog, and a PDF. The institution may regard all five as available. The student experiences five separate systems and must assemble the answer under time pressure.

  • Situation-first language: “I may need to drop a class.”
  • Eligibility and timing: who can use the service, and when.
  • A concrete next action: call, book, visit, upload, or ask.
  • A named owner and a freshness date for every resource.

What a university resource layer should do

A resource layer is not another portal. It is a translation and orchestration layer across existing services. It should accept ordinary questions, clarify the situation, identify the responsible source, expose time-sensitive constraints, and preserve enough context that the student does not have to start over at the next handoff.

The source of truth must remain visible. The system should link to the governing office or policy, distinguish general guidance from a determination, and say when information was last verified. If the institution cannot name an owner for a resource, AI will not make that resource reliable.

How to audit discoverability

Start with real student language from advising notes, search logs, help-desk categories, orientation questions, and student interviews. Test whether someone can reach the correct next step without already knowing an office name. Include edge cases: after-hours questions, multiple simultaneous needs, commuter and transfer experiences, and students who are unsure whether their concern is academic or personal.

Measure successful arrival, not clicks. A search result is not a handoff. The useful metrics are whether the student reached the correct service, understood what to prepare, completed the next action, and avoided unnecessary transfers.

Decision framework

Five questions to take into the room.

  1. 01

    Collect the situations students describe in their own words.

  2. 02

    Map each situation to an authoritative owner and source.

  3. 03

    Add eligibility, timing, preparation, and next-action fields.

  4. 04

    Design warm handoffs that carry context forward.

  5. 05

    Test successful arrival with students who do not know the org chart.

Questions this note answers

  • Why do students miss services that already exist?
  • How should a university organize information for AI search?
  • What is the difference between a portal and a resource layer?
  • How can a chatbot route students without becoming the source of truth?

About the evidence

Project Ardi was one 44-day product pilot at CU Boulder with no comparison group. It measured conversations and product behavior, not academic outcomes. “2,000+” is Ardvarq's company-reported estimate of student users; 903 is the tracked conversation count. Read the research notes and study limits.