AI change management side quest · data contract

What the pilot dashboard needs from Greenlight.

For the reporting review with Jason Dowden. Three feeds, fourteen signals, six questions. The answers fill Section B2 of the assessment and decide which triggers the pilot can run.

ForJoey Fogle

FeedsSuccess Metrics Framework v1.0, Section 2a to 2c; CSA Section B2

StatusQuestions open.

ByDallas Andrews

In one screen

The ask. Per-user, per-day usage with the agent used, the seat roster, and the manager report's access log. Daily is fine.
Why. Eleven of fourteen signals compute from those three feeds. The other three need a source outside Greenlight.
Then. Sample rows out, export in, first weekly review.

Three feeds

Everything the framework measures traces to three tables.

Feed 1

Usage log

One row per user, per day, per agent or skill.

user email or stable id
date
agent name of the agent, skill or project used
uses conversations started, or messages sent, one of the two

Drives active use, time to first use, time to sustained use, feature depth, team spread, and three of the four triggers.

Feed 2

Seat roster

One row per person in the pilot.

user
team and manager
seat_granted the date the tool became available to them

Time to first use counts from the seat date, not the pilot start.

Feed 3

Manager report access

One row per open of the manager report, or of the dashboard.

manager
opened_on

Drives Manager Dashboard Engagement. If the report has no access log, the dashboard can keep its own.

Section B2 · signal by signal

Fourteen signals, each with the field it needs and a column for the answer.

The three blank columns are Jason's: collectable, latency, source. The engine column is what this side computes from the three feeds.

Framework signal	Needs	Engine	Collectable	Latency	Source
Active Usage Rate	Feed 1 counts	computed	yes / no / proxy	rt / daily / weekly / lag	export / api / admin view
Time-to-First-Use	Feed 2 seat date, Feed 1 first row	computed	yes / no / proxy	rt / daily / weekly / lag	export / api / admin view
Time-to-Sustained-Use	Feed 1 counts (derived)	computed	yes / no / proxy	rt / daily / weekly / lag	export / api / admin view
Feature Utilization Depth	Feed 1 agent column, list of available agents	computed	yes / no / proxy	rt / daily / weekly / lag	export / api / admin view
Team Adoption Variance	Feed 2 manager per user (derived)	computed	yes / no / proxy	rt / daily / weekly / lag	export / api / admin view
Manager Dashboard Engagement	Feed 3	computed	yes / no / proxy	rt / daily / weekly / lag	export / api / admin view
Coaching Conversation Rate	Manager self-report per at-risk user (a form, not Greenlight)	proxy	yes / no / proxy	rt / daily / weekly / lag	form / 1:1 notes
Intervention Response Rate	Manager-logged response per fired alert (same form)	computed	yes / no / proxy	rt / daily / weekly / lag	form / tracker
Usage Drop	Feed 1 (derived)	computed	yes / no / proxy	rt / daily / weekly / lag	export / api / admin view
Adoption Plateau	Feed 1 (derived)	computed	yes / no / proxy	rt / daily / weekly / lag	export / api / admin view
Zero-Use Persistence	Feed 1 and Feed 2 (derived)	computed	yes / no / proxy	rt / daily / weekly / lag	export / api / admin view
Workflow Deviation Rate	Process observation or a self-report field; not in usage telemetry	no source	yes / no / proxy	lag	observation
Escalation Pattern Shift	The pilot team's own case or call routing data	no source	yes / no / proxy	lag	pilot team system
Sustained Workaround	Process observation or self-report	no source	yes / no / proxy	lag	observation

Three rows have no telemetry source

Workflow deviation, escalation shift and sustained workaround live in the pilot team's own process, not in any AI tool. Carry them as observed or self-reported, or leave them out with a note.

Six questions for the review

The answers settle latency, access, and who may see what.

1What exists. Which daily, weekly and per-user reports managers already have. Can we see one.
2How it comes out. CSV, API, admin view, or the Zuar path Matt Graves's trackers took. Something the team can run without asking.
3Granularity. Per user per day, with the agent named. Per week works, but time to first use gets rougher.
4History. How far back the log goes. The baseline needs the weeks before the pilot.
5Who sees per-user rows. Proposal: their manager and the project team; leadership sees team level. Jason and Kathryn to confirm.
6Where the dashboard lives long term. If it's Zuar or an internal page, Feed 3 can come from there.

When the answers land

Export in, engine reruns, real weeks on the dashboard. A "no" greys its row rather than removing it, so the gap stays visible.