AI Operating SystemPrivate Credit
End-to-end, with exact prompts

Real workflows

Pull these up when you have a real deal or portfolio task in front of you. These aren’t before/after snippets — they’re full sequences. You drive; the AI absorbs everything around the work so that when you write, every piece of context is at your fingertips. The example deals are fake; the structure is exactly what you’d run tomorrow.

The principle

Use AI as a reasoning partner, not a ghostwriter. It does the slow synthesis. You keep the judgment — and the judgment is what your job is.

From CIM to IC memo

A teaser lands. Then a CIM, a model, management calls, notes, and finally a draft memo. It used to take a week of grinding — 25 to 30 hours. With AI as your reasoning partner it takes about four hours of focused work across two days, and the result is better, not worse. Written for Copilot with Notebooks (where most firms live); Claude/ChatGPT notes where it matters.

The setup 15 min · once per deal

Build the workspace before you write a single prompt. Create a Copilot Notebook named for the deal — Project Aurora — and add every document you have: teaser, CIM, preliminary model, management deck, sector research, your call notes. Add a context page with your About-me block plus three deal-specific lines:

  • My role on this deal: [originator / lead / supporting analyst].
  • My PM is [name] and cares most about [the two or three things they consistently flag].
  • Our hold size would be ~[$X]M in [unitranche / first lien / second lien].

Run the Overview and read it. It’s not your work product — it’s a first pass at what the materials collectively say. Note anything that surprises you; that’s where the interesting questions live. (No Notebooks? Open one chat and keep it open across the steps — it’s just less powerful.)

Step 1 — The 20-minute read 20 min

Come out knowing what the deal is, in your own words, before spending real time.

1.1 · Plain English

Read the CIM and the teaser and tell me, in plain English: what does this company actually do, who pays them, and how do they make money? Skip the marketing language. If their description uses "platform" or "solution," translate it to what a normal person would understand.

1.2 · The thesis, de-spun

What is the sponsor's growth thesis, in three sentences? Strip out the optimistic adjectives.

1.3 · Five curiosities

Tell me the five things about this business I should be most curious about after one read of the CIM. Not risks yet — just things that don't quite add up or that I'd want to know more about.

Save the output as 01 — The Deal in Plain English. You’ll return to it when writing the memo.

Step 2 — Quantitative hunt 30 min

Make Copilot do the slow, boring part: pulling every number that matters into tables you can compare.

2.1 · Market claims

Pull every quantitative claim about market size, growth rate, or market share in this CIM. List each with the page number, the source it cites (or "no source"), and your read on whether the claim looks defensible. Format as a table.

2.2 · Customer data

Now pull every customer-related number: total customer count, top-customer percentage of revenue, net retention, churn, average contract length — whatever is mentioned. Same table format. Page numbers required.

2.3 · EBITDA bridge

Find the EBITDA bridge. List every adjustment from reported to adjusted EBITDA. For each: the dollar amount, the rationale the CIM provides, and your assessment of how it survives credit-committee scrutiny (defensible / arguable / aggressive / not going to fly). Order from most defensible to least.
Verify three numbers yourself

Go back to the CIM and confirm three of the figures against the page reference — not because Copilot is unreliable, but because you need to be the one who’s touched the numbers when your IC asks.

Step 3 — Walk through the model 45 min Excel Copilot

Open the sponsor’s preliminary model and click the Copilot button.

3.1–3.3 · Structure, drivers, margins

Walk me through this model: the tabs, what each does, and the four or five assumptions that drive everything else. Then: what's the base-case revenue growth assumption, where is it set, and what's the underlying logic — if it isn't documented, say so. Then: are EBITDA margins expanding, flat, or compressing across the projection, by how much, and where is the assumption that drives it?

3.4 · Stress test

Stress test: drop revenue growth by 200bps in years 2–5 and EBITDA margins by another 100bps in the same period. Tell me what happens to total leverage at year 3, fixed-charge coverage at year 3, and maintenance-covenant headroom in each year. Build a new tab with the output and walk me through what you did.

3.5 · One structural flaw

If you had to flag one structural issue with how this model is built — something that should make me trust the output less — what would it be?

No Excel Copilot? Identify the key assumptions manually, then describe them in Claude/ChatGPT and run the stress logic in text. Slower; same thinking.

Step 4 — The risk inventory 45 min

Now the work that justifies your role. The AI generates the inventory; you decide what matters.

4.1 · Ten risks, ranked

Based on everything in this notebook — CIM, model, my call notes, the sector research — give me the ten biggest risks to this credit, ordered most material to least. For each: a one-sentence description, the evidence from the materials (with references), and what mitigation the sponsor currently offers, if any.

4.2 · The skeptical PM

Now be more aggressive. For each of the top five risks, tell me what the most skeptical PM at my firm would say — not the polite version, the version where they've seen this play out badly three times.

4.3 · What’s missing

What's missing from this risk list? What should I worry about that isn't in the materials at all because the sponsor wouldn't volunteer it?
4.4 · Your judgment — no prompt

Pick the three risks you’ll lead with. Criteria: materiality if it goes wrong, how hard it is to mitigate, how clearly your IC will care. Write the three down with one sentence each on why it made the cut. Save as 04 — Risk Lead.

Step 5 — Management meeting question set 20 min

5.1 · Question set on the lead risks

Based on the three lead risks I just identified, build a management-meeting question set. For each risk: (a) the framing question I'd open with, (b) two or three follow-ups to push if their first answer is generic, and (c) one question saved for the end to test whether they're being straight. Then add five questions about the business itself, not the risks.

Save as 05 — Management Q Set. One of the highest-leverage uses of AI in the whole workflow.

Step 6 — After the management call 30 min

You ran the call and took notes (Teams Copilot transcribed them — even better). Now process what you heard.

6.1–6.4 · Summarize, grade, update, follow up

Here are my notes from the management call. (1) Summarize the call in five bullets — what did they say about each of my three lead risks? (2) Where did management give a sharp, well-evidenced answer, and where did they fall back on platitudes? Be specific about which felt strong and which felt soft. (3) Did anything change my read on the deal, and how would I update the risk list? (4) Draft the three follow-up questions I should send the sponsor's deal team in the next 24 hours.

Save as 06 — Post-Mgmt Read.

Step 7 — First draft of the memo 45 min

By now you’ve done four hours of structured work and the notebook is rich with context. This is why the draft is dramatically better than a cold prompt.

7.1 · Draft the memo

Using everything in this notebook — the deal description, the quant hunt, the stress-test output, the risk lead, the call notes — draft a first-pass IC memo. Structure: (1) the deal in two sentences; (2) why we like it — three bullets, each one specific reason, no marketing language, no "compelling," no "differentiated"; (3) the three biggest risks, each with the evidence and how we get comfortable; (4) what we're still trying to figure out — two or three open questions for confirmatory diligence; (5) recommendation — one direct paragraph. Match my firm's tone: direct, analytical, no hedging, no corporate adjectives. If my notes contain a sponsor phrase that sounds like marketing, don't put it in.

7.2–7.3 · Self-critique & rewrite

Now critique the draft you just wrote. Where is it weakest, and where will I get pushed back on if I send this to my PM tomorrow morning? Then rewrite the two weakest sections based on your own critique.

Step 8 — Your edit 60 min

The AI got you 60% there. The remaining 40% is the part that makes the memo yours and demonstrates why you have the job.

  1. Read it out loud. Any sentence that could have been written by anyone — rewrite it.
  2. Check every number against source. Trust no figure you haven’t personally verified at least once.
  3. Add the one or two things only you know — a texture from the call, a read on the sponsor, the thing your PM mentioned offhand. This is your value-add.
  4. Cut the hedging. Kill “potentially,” “may,” “could be,” “we believe.” Be willing to be wrong on the record.
  5. One final adversarial pass. “You are the most skeptical PM at my firm. Read this and tell me the three questions you’d ask first.” Add the answers if they’re not already there.
What you just did

A memo presentable to your IC tomorrow, that you can defend line by line, produced in roughly four hours instead of a week. First time it feels slow and weird. Second time, obvious. Third time, it’s the way you were always supposed to work.

Do this today

Don’t run a whole workflow cold. Pick the next real task on your desk — a CIM that just landed, or a portfolio name due for review — and run just the first two or three steps. Build the workspace, get the plain-English read, do the quant hunt. You’ll feel the difference before you’re halfway through.

← BackThe first 30 days