Turn an AI Adoption Resistance Case Into a Change Management Brief

The task

You're leading an enterprise AI deployment and a business unit is quietly (or loudly) refusing to use the tool. Before the next steering committee, you need a short brief that names the root cause, gives the sponsor something to say, and lists the next two moves. This workflow turns a messy field note into that brief in one sitting.

Before AI

Normally this means a 45-minute stakeholder interview, a call with the vendor CSM, a second call with the local change lead, then an hour wrestling a slide together. The output is usually a status update dressed up as analysis — root causes get skipped because you ran out of time. As Jun Tay, finance technology specialist at DBS Bank in Singapore, points out, the gap between deploying AI and anyone using it well is where adoption is won or lost, and that gap is exactly what these rushed briefs fail to close.

The framing behind this workflow is Tay's argument that underinvestment in change management — not the platform — is what kills adoption. We use it to force the model past "run more training" as a default recommendation.

The workflow

Step 1 — Diagnose the resistance. Paste the raw field note. The prompt below extracts the friction pattern, separates surface complaints from underlying causes, and flags what's missing.

You are a senior change management consultant advising on enterprise AI deployments. I will paste a field note describing stakeholder resistance to an AI tool rollout. Do the following:

1. Summarize the situation in 3 bullets: who, what tool, what behavior.
2. Separate SURFACE complaints (what people are saying) from LIKELY ROOT CAUSES (what's actually driving it). Use two labeled lists.
3. Classify the dominant resistance pattern using this taxonomy: (a) accountability ambiguity — unclear who owns the AI's output, (b) workflow mismatch — tool doesn't fit the actual job, (c) trust deficit — output quality or explainability concerns, (d) status threat — the tool changes power or expertise dynamics, (e) incentive misalignment — no reason to use it, existing metrics penalize using it.
4. List what's MISSING from the field note that you'd want before finalizing a diagnosis (max 4 items).

Keep it under 400 words. Be blunt. Do not recommend "more training" unless the evidence actually supports it.

Here is the field note:

Deployment: "Lumen Copilot" (GenAI drafting assistant from vendor Northwind) rolled out 11 weeks ago to the Commercial Credit team at Meridian Regional Bank — 42 relationship managers (RMs), 6 team leads, 1 division head (Priya S.).

Adoption: 18% weekly active after week 3, dropped to 9% by week 10. Target was 60%.

Field observations (from 4 RM interviews + 1 team lead 1:1, week 10):
- RMs say the memos it drafts are "80% there but the last 20% takes longer than writing from scratch."
- Two RMs said their team lead flagged an AI-drafted credit memo in committee as "not sounding like our house style" — since then those RMs stopped using it.
- Team lead Marcus T. told me privately: "If a deal goes bad and the memo was AI-drafted, whose neck is on the line? Mine? The RM's? Northwind's? Nobody's said."
- Compliance sent an email in week 6 saying AI-generated content must be "reviewed and attested" — no template provided for the attestation.
- Priya (division head) is publicly supportive in town halls but has not used the tool herself and did not attend the vendor's advanced-use workshop.
- Comp plan for RMs rewards deals closed and portfolio quality. No mention of tool usage. Bonus cycle closes in 8 weeks.
- The other pilot site (Wealth Advisory) hit 55% adoption using the same tool. Difference noted: their division head demoed her own use case at kickoff.

Step 2 — Build the brief. Turn the diagnosis into something a sponsor can act on this week.

Now produce a one-page Change Management Brief for the executive sponsor (the division head). Structure it exactly as follows, using these headings:

**Situation** (2 sentences, no jargon)

**Root Cause — Primary** (1 paragraph, name the dominant pattern from step 1 and explain the mechanism in plain English)

**Root Cause — Contributing** (bulleted list, max 3)

**What Won't Work** (2-3 bullets — call out the obvious-but-wrong moves, e.g. "another training session," "a stern email from Compliance")

**Recommended Moves — Next 2 Weeks** (numbered list of 3-4 concrete actions with an owner suggestion for each; each action must be executable without buying anything new)

**Sponsor Talking Points** (3 short quotes the division head can actually say out loud — one for a team huddle, one for a 1:1 with the skeptical team lead, one for the next steering committee). Each under 30 words. Sound like a human, not a press release.

**Success Signal at 30 Days** (one measurable behavior, not a usage percentage)

Keep the whole brief under 500 words. Write for a busy executive who will skim.

Step 3 — Pressure-test. Before you send it, have the model argue against itself.

Now play the role of a skeptical Chief Risk Officer reading this brief. Give me:

1. Three questions you would ask that the brief does not currently answer.
2. One recommendation in the brief that could backfire, and how.
3. One thing in the original field note the brief may have under-weighted.

Be specific. No generic "consider stakeholder alignment" comments.

Gotchas

• Garbage in, garbage out. If your field note is just "the team hates it," step 1 will invent root causes. Force yourself to capture at least one direct quote, one usage number, and one incentive-structure detail before running this.
• The taxonomy is a lens, not the truth. Real resistance is usually 2-3 patterns tangled together. If the model picks one confidently, re-run step 1 asking it to argue for the second-most-likely pattern.
• Sponsor talking points need a human pass. LLMs default to corporate-speak. Read them aloud. If your sponsor wouldn't say it in a hallway, rewrite it.
• Don't skip step 3. The brief will read as clean and confident even when it's wrong. The CRO role-play is where you catch the recommendation that would blow up in committee.
• This does not replace the interview. It replaces the writeup. You still need to have actually talked to the resistant users — otherwise you're generating fiction with structure. As DBS's team frames it in their own responsible AI adoption guidance, humans remain accountable when AI fails or leads to a poor decision — you cannot sue an AI in the courtroom, and that principle applies to the briefs you send upstairs too.

Time saved

Roughly 60-90 minutes per stakeholder case, mostly recovered from the "staring at a blank slide" phase and the second round of edits. The interview time doesn't shrink. What shrinks is the gap between having the observations and having a document a sponsor can act on. Conservative estimate — if you run this weekly across a multi-site rollout, budget 4-6 hours saved per month per workstream lead.

What we got when we ran it

We executed every prompt above, in order, against the sample input (model: claude-sonnet-4-6). Outputs below are real and unedited apart from length truncation.

Step 1 output

Lumen Copilot Resistance Diagnosis — Meridian Commercial Credit --- ### 1. Situation Summary - **Who:** 42 relationship managers and 6 team leads in Meridian Regional Bank's Commercial Credit division, overseen by division head Priya S. - **What tool:** Lumen Copilot, a GenAI drafting assistant used to produce credit memos. - **What behavior:** Adoption collapsed from 18% to 9% by week 10 against a 60% target; tool is being actively avoided after early negative signals from leadership and compliance. --- ### 2. Surface Complaints vs. Root Causes **Surface Complaints** - The drafts require too much editing — "80% there" isn't good enough - Output doesn't match house style - Compliance attestation requirements are unclear **Likely Root Causes** - Accountability for AI-drafted memos is undefined. Marcus T. named it directly: nobody has said whose neck is on the line if a deal sours. That ambiguity is rational deterrence, not technophobia. - Two RMs were publicly embarrassed in committee. That story has traveled. The tool now carries reputational risk. - Priya's visible non-use is a green light to opt out. People watch what leadership does, not what they say in town halls. The Wealth Advisory contrast confirms this causally. - Compliance created friction with no resolution path. An attestation requirement with no template is a compliance trap, not a compliance process. - The comp plan creates zero positive incentive and the bonus cycle closes in 8 weeks — RMs are in execution mode on deals, not tool experimentation. --- ### 3. Dominant

…output truncated at 250 words.

Step 2 output

# Change Management Brief — Lumen Copilot Recovery Prepared for: Priya S., Division Head, Commercial Credit Date: [Insert] | Status: Action Required --- ## Situation Lumen Copilot adoption has fallen to 9% — not because the tool is broken, but because your team has rational reasons to avoid it. Two RMs got burned in committee, nobody knows who owns a bad AI-drafted memo, and the bonus clock is ticking. --- ## Root Cause — Primary Accountability ambiguity. When Marcus T. asked "whose neck is on the line if a deal goes bad," nobody answered him. That silence is the adoption killer. RMs are not being irrational — they are correctly identifying that the accountability structure around AI-drafted memos is undefined. Until ownership is explicit and documented, using the tool is a career risk with no compensating upside. Everything else is secondary to fixing this. --- ## Root Cause — Contributing - Visible leadership gap: You haven't used the tool publicly. The Wealth Advisory site hit 55% adoption because their division head showed her own use case on day one. Your team is taking cues from your behavior, not your town hall remarks. - Compliance left a live wire: The week-6 attestation email created a requirement with no process. It reads as a trap, not a guardrail. - Bonus cycle timing: RMs are eight weeks from close. Experimentation is a luxury they don't feel they have right now. --- ## What Won't Work - Another training session. Proficiency is not the

…output truncated at 250 words.

Step 3 output

CRO Reading — Skeptical Review --- ### 1. Three Questions the Brief Doesn't Answer **"Who actually authorized the one-paragraph accountability policy?"** The brief tells Priya to draft accountability language with Legal/Compliance in five days. That's fine as a starting point. But credit memo accountability at a regulated bank isn't a paragraph Priya can write on a Tuesday — it touches model risk governance, OCC guidance on AI in credit decisions, and potentially your SR 11-7 model risk framework. Has Model Risk Management been in the room at all? If that policy gets written without them and a deal later goes bad, the one-paragraph fix becomes Exhibit A in a regulatory conversation nobody wants. **"What exactly is Northwind's contractual liability for output errors?"** Marcus T.'s question — whose neck — applies externally too. The brief assumes accountability sits cleanly with the RM and team lead. But if Northwind's output is factually wrong and that error survives attestation and contributes to a credit loss, what does the vendor contract actually say? The brief recommends closing the internal governance gap without confirming the external one exists or has been reviewed. **"Why did the Compliance attestation email go out without an accompanying process?"** Someone in Compliance sent a week-6 email creating a requirement they couldn't operationalize. That's a controls breakdown, not a communications hiccup. Before Priya convenes a five-day fix, I want to know whether Compliance assessed the tool at all before deployment — and if they did, why the attestation framework wasn't

…output truncated at 250 words.