1. Choose where to start
Look at every candidate process before picking one. Score each on five factors, then place it on a grid of impact against risk.
- Volume
- How many requests or cases a term, and how many people handle them.
- Time and cost
- Staff hours spent, plus how long people wait.
- Effect on people
- How much the delay or confusion affects students and staff.
- Data sensitivity
- Low, medium or high, using the same scale as the catalog.
- Cross-office dependency
- How many offices touch it, and how many rely on it.
High impact, low riskStart here. Good fits for Apps Script and the fast track.
High impact, high riskWorth doing, with the full seven steps and the right reviewers.
Low impact, low riskLet offices handle these themselves, within the guardrails.
Low impact, high riskUsually not worth it. Consider retiring the process instead.
Tag every candidate with its HERM capability. Grouped by capability, the list shows where offices are duplicating each other and which capabilities nobody has touched. For a fuller prioritization method, see Campus AI Compass ↗.
3. Set up the operating model
Office-level fixes add up only when someone owns the standards. Settle these four questions once, centrally.
- Who owns what
- A central team or a network of local leads sets standards and keeps the inventory. Process owners own outcomes. Named automation stewards own each tool. A small central team with a lead in each division is a practical middle ground.
- How requests come in
- One short intake form, triaged weekly with the five scores above. Every request gets an answer: fast track, workshop, enterprise queue or not now.
- What gets reused
- Grow a shared library from the scenario catalog and the patterns. Before anyone builds, check whether another office already has the same tool.
- Guardrails for staff builders
- The rules on Build it safely, an inventory every production script must be listed in, and a review date on each one. With those in place, staff can keep building without leaving IT a pile of unowned tools.
People need skills too. Process owners need the one-hour workshop, builders need the prompting method, and leaders need enough AI fluency to ask the questions below.
4. Ask for evidence
Require a baseline before launch and a decision at every review. Across the portfolio, track:
- Hours returnedAnd what staff did with them.
- Cycle timeFrom request to outcome, against the baseline.
- Exception rateCases the workflow couldn't handle.
- AdoptionShare of cases that go through the new path.
- Maintenance loadHours spent fixing and updating tools.
- FairnessOutcomes across student groups.
Consolidate or retire whenseveral tools serve the same capability, maintenance takes more time than the tool saves, a tool has no owner, or an enterprise system now does the job.
Per-workflow measures are on Evaluation. For the full method, see the AI Cost, Use, and Outcomes Framework ↗.