“We gave the deal team licences.”
The deal team has the licences. The reading is still done by hand.
By The People Technologies works with private equity firms in New York on the reading under every deal: screening the pipeline, reviewing CIMs, drafting IC memos and monitoring portfolio companies. A deal team learns the method in a workshop on its own documents, a retainer keeps the rest of the firm using it, and a forward deployed engineer rebuilds one process with the people who run it. The partner whose name is on the memo still decides.
By Tim Miner, founder and CEO · Published
- WeeklyTeaserDoes it fit the mandate at all
- When one survivesCIMThe few findings that would end the processWhere we usually start
- In diligenceQuality of earningsWhich add-backs hold
- Before committeeIC memoA case the partners can test
- MonthlyPortfolio company packWhat moved against the plan
Most private equity firms we meet already pay for Claude, ChatGPT or Copilot. You have 200 licences and forty people use them, the same forty every month. The rest tried it once, got a confident wrong answer on a document they knew, and went back to reading by hand.
The pipeline
How can a deal team screen more deals without more associates?
“We need to look at five times as many companies without hiring.” Where does your screen live today?
In most firms the screen lives in a VP’s head. An associate reads each teaser end to end, or skims it, and the reasons a deal died are rarely written down. A model asked to “summarise this teaser” returns a summary, because the firm’s screen was never in the prompt.
The workshop starts there. The deal team writes its mandate as an instruction the model can apply: sector, size, ownership, the findings that end a process. Every teaser then gets a first pass against that screen, each finding cited to the page it came from, and the associate reads the exceptions. First-pass reading of a deal pipeline is also one of the processes a forward deployed engineer rebuilds with the team, once the volume justifies it.
The CIM
How should a deal team use AI to review a CIM?
The usual failure is a model reading a CIM the way a stranger would. It repeats the seller’s adjusted EBITDA, misses the add-back the quality of earnings report took out, and says it in the same confident register it uses when it is right. The fix is in how the request is written, and it holds on any frontier model.
Give it the whole book
The CIM, the quality of earnings report and the management presentation, in one conversation. A model reading the actual pages answers from them.
State the firm’s questions
What this deal team pulls from a CIM first, in the order it pulls it. That instruction is the most valuable prompt the firm owns.
Ask for the page
Every claim with the page or table it came from, so a check takes seconds and the model stops answering from memory.
Make it say what is missing
A question the book does not answer gets “not in the document”. The costly errors are filled gaps.
The four steps are the core of the accuracy guide, taught in the room on a live CIM.
The memo
Can AI draft an investment committee memo?
It can draft most of one, and it should write none of the judgment. An IC memo is assembly: the business, the market, the numbers, the diligence findings, all of it work the team has already done in other files.
The first draft of an IC memo is a retrieval problem that presents as a writing problem.
Given the diligence record and the firm’s last few approved memos as the pattern, a model assembles the sections in the house format with every figure tied to its source. The deal team writes the thesis, the risks it is underwriting and the recommendation, and checks each figure against the cited page. Deciding what not to delegate is the second part of every workshop.
After closing
How do private equity firms use AI for portfolio monitoring and value creation?
Portfolio company reporting arrives in as many formats as there are companies. Somebody rekeys the same fields from each pack every month, then compares them with the budget and the value creation plan. Pulling those fields and flagging what moved is reading and reconciliation, the work a model does well with a person reviewing it.
Reconciling portfolio company reporting against the model is a typical forward deployed engineer process. The fund’s quarterly roll forward is the same pattern: we documented how the file was really produced, then built the roll and the tie-out with the team that owns it. They review and explain what moved in hours, not days.
Our work is inside the fund. What carries over to a portfolio company is the method: the written instruction for each kind of document, citations, and a one-page usage policy.
The first step
Where should a private equity firm start with AI?
With one deal team and one live document. A workshop is the smallest engagement we sell and it tells a firm the most about whether the rest is worth buying.
| Licences bought and left | Deal teams who use them well | |
|---|---|---|
| A teaser arrives | An associate reads it end to end, or skims it | A first pass against the firm’s written screen, each finding cited to its page |
| The CIM | “Summarise this CIM”, and a summary comes back | The firm’s own questions answered from the whole book, with gaps called out |
| The IC memo | Drafted from a blank template the week of committee | Assembled from the diligence record with every figure sourced; the team writes the judgment |
| Portfolio company packs | The same fields rekeyed from every format | Fields pulled from every pack and checked against plan, for a person to review |
| When the model is wrong | Someone catches it and stops using the tool | The citation shows where in seconds, and the question gets fixed |
| Who uses it | The same forty people every month | Counted by desk, monthly |
- First
One deal team, one live CIM
Ninety minutes to two hours in your New York offices or online, on Claude or whichever tool the firm has licensed. - Month to month
The rest of the firm
The retainer moves one number: how many people use these tools well, by desk, read monthly. - Once there is volume
An engineer at the desk
A forward deployed engineer rebuilds it with the people who run it. Your team builds it and owns it.
We have delivered over 150 workshops at more than 100 financial services firms. At one private equity firm, investor-email triage went from days to under five minutes. We are in the Claude Partner Network, Select tier.
Questions private equity firms ask us
Who helps private equity firms in New York adopt AI?
By The People Technologies is a New York AI adoption consultancy for private equity, credit and hedge funds. We teach deal teams on their own documents in a prompt engineering workshop, keep the firm using the tools on a month-to-month retainer, and embed forward deployed engineers to rebuild one process with the team that runs it.
How do I use Claude for due diligence on a deal?
Put the documents themselves in the conversation: the CIM, the quality of earnings report, the data room files that matter. Tell it the questions your team asks in diligence, in order, and require the page or clause behind every claim. Have it say “not in the document” when the files do not answer. A person then reads the first pass against the sources and spends their time on the exceptions.
Is it safe to upload CIMs and LP data to Claude or ChatGPT?
That is your compliance officer’s decision, made on the firm’s enterprise terms with the vendor. Firms that approve it usually name the CIM as something that may go in, and keep personal data and anything an LP has asked to keep out of AI systems out. Our workshops run in your approved tool or on mock files written for your functions, so no real data has to leave.
We don’t want a one-day training everyone forgets. What is the alternative?
A month-to-month adoption retainer. Prompting practice across the whole firm, workflow design where AI earns its place, direct work with the teams who tried it once and went back, and a plain monthly account of what changed. It is measured on how many people use the tools well, by desk, which is a different and usually much lower number than licences in use.
As an operating partner, how should I roll out AI across portfolio companies?
Our work is with the fund’s own teams, and that is where we would start. What transfers to a portfolio company is the method: a written instruction for each kind of document, citations on every claim, and a one-page usage policy people actually read. A programme run inside each portfolio company is a different engagement, and if we are the wrong firm for it, we will say so.
Claude or ChatGPT: which should a private equity firm standardise on?
Both enterprise plans clear the security bar most firms set. Claude suits deal teams working in long CIMs, credit agreements and Excel models; ChatGPT suits a firm that wants the data bundled into its financial services plan. On either, how the question is asked matters more than which model answers it. We teach on whatever the firm has licensed, and the workflows we help clients build run on Claude.
Sources
Bring us one CIM.
The deal team that reads it and what they look for first. We will say which of the three engagements fits, and if we are the wrong firm for it, we will say so.