# D4 · Adoption and Change Management
Adoption curves and champion networks, use-case libraries, incentives, manager behaviour, communication, and the classic failure of licences bought and never used.
import { Accordions, AccordionItem, Tabs, TabItem } from '@prosefly/astro-components';
This domain carries **16%** of the mock — roughly **8 of 50 items**. It tests the difference between *deploying* AI and *adopting* it. The single most common enterprise-AI failure is not technical: it is licences bought, announced, and left unused. This domain covers adoption curves, champion networks, use-case libraries, incentives, the decisive role of first-line managers, and the communication that turns "another tool we've been told to use" into "the way we work now". It is the human counterpart to the [strategy](/openai/leadership/domains/d2-strategy-and-roadmap/) domain's embed phase and depends on the [workforce](/openai/leadership/domains/d5-workforce-and-enablement/) domain's enablement.
## What you need to know
Adoption follows a curve: a few innovators and early adopters move first, an early and late majority follow if you make it easy and rewarded, and laggards move last. Champions — respected practitioners, not the AI team — carry adoption further than any top-down mandate. A living **use-case library** turns one team's win into every team's starting point. Incentives and, above all, **manager behaviour** decide whether the majority moves: people copy their manager, not the CEO's email. The signature failure to recognise and prevent is the **licence graveyard** — seats provisioned, a launch email sent, and usage flat within a month because nobody changed how the work is actually done.
## Learning objectives
By the end of this page you should be able to:
1. **Locate** a population on the adoption curve and choose the right intervention for each segment.
2. **Design** a champion network that scales adoption beyond the central team.
3. **Build** a use-case library that makes good practice the default starting point.
4. **Align** incentives and manager behaviour so the majority adopts.
5. **Communicate** change in a way that reduces fear and models real use.
6. **Diagnose and prevent** the licences-bought-but-unused failure.
---
## 4.1 The adoption curve
Adoption is not uniform. The population splits into segments that need different things, and the make-or-break gap sits between the early adopters and the majority.
| Segment | Share (rough) | What they need |
| --- | --- | --- |
| Innovators | ~2–3% | Access and freedom; they will find uses themselves |
| Early adopters | ~13% | Recognition and a channel to share wins |
| Early majority | ~34% | Proof it works for *their* job, and it is easy |
| Late majority | ~34% | Peer pressure, manager expectation, support |
| Laggards | ~16% | Clear expectation and removal of the old path |
```text
usage
▲ early late
│ early majority majority
│ adopters ████████ ████████ laggards
│ ████ ░░
│ ██ THE CHASM ↑
│██ innovators (make it easy +
└──────────────── proven + rewarded)──────────► time
```
The programme's job is to get the majority across the chasm, and the majority does not respond to novelty — it responds to *proof for its own role* plus *low friction*. Announcements move innovators; they do nothing for the majority.
:::tip[Assessment signal]
Stems describing **"usage spiked then fell", "innovators love it but nobody else uses it", "how do we reach the majority"** are testing the chasm. Correct answers reach the majority with role-specific proof, manager expectation and support — not another all-staff announcement.
:::
## 4.2 Champion networks
Champions are respected peers embedded in the business who demonstrate real use in the local context. They out-perform central mandates because the majority trusts someone who does their job, not the AI team.
| Do | Don't |
| --- | --- |
| Pick credible practitioners, one per team/function | Appoint volunteers nobody respects |
| Give them time, recognition and early access | Add it as unpaid extra work |
| Equip them with the use-case library and a channel | Leave them to invent everything alone |
| Measure local adoption, not champion activity | Reward attendance at champion meetings |
A champion network converts a central programme into local behaviour change. It is the mechanism that carries adoption from early adopters into the majority, one team at a time.
## 4.3 Use-case libraries
A use-case library is a curated, living collection of proven ways people in *this* organisation use AI, with the prompt or workflow, the role it suits, and the benefit seen. It removes the blank-page problem that stalls the majority.
```text
Title : Weekly account summary for CSMs
Role : Customer success
Problem : 45 min every Monday compiling account status
Approach : Prompt template + connected CRM notes
Benefit : ~30 min saved / CSM / week; more consistent format
Owner : J. Okafor (CS champion)
Reviewed : governed workspace only; no customer PII in prompt
```
- Turns one team's discovery into every peer's starting point.
- Encodes the *governed* way to do it (data rules baked in), reducing risk.
- Grows from champions and users, so it stays real rather than aspirational.
- Gives the majority "for my role, here is exactly what to do", which is what moves them.
A stale, top-down library dies; a living one curated by champions and fed by users becomes the first place people look.
## 4.4 Incentives and manager behaviour
Incentives set direction, but **manager behaviour is the strongest single adoption lever.** Teams copy their immediate manager's habits, not a mandate from three levels up.
| Lever | Weak version | Strong version |
| --- | --- | --- |
| Recognition | Generic "innovation award" | Spotlight real, role-specific wins in team meetings |
| Expectation | CEO email says "use AI" | Manager expects and reviews AI use in normal work |
| Time | "Do it on top of your day job" | Protected time to learn and to share |
| Metrics | Track licences issued | Track active use and value in the workflow |
| Modelling | Leaders talk about AI | Managers visibly use AI in front of their teams |
The uncomfortable truth for the exam: a manager who does not use AI themselves will not produce a team that does, regardless of incentives. Adoption programmes therefore target managers first.
## 4.5 Communication that reduces fear
Bad AI communication triggers the workforce risk from [D3](/openai/leadership/domains/d3-governance-and-risk/): fear of replacement produces quiet resistance. Good communication is specific, honest and repeated.
| Message goal | Weak | Strong |
| --- | --- | --- |
| Purpose | "We're an AI-first company now" | "Here is the task this removes and what you do with the time" |
| Job security | Silence, which breeds rumour | Honest statement of intent (augment vs reduce) — see [D5](/openai/leadership/domains/d5-workforce-and-enablement/) |
| How-to | One launch webinar | Role-specific, repeated, with the use-case library |
| Feedback | One-way announcements | A channel where problems are heard and fixed |
Silence is the worst option: in the absence of a clear message, people assume the most threatening one and disengage.
## 4.6 The licence graveyard
The defining failure of enterprise AI: buy seats, send a launch email, celebrate the rollout — and watch active usage collapse within weeks because nothing about how the work is done actually changed.
```text
LICENCE GRAVEYARD LIVING ADOPTION
───────────────── ───────────────
buy seats buy seats for a proven use case
send launch email vs managers model + expect use
measure licences issued measure active use + value
usage: spike → flatline champions + library sustain it
root cause: deployment, root cause fixed: behaviour
not adoption change, not tool access
```
The diagnosis is always the same: the organisation treated a *behaviour change* as a *procurement event*. The cure is proof-for-role, manager modelling, a use-case library, a champion network and metrics that count *active use and value*, not seats sold.
:::tip[Assessment signal]
Stems reporting **"we bought 5,000 licences but usage is 8%", "adoption stalled after launch", "the rollout email went out but nothing changed"** are the licence-graveyard signal. Correct answers fix behaviour (managers, champions, role proof); distractors buy more licences, send another email, or blame the tool.
:::
## Decision framework
### The A-D-O-P-T diagnostic
When adoption stalls, run these five checks in order to find the real blocker before spending more.
| Letter | Check | If failing |
| --- | --- | --- |
| **A**ctive use | Are people *using* it, or just licensed? | You have a graveyard; fix behaviour, not seats |
| **D**emonstrated value | Is there role-specific proof it helps? | Build use-case library entries per role |
| **O**wnership by managers | Do managers model and expect use? | Target managers first; make it their metric |
| **P**eer channels | Do champions and a library exist? | Stand up a champion network and library |
| **T**ime and incentives | Is there protected time and recognition? | Reallocate time; recognise real wins |
The order matters: measuring active use first tells you whether you have an adoption problem at all, before you invest in champions or communication.
## Common mistakes
| Mistake | Why it happens | What to do instead |
| --- | --- | --- |
| Measuring licences issued as success | It is the easy number | Measure active use and value in the workflow |
| One launch email as the change plan | Rollout feels like adoption | Sustained, role-specific enablement plus manager modelling |
| Mandate from the top with no manager buy-in | Authority feels sufficient | Target first-line managers; teams copy them |
| Appointing champions nobody respects | Anyone can volunteer | Choose credible practitioners and give them time |
| A top-down, static use-case library | Central control feels tidy | A living library fed by champions and users |
| Silence on job impact | Avoiding a hard conversation | Communicate intent honestly to prevent fear-driven resistance |
| Blaming the tool when usage stalls | Easier than changing behaviour | Diagnose with A-D-O-P-T; the blocker is usually behaviour |
| Buying more licences to "boost adoption" | Procurement is a familiar lever | More seats do not create use; fix the behaviour first |
## Scenario challenge
**Scenario.** Nine months ago your company bought 6,000 ChatGPT Enterprise seats and announced them with a keynote and an all-staff email. Today the dashboard shows 11% weekly active use, concentrated in the product and data teams. A board member asks whether the investment was a mistake and suggests either "buying a better tool" or "mandating usage with a target for every employee". Your engagement data shows two things: the teams using it heavily had engineers who already loved it, and a survey found many staff "don't know what they'd use it for in my role" and "my manager never mentions it".
**Expert reasoning trace.**
1. **Name the failure precisely.** This is a licence graveyard, not a tool failure. Usage concentrated in innovator-heavy teams (product, data) and flat elsewhere is the classic chasm signature: the early adopters moved, the majority never did. Buying a different tool would reproduce the same curve.
2. **Reject the two board suggestions as written.** "Buy a better tool" misdiagnoses a behaviour problem as a product problem. "Mandate a usage target for everyone" produces gaming (open-and-close sessions to hit the number) and resentment, not value — activity is not adoption.
3. **Run A-D-O-P-T.** Active use is low (confirmed). Demonstrated value: the survey's "don't know what I'd use it for" is a *role-proof* gap — there is no use-case library for their jobs. Ownership: "my manager never mentions it" is the decisive finding — managers are not modelling or expecting use. Peer channels: no champions outside the early teams. Time/incentives: none allocated.
4. **Sequence the fix by leverage.** First, managers: make role-relevant AI use a normal expectation that managers model and review — because the majority copies managers, this is the highest-leverage move. Second, champions: recruit a credible practitioner in each lagging function. Third, use-case library: harvest the product/data teams' wins and, via champions, create entries for finance, sales, operations so every role has a "here's exactly what to do". Fourth, communicate honestly about intent to defuse the unspoken job-loss fear surfacing as disengagement.
5. **Change the metric.** Report *active use and value by role*, not seats issued or raw logins, so the board sees real movement across the chasm rather than a vanity mandate.
**Board-ready outcome:** the diagnosis is behaviour, not tooling; the two proposed fixes are declined with reasons; the plan targets managers first, then champions, then a role-specific use-case library, with honest communication and value-based metrics — the levers that actually move the majority.
## Assessment traps
| Trap | Why it is tempting | The discriminator |
| --- | --- | --- |
| Buy a different tool to fix low usage | Reframes behaviour as product | A graveyard is a behaviour failure; a new tool repeats the curve |
| Mandate a usage target for every employee | Feels decisive and measurable | Mandates create gamed activity, not adoption; target managers and role proof |
| Send another all-staff announcement | Communication feels like action | Announcements move innovators only; the majority needs role proof + managers |
| Count logins or seats as adoption | Easy, available numbers | Measure active use *and value in the workflow* |
| Appoint eager volunteers as champions | Willingness looks like fit | Champions must be respected practitioners, or peers ignore them |
| Stay silent on job impact to avoid conflict | The conversation is hard | Silence breeds fear and resistance; communicate intent honestly |
## Practice questions
Each item states how many responses to select. Commit before revealing.
A. The tool is broken.
B. Early adopters moved but the majority has not crossed the chasm.
C. Everyone is using it as intended.
D. Licences were never provisioned.
**Answer: B.** Concentrated use among innovator-heavy teams with flat use elsewhere is the classic chasm signature. The tool is not broken (A), adoption is clearly incomplete (C), and licences were provisioned (D) — they are simply unused.
A. Another all-staff email.
B. First-line managers visibly using AI and expecting it in normal work.
C. A larger licence count.
D. A more advanced model.
**Answer: B.** The majority copies its immediate manager; manager modelling and expectation is the strongest adoption lever. Emails (A) reach innovators, more seats (C) do not create use, and model capability (D) is not the blocker.
A. It undercounts usage.
B. Licences issued measures deployment, not adoption; it says nothing about active use or value.
C. It is too expensive to track.
D. It should be reported monthly, not once.
**Answer: B.** Seats issued is a procurement number, not an adoption or value number. It does not undercount usage (A), cost (C) and cadence (D) are not the issue — the metric measures the wrong thing.
A. Select respected practitioners embedded in each function.
B. Give champions protected time, recognition and early access.
C. Reward champions for meeting attendance.
D. Appoint whoever volunteers first.
E. Have champions report only to the central AI team.
**Answer: A and B.** Credible practitioners with time and recognition drive local adoption. Rewarding attendance (C) measures the wrong thing; volunteers (D) may lack credibility; reporting only upward (E) misses the point of local peer influence.
A. Send more general training on AI.
B. Build role-specific use-case library entries showing exactly what to do for their jobs.
C. Increase the licence count.
D. Mandate daily usage.
**Answer: B.** The gap is role-specific proof; a use-case library for their roles removes the blank-page problem. General training (A) is not role-specific, more seats (C) are unused, and a mandate (D) produces gamed activity.
A. It will cost too much.
B. It produces gamed activity (logins to hit a number) rather than genuine value-adding use.
C. It is technically impossible to measure.
D. It will overload the model.
**Answer: B.** Usage mandates drive superficial activity, not adoption or value. Cost (A), measurability (C) and model load (D) are not the central problem — gaming is.
A. The living one is longer.
B. The living one is continuously fed and curated by champions and users, so it stays real and role-relevant.
C. The living one is owned solely by the AI team.
D. The living one contains only executive examples.
**Answer: B.** A living library grows from real use by champions and staff, keeping it relevant. Length (A) is not the point, sole central ownership (C) makes it stale, and executive-only examples (D) miss the majority's roles.
A. Say nothing until the strategy is final.
B. Communicate the intent honestly (augment vs reduce), name the tasks affected, and open a feedback channel.
C. Announce that the company is 'AI-first' and move on.
D. Threaten disciplinary action for non-use.
**Answer: B.** Honest, specific communication with a feedback channel reduces fear-driven resistance. Silence (A) breeds rumour, a slogan (C) says nothing concrete, and threats (D) deepen resistance.
A. Whether more licences are needed.
B. Whether there is genuine active use, to confirm an adoption problem exists before investing further.
C. Which model is fastest.
D. The vendor's roadmap.
**Answer: B.** A confirms active use first, so you know you have an adoption problem before spending on champions or communication. Licences (A), model speed (C) and vendor roadmap (D) are not the first diagnostic step.
A. Seats were bought and a launch email sent, but nothing changed about how work is done.
B. Usage spiked at launch then flattened within weeks.
C. Champions in every function actively share wins.
D. Managers model and expect AI use daily.
E. Metrics track value delivered in the workflow.
**Answer: A and B.** Buying seats without changing the work, and a spike-then-flatline usage pattern, define the graveyard. Active champions (C), manager modelling (D) and value metrics (E) are hallmarks of *successful* adoption, not the failure.
A. Send those teams extra licences.
B. Make role-relevant AI use an expectation managers model and review, since teams copy their managers.
C. Escalate to the CEO for another company-wide email.
D. Replace the tool for that division.
**Answer: B.** The blocker is manager behaviour; making use a modelled expectation is the highest-leverage fix. Extra licences (A) go unused, another email (C) reaches only innovators, and swapping tools (D) misdiagnoses the problem.
A. Total licences purchased.
B. Weekly active users applying it to real, role-relevant tasks with evidence of value.
C. Number of launch emails sent.
D. Attendance at the kickoff webinar.
**Answer: B.** Active, value-adding use by role is the indicator that adoption is real. Licences purchased (A), emails sent (C) and webinar attendance (D) all measure deployment or activity, not adoption.
## Key takeaways
- Adoption follows a **curve**; the programme's job is to move the **majority across the chasm** with role proof and low friction, not to impress innovators.
- **Champion networks** of respected practitioners carry adoption further than any top-down mandate.
- A **living use-case library** removes the blank-page problem and bakes in the governed way to work.
- **Manager behaviour is the strongest single lever**: teams copy their managers, so target managers first.
- **Communicate intent honestly**; silence about job impact breeds fear-driven resistance.
- The **licence graveyard** — seats bought, email sent, usage flat — is a behaviour failure, not a tooling one.
- Diagnose stalls with **A-D-O-P-T**; measure **active use and value**, never seats issued.