Campus AI Development High-value opportunities

1.1 — High-value opportunities

Where building with AI pays off

Cheap building doesn’t make every idea worth building. The strongest candidates share a few traits: a job that’s poorly served today, people who know the work, data that can stay inside approved tools and an outcome someone can see.

What strong candidates have in common

  1. 01A niche or poorly served processToo specific, too small or too local for a vendor to serve well, and too low on the list for central IT to staff.
  2. 02Manual work with a clear expertHigh volumes of spreadsheets, re-keying or email routing, and a person who knows exactly how the work should go.
  3. 03A well-understood job to be doneThe people involved can say what “done” looks like. Exploration is fine; an unclear purpose isn’t.
  4. 04Data that fits approved toolsPublic, internal or moderate-sensitivity data that can stay inside tools approved for its level. See data classification.
  5. 05An outcome you can measureHours saved, errors avoided, requests closed faster, a backlog cleared. Something that will show whether it worked.
  6. 06A faster path than the alternativesBuilding gets value sooner than a procurement cycle or a new hire would, and the institution can still own the result.

Use these as signs. A candidate missing one or two can still be worth building. One missing most of them usually isn’t.

Good fits, poor fits

What the line looks like in practice

The same tools serve both columns. What differs is what the software touches and what happens when it’s wrong.

Good fits

Operational systems the market skipped

  • Facilities data scattered across spreadsheets, legacy systems and scanned leases, brought into one system (Emerson’s GRID). Case
  • Help desk forms that grew into guarded self-service for routine account tasks (Emerson’s PAM)
  • A unit dashboard built from internal data by the staff who use it. Scenario
  • A course tool or study aid at T1–T2. Scenario
  • An analysis pipeline a research software engineer reviews. Scenario
Proceed with caution

Where cheap building is the wrong reason

  • Core systems of record (student information, ERP, HR, finance): configure or extend the platform instead. Platforms
  • High-stakes automated decisions about admissions, aid, discipline or grades. Case
  • Heavily regulated data (HIPAA, CUI, export-controlled, identifiable human subjects): approved enclaves only
  • A tool that duplicates an enterprise capability. Enterprise fit
  • Anything nobody will own after launch. Case

A short filter

Five questions before the six

This filter doesn’t replace the six questions or the Strategic Compass. It tells you which one to start with and gives you your first answers.

  1. 01What job is poorly served today, and who does it now?
    Names the domain expert and the current workaround.Feeds Six questions: Audience
  2. 02Does a platform or contract the institution already has do this?
    If yes, configure before you build.Feeds Enterprise fit
  3. 03What is the most sensitive data it would touch?
    Sets the ceiling on tools, tier and approach.Feeds Six questions: Data
  4. 04What would change if it worked, and how would you know?
    Becomes the success measure.Feeds Compass: Plan and Track
  5. 05Who will own it a year from now?
    No answer usually means no build, or a time-boxed experiment.Feeds Six questions: Accountability, Lifespan

Where the answers go

Institutional initiative, or not?

  • Run the Compass first if it needs funding or staff time beyond one person, serves others, touches confidential data or affects decisions about people. Compass crosswalk
  • Go straight to the six questions for a personal tool or throwaway prototype at T1–T2. The seven steps
  • Either way, the answers carry into the chooser and the control tiers.

Problem types

What commonly fits on campus

Five kinds of work show up again and again. Each still runs through the data check and the control tiers.

Workflow automation

Intake, routing and approvals

Request forms, approval chains, reminders and status updates that now live in email and shared inboxes.

Functional analyst scenario →
Internal tools

Trackers that replace spreadsheets

Inventories, schedules, small admin apps and reports for one office or department.

Business user scenario →
Data integration

One view of scattered sources

Domain-specific data pulled from several systems into one place, as with Emerson’s facilities system and institutional research data hub.

Emerson case →
Access and identity helpers

Guarded self-service

Routine account tasks handled by front-line staff within guardrails. Needs the identity team as a partner and usually sits at T3 or above.

Operating controls →
Faculty and research support

Course tools and lab utilities

Study aids, assignment helpers, analysis pipelines and instrument scripts, with review scaled to whether findings or grades depend on them.

Research scenario →

Next: weigh what a candidate will cost to own in the economics of building, then run it through the six questions.