What “good” looks like

Before & after

Every pair below is the same person asking the same question. The only difference is whether their AI was configured using the setup guide or not. This is the gap you’re closing.

1 · Summarizing a deal

You: “Summarize this CIM for me.”

Default AI

“This Confidential Information Memorandum presents an investment opportunity in a mid-market company. The company operates in the healthcare services sector and has demonstrated consistent revenue growth over the past five years. Key highlights include strong EBITDA margins, a diversified customer base, and an experienced management team…”

Generic. Could describe any company. You’d redo the whole thing.

Configured AI

“This is a $45M revenue healthcare staffing company, PE-backed by [Sponsor] since 2022, seeking a $120M first-lien term loan to refinance and fund a tuck-in. Three things that matter for your review:

1. Revenue concentration — top 3 hospital systems are 62% of revenue. Lose one to an in-house model and you lose 20%+ overnight.
2. EBITDA adjustments — $8M of add-backs on a $14M base. 57% of adjusted EBITDA is ‘run-rate synergies’ and ‘one-time integration costs.’ Your committee will push hard.
3. Cash flow — capex-light, but working-capital swings are real on 60–90 day hospital collections. DSC looks fine on paper, tightens fast in a slowdown.

Want me to build the risk section, or dig into the adjustments first?”

It knows your role, your firm’s standards, and what you’ll get asked. It’s already on your next step.


2 · Writing an email

You: “Write a follow-up email to a potential client.”

Default AI

“Subject: Following Up on Our Recent Conversation

Dear [Name], I hope this email finds you well. I wanted to follow up on our recent conversation regarding potential collaboration opportunities. I believe there are several synergies between our organizations that could be mutually beneficial…”

Straight to the trash folder. Nobody reads this.

Configured AI

“Subject: Quick follow-up — next steps from Tuesday

Hi [Name], good meeting you on Tuesday. Wanted to keep this simple — you mentioned the board conversation is coming up. If it’d help, I can put together a one-page summary of the key terms we discussed that you could share with them. Sometimes that makes the internal conversation easier.

Either way, let me know how it goes. Happy to adjust anything based on their feedback. Best, [You]”

Short. Useful. Gives him something to work with. Doesn’t sound like a robot.


3 · Stress-testing your own work

You: “I think this company’s margins are sustainable. What do you think?”

Default AI

“That’s a reasonable assessment. The company has demonstrated consistent margin performance over the past several years, which suggests sustainability. Their market position and operational efficiency support the view that margins can be maintained…”

Worthless. It just agreed with you.

Configured AI

“I’d push back. Three things bother me:

1. Labor costs — 68% of COGS is labor, in a market at 3.2% unemployment. Wage pressure is real and they have no pricing power: top 3 clients are on fixed-rate contracts through 2027.
2. The 2024 expansion — margins jumped 300bps in one year, tracking to a one-time contract win, not operations. Underwrite to TTM and you underwrite to the peak.
3. The add-backs — strip the ‘non-recurring’ costs that have recurred three years running and normalized margins are ~16%, not 20%.

What margin are you underwriting in your base case? I can stress it against the covenant package.”

It found the holes — before you walk into committee.


4 · Learning something new

You: “Explain covenant-lite structures to me.”

Default AI

“Covenant-lite, or ‘cov-lite,’ refers to loan agreements that lack traditional maintenance covenants. Instead of requiring the borrower to maintain certain financial ratios on an ongoing basis, cov-lite loans typically only include incurrence covenants that are tested when the borrower takes a specific action…”

A textbook answer. Fine for an exam. Useless on a live deal.

Configured AI

“In your world, cov-lite means lending with fewer tripwires.

Normal loan: borrower must prove every quarter that leverage is below X and coverage above Y. Miss, and you have a seat at the table to renegotiate or accelerate.
Cov-lite: those quarterly tests don’t exist. They’re only tested when they do something — more debt, a big acquisition, a dividend. If they’re just slowly deteriorating, you don’t find out until it’s too late.

Why it matters: your downside protection is weaker; you’re betting on the projections with fewer off-ramps. Your IC will want to know why you’re comfortable.
The question to ask: what’s your early-warning system if this underperforms? The covenant package won’t tell you.”

Same concept, framed for your actual job and the decisions you make.


The pattern

1 · Knows your context

It skips the generic stuff because it already knows who you are.

2 · Writes for your audience

For your IC, not for a textbook.

3 · Challenges your thinking

It looks for the holes instead of validating you.

4 · Suggests the next step

Because it knows what your workflow looks like.

This is what the five rules and the setup create. Not better prompts. Better infrastructure.

Do this today

Run example 3 on a thesis you actually hold right now. Tell your configured AI: “Tear apart my view on [deal]. Find the three weakest points before my IC does.” See if it surfaces something you hadn’t considered. It usually will.