ai-adoption · deliverable

“AI at Your Desk” — Department AI Adoption Program (design memo)

Client: Helios Facilities Group (fictional; Kit 3 universe) — 300 desk employees across Finance, HR and Operations-planning.

Audience: general employees — not trainers, not developers. People with a Monday-morning workload and no AI background.

One-line: move 300 desk workers from “AI is a chatbot I’ve heard of” to “AI is a governed habit in my week” in 60 days, without a single policy-breach incident.

1. Anti-generic anchor: this program is about three people, not “AI adoption”

Abstract adoption programs produce abstract adoption (zero). Everything below is designed around three named personas and their real Monday-morning tasks:

Persona	Desk	Monday-morning task	Before AI	After AI (trained)
Marta, 54	Finance (AP)	Invoice-line anomaly spotting before the weekly payment run: ~240 lines, manual scan for duplicates, threshold splits, VAT mis-coding	~3.5 hours, fatigue errors cluster after hour 2	AI pre-screens lines into an anomaly table (RTF + decomposer patterns); Marta verifies and decides. ~1 hour, and the boring 240-line scan is now a 40-row review
Jeroen, 31	HR	Job-ad drafting + CV-screening notes for the Facilities Coordinator opening (38 CVs)	Ads: 2 hours drafting + legal check round-trip. Screening: ~15 min/CV, style drifts by CV #20, bias risk rises with fatigue	Ads: draft + critique-loop in 25 min. Screening: structured notes per CV with the bias checklist baked into the prompt (few-shot pattern); ~6 min/CV, consistent style, auditable notes
Sanne, 42	Ops-planning	Schedule-variance summary from last week’s shift logs for the Monday management meeting	90 minutes copy-pasting from three log exports into a summary managers half-read	Decomposer prompt turns logs into per-site variance + cause attribution in 20 minutes; Sanne sanity-checks the numbers and writes the “so what”

Design consequence: every example prompt, every huddle script and every success story in this program uses these tasks. If a training example doesn’t map to a Marta/Jeroen/Sanne task, it gets cut.

2. The 4-stage adoption arc

Stage	What it means	Assets	Exit criteria
1 · Curious	Knows what the tools can/can’t do for their desk; has seen a live demo on their own task type	30-min desk-specific demo session · governance basics (04) · red-flag list	Can name one task of their own that AI could help with, and one thing that must never go into a tool
2 · Capable	Has applied ≥2 prompt patterns to real work with a usable result	Prompt-pattern trainer (02) · department example pack · sandbox time (2×30 min in work hours)	Produces one real work output using a pattern, self-assessed “usable with light edits”
3 · Confident	AI use is a weekly habit, not an experiment; knows when not to use it	Team “first AI habit” (manager pack 03) · AI champion floor support · critique-loop discipline	Weekly active use for 3 consecutive weeks; can state the escalation path from memory
4 · Governed	Follows disclosure + data-classification norms without prompting; flags near-misses	Governance basics (04) · quarterly refresh · champion network	Zero unforced norm violations in the quarter; at least one disclosed AI-assisted client-facing deliverable with a human name on it

3. Audience segmentation — real tasks, real time math

Group	n	Anchor task (before → after)	Primary patterns
Finance	~90	Payment-run anomaly spotting: 3.5h → ~1h verify-and-decide · vendor email drafting with critique loop	RTF, Decomposer, Critic-loop
HR	~60	Job-ad drafting: 2h → 25 min · CV-screening notes: 15 → 6 min/CV with bias checklist in the prompt	Few-shot, Critic-loop, Decomposer
Ops-planning	~70	Schedule-variance summary: 90 → 20 min · recurring report skeletons	Decomposer, RTF
Other desks	~80	Join wave 2 with their own anchor task, chosen with their team lead	RTF first, then need-based

Time figures are self-reported estimates collected in the pilot and reported as such — see objectives below.

4. Rollout plan

Pilot squad (weeks 1–3): 12 volunteers — 4 per department, mixed seniority, including at least 2 self-declared skeptics (skeptic feedback is the most valuable feedback). Full trainer + manager pack + weekly 20-min retro.
Wave pattern (weeks 4–8): wave 1 = Finance + HR (~150), wave 2 = Ops-planning + other desks (~150). Each wave: demo session → trainer access → two sandbox slots → team habit selection. Waves are capped so floor support never thins below 1 champion per 25 trainees.
Floor-support model — “AI champions”: 12 champions (one per ~25 seats), 2 hours/week protected time, a shared channel, and a monthly pattern-exchange meeting. Champions do not do people’s work for them — they unblock, demonstrate, and escalate tool/policy questions.

5. Measurable objectives (pre-committed)

Objective	Target	Instrument
Weekly active AI-tool usage among trained staff by day 60	≥60%	Tool-license activity log + monthly 3-question pulse
Policy-breach incidents (red-data in external tools, undisclosed client-facing AI text)	Zero	Champion escalation log + quarterly audit sample
Task-time savings, self-reported, pilot cohort	≥2 h/week median	Pilot retro survey (weeks 3 + 6), reported with spread, not just median
Pattern competence	≥80% of trainees pass all 4 pattern checks	Trainer check scores
Disclosure norm adherence	100% of sampled client-facing AI-assisted deliverables carry a human name	Quarterly audit sample (n≥20)

6. What this program deliberately does NOT do

No “100 prompt tricks” content dump — four patterns, deeply practiced, beat forty patterns skimmed.
No mandatory usage quotas — quotas produce fake usage; the habit model produces real usage.
No productivity claims beyond self-reported time savings in year 1 — see honest attribution in the measurement note (Kit 5 discipline carries over).

Fictional disclosure: Helios Facilities Group, personas and figures are a fictionalized composite built from realistic office-work patterns. Memo drafted with AI assistance; program design, targets and final copy are human-owned.

Prefer the source? Download the raw markdown.