30/60/90 decision record and measurement sheet
One decision record per pilot. Fill it in as you work through Lesson 6.1, and use Lesson 6.2 for the measures. Every field asks for your own numbers; none of them comes with a default.
The site does not send or save your worksheet entries. Printing and the Markdown download happen in your browser. Print or download a copy before leaving; do not rely on the browser to retain your entries.
See a filled example (Illustrative example, not a real pilot)
Illustrative example, not a real pilot. The setting is invented and the values are relative on purpose: use your own measured baseline, never these.
- Pilot name
- Sample-request intake at a university core facility
- 30/60/90 review dates
- Three dates set on the decision owner's calendar on the day the record was opened: day 30, day 60 and day 90.
- Organization or unit
- A university core facility that runs sample analyses for other labs
- Problem and affected users
- Requests from other labs wait days before analysis starts. The requesting researchers wait, and the two intake staff spend part of each day chasing missing details by email.
- Current workflow and baseline, with source
- Calendar days from request to analysis starting, and the share of requests that needed at least one email back, both counted from the shared inbox for last quarter. Intake staff time comes from the staff's own estimate and is labelled as an estimate.
- Proposed change, and the non-AI alternative
- Non-AI fix first: the form gained the missing required field and all requests now go to one named inbox. That removed much of the wait. The delay that remained was sorting each request to the right instrument team, so the pilot tests an assistant that reads each request and suggests the team and any missing detail, with staff confirming every suggestion.
- Unknowns to test
- Whether checking the suggestions takes less staff time than sorting by hand; whether unusual sample types are sorted correctly; whether the requesting labs notice any difference.
- Team and named decision owner
- The two intake staff, the instrument lead who knows what a correct routing looks like, and the facility's IT contact. Decision owner: the facility director.
- One obstacle to remove
- The request to connect the assistant to the intake inbox sat in the research computing queue with no owner. The queue protects research data. The associate dean for research named a reviewer and a decision date, and the request was limited to request forms, with no sample data.
- Pilot boundaries
- Request forms only: last quarter's forms for a first check, then new requests for the rest of the pilot. No sample results and no billing records. Staff confirm every suggestion before a request moves, and the assistant sends nothing to the labs.
- Applicable policy or compliance constraints
- The university's data classification for research administration records. The privacy office reviewed the vendor's retention terms. No human-subjects data is involved, which the research ethics board (IRB) office confirmed in writing.
- Accountable reviewers
- Privacy office lead (data handling), research computing security reviewer (the connection), instrument lead (routing accuracy), facility director (the decision).
- Pass bar and stop condition
- Pass: calendar days from request to analysis starting fall to about half of the baseline measured after the form fix, with no more misrouted requests than staff routing produces today. Stop: misrouting rises above today's level in two consecutive weeks, or any sample data is found in the tool.
- Total cost by category
- Implementation: the vendor subscription for the pilot period, the IT contact's time to connect the inbox, and intake staff time to count the baseline from the shared inbox. Review: the privacy and security reviews. Training: one session for intake staff. Support: the IT contact, on a stated schedule. Ongoing operation: the subscription and the instrument lead's monthly accuracy check. All estimates, labelled as such.
- Operating owner
- The intake supervisor, named before the pilot started. The support route and a budget line were confirmed before the day-90 decision.
- Measured pilot results at day 60 / day 90 (against baseline)
- Day 60: days to analysis starting at about half of the post-fix baseline; misrouting at today's level; staff say checking a suggestion is quicker than sorting by hand. Recorded as met, with the day-90 decision kept on its date. Day 90: days to analysis starting held at about half of the post-fix baseline across the period; misrouting slightly below today's level. Source for both: the shared inbox, counted the same way as the baseline.
- Final decision: expand, revise, defer or stop
- Expand, recorded on the day-90 date by the facility director: extend the assistant to requests that staff log from phone calls. Reason: the bar was met against a measured baseline, and the operating owner, support route and budget were in place first.
Three calendar dates, set before the pilot starts. The day-60 review decides whether the pilot continues.
By day 30: the problem, the baseline, the people
The workflow, who does it today, and who waits on it.
How long it takes end to end, how often it is wrong or redone, what it costs. Name where each number came from and the period it covers.
What the pilot would do differently, and the fix that needs no AI, considered first.
What you do not know yet that the pilot has to answer.
Who does the work, and the one person who decides at day 60 and day 90.
The queue, access or sign-off you will own clearing, the purpose it serves, and by when.
Before pilot day one
Review proportionate to the consequence. Lesson 5.2 covers the questions, and its dated note covers federal policy constraints. Lesson 5.2
Which users, which data, which systems, for how long. What the pilot will not touch.
Privacy, records, security, procurement, research ethics board (IRB). Federal programs: see the note in Lesson 5.2.
Who signs for security, privacy, reliability and accountability, by name.
What result expands the pilot, and what result stops it. Written before day one.
Implementation (including baseline collection and evaluation), review, training, support, ongoing operation. Estimates are fine; label them.
Who runs it after the pilot. Named before the pilot starts; support and budget confirmed before any expansion.
At day 60 and day 90: the results and the decision
The same measures as the baseline, taken at each review, with where each number came from. Say whether the pass bar or the stop condition was reached, or record that the evidence cannot settle the bar, the missing evidence and a cap on the time or money spent collecting it. A second review that still cannot judge is a stop.
One word, the date, the owner, and the one-sentence reason. Lesson 5.1 says when each one fits.
Back to the course overview.