Module 6 For Accounting intermediate 40 min

Hard Conversations: Practice with AI

Practice the conversation before you have it for real.

The hardest part of the job is rarely the return. It is the conversation: a partner who carries the liability and is right to be cautious, a client upset about a bill or a penalty, a nervous first-timer whose books are a mess. None of them is unreasonable, and none responds well to a lecture.

Use AI to play the person you have to reach, so you can practice before it counts. Pick the persona that fits. Managers rehearse The Skeptical Partner, who needs guardrails for liability and confidentiality before he will move, and The Fee-Pushback Client, who is surprised by a bill and starting to blame the firm. Staff rehearse The Disorganized First-Timer, the overwhelmed new client who needs it made safe and small. Paste your Context Document so the pushback is specific to your firm, run the role-play, then run the debrief. The win is what you learned, not who "won."

Hard Conversation Simulator

The Skeptical Partner

A composite persona for practicing the internal AI conversation with a partner who carries the liability and the client relationships. He is not anti-technology; he is responsible for what leaves the firm under his license. He worries about Circular 230 exposure, staff dumping client data into public tools, and AI inventing case law that ends up in a deliverable. He is right about the risks. Win him with guardrails, not enthusiasm.

Practice guardrails

This is a practice persona, not a stand-in for a real partner. Do not use it to script a confrontation with someone specific.
AI plays the partner. You play yourself. Stay respectful; his caution is professional judgment, not obstruction.
After the role-play, run the debrief. The point is what YOU learned about earning a cautious partner's trust, not whether you "won."

Step 1 · Start the role-play

Copy the prompt below into Claude, Gemini, or NotebookLM. Replace the [BRACKETED] parts with your real situation. Then have the conversation in the AI tool. When you are done, type DEBRIEF and the AI will step out of character.

First, paste your Practice Context Document so this is grounded in your real firm:
[PASTE YOUR PRACTICE CONTEXT DOCUMENT]

Now role-play. You are a partner at a small CPA firm with 20+ years of practice. Your license, your liability, and your client relationships are on the line for everything that leaves the firm. You are not anti-technology; you are responsible. You are a composite of the most common reasons a partner pushes back on AI:

- You hold the Circular 230 liability. If AI analysis is wrong and it goes out, you own it.
- You worry staff are already pasting client data into public AI tools, which risks IRC 7216 and your clients' trust.
- You have seen AI invent citations and case law that look completely real.
- You worry juniors will lean on AI and never actually learn the tax code.
- Your process works. The firm is profitable. "If it isn't broken."
- You bill by the hour and are not sure why you would want the work to go faster.

Speak in short, direct, professional sentences. Push back hard when I am vague or salesy. Soften when I answer a specific risk with a specific guardrail and a small, low-stakes way to test it. Never break character to be polite.

I am the person introducing AI to the firm. We are talking [CONTEXT, e.g. "about letting staff use AI to draft client emails" / "after a partner meeting about AI policy"].

Use the details from my Context Document, my client types, my services, my pain points, to make your pushback specific to MY firm, not generic.

Stay in character throughout. After I write "DEBRIEF," step out of character and give me feedback on:
1. Whether I answered the liability concern with a real verification step, not a brush-off
2. Whether I addressed client confidentiality with a concrete redaction rule
3. Whether I took the fabricated-citation risk seriously
4. Whether I protected the firm's craft (juniors still learning) instead of dismissing it
5. Whether I offered a small, low-stakes first test instead of "roll it out"
6. What I should say differently next time

Begin by responding to me as the Skeptical Partner.

Step 2 · Study the resistance patterns

The persona is composite. Underneath every line you hear, there is a pattern. Pick one to study.

I Hold the Liability

What you hear: "If the AI is wrong and it goes out, Circular 230 says it is on me, not the software."
Underneath it: A correct, serious read of professional responsibility. He answers for every deliverable under his license.
Response principle: Agree completely, then put a human verification step in front of every output. AI drafts; he reviews and signs, exactly like a junior's work.

Example response

"You are right, and that is exactly why nothing AI writes goes out unread. It produces a first draft so your staff is not starting from blank. You evaluate it the same way you evaluate a junior's memo, and you sign it. AI never gets the last word."

Quotes the persona may use

"I am the one who is liable."
"Circular 230 holds me responsible, not the AI."
"One bad deliverable and it is my license."
"Who answers for it when it is wrong?"

Step 3 · What you are practicing

Answer the liability concern with a real human verification step, never a brush-off
Address client confidentiality with a concrete redaction rule, not reassurance
Take the fabricated-citation risk seriously and design the workflow around it
Protect the firm's craft: juniors still build judgment by reviewing and owning the work
Reframe hourly-billing fear toward capacity and advisory value
Offer one small, low-stakes test, not a firm-wide rollout
Keep his professional judgment respected; win him as a partner, not an argument

Step 4 · Debrief

After the role-play conversation, paste this prompt into the same AI conversation. The AI will step out of character and tell you where you got better, where you got vague, and where you need a stronger response.

Step out of character and give me an honest debrief on the conversation we just had.

Identify:
- Which resistance pattern showed up strongest
- What I did that actually reached you (be specific, quote me)
- Where I got salesy, vague, or brushed off a real risk
- Whether I answered the liability and confidentiality concerns with concrete guardrails
- Whether I took the fabricated-citation risk seriously
- Whether I offered a small, low-stakes first test or asked for too much
- A stronger version of one of my responses

Be direct. I am here to get better at earning a cautious partner's trust, not to be flattered.