AI for private equity
AI for private equity firms
Deal intel, diligence, and investor reporting that runs itself. 5x ROI in 30 days, or we work free.
- Buyout
- Growth equity
- Fund of funds
- Family office
- Secondaries
Teams we build for
- Hoyes Michalos
- Nurse Next Door
- Fedi
- UBC Sauder
- Merchant House Capital
- Picton Investments
- Campbell Froh May & Rice LLP
- Barnakl
- Hungerford
- Breez
Teams we build for
- Hoyes Michalos
- Nurse Next Door
- Fedi
- UBC Sauder
- Merchant House Capital
- Picton Investments
- Campbell Froh May & Rice LLP
- Barnakl
- Hungerford
- Breez
Teams we build for
- Hoyes Michalos
- Nurse Next Door
- Fedi
- UBC Sauder
- Merchant House Capital
- Picton Investments
- Campbell Froh May & Rice LLP
- Barnakl
- Hungerford
- Breez
Deal intel prep eats the most expensive hours in the building.
Comps, filings and news pulled by hand before every committee meeting, by people paid for judgment. The reading is the cost. The judgment was never the slow part.
Investor reporting is a quarterly fire drill.
Senior people formatting LP updates at quarter-end instead of working the portfolio. The LPA already says what may be disclosed. Nobody has time to check every page against it.
Diligence findings die in inboxes.
The firm has looked at this space before. What it learned lives in one principal's sent folder. Every new deal re-learns it at full price.
Portco reporting packs land monthly and get skimmed.
A covenant drifting at one portfolio company surfaces at the board meeting, not the week it moved. Monitoring is rationed by attention, and attention follows the live deal.
Every DDQ starts from a blank page.
The answers exist in the last fundraise's responses and the fund documents. Rebuilding them by hand is the tax on every new LP conversation.
Private equity, before and after
The manual path is dashed: Analysts pull comps by hand, LP updates built at quarter-end, Findings buried in inboxes. The system path replaces it, and a person approves before anything ships: Committee briefs, every line cited, Reporting assembles itself, Every past deal searchable.
Before: by hand
- 01Analysts pull comps by handhuman
- 02LP updates built at quarter-endhuman
- 03Findings buried in inboxeshuman
After: the system
- 01Committee briefs, every line cited
- 02Reporting assembles itself
- 03Your approvalhuman
- 04Every past deal searchable
The research
88% of agentic AI pilots never reach production. Scope and a measurement gate are what separate the other 12% (IDC, 2026).
IDC · 2026
Proof from the pattern
Our closest published work is an outbound and research engine built for a private capital firm. The mechanics below are what we would build and measure with you, against a baseline you sign before the build.
We guarantee 5x ROI inside 30 days of deployment, in writing, measured against a baseline you sign before we build. If the system misses the bar, we keep working for free until it clears.
The numbers in private equity firms
88% of agentic AI pilots never reach production. Scope and a measurement gate are what separate the other 12% (IDC, 2026).
IDC · 2026
89% of midsize businesses plan to put AI to work in 2026 (JPMorgan Chase, 2026).
JPMorgan Chase · 2026 · 1,469 business leaders, November 2025
95% of GenAI pilots show no measurable P&L return. The winners pick use cases by P&L impact, not novelty (MIT, 2025).
MIT NANDA · 2025 · 150 interviews, 350-employee survey, 300 deployments analyzed
Who this is built for
Your calendar goes back to founders and bankers.
The scarcest thing in the firm is not capital, it is your attention. Every hour spent assembling a brief is an hour not spent with a management team, and those are the hours that move deals. The committee still decides. You still own every position. The assembly work underneath the judgment is what moves.
The LP letter becomes a review, not a weekend.
Quarter-end arrives on a schedule, and it still lands like a surprise. The reporting agent assembles the letter from the fund admin extract, holds it to the LPA's disclosure rules and your side letters, and hands you a draft with every number traced. You read, correct, and approve. No page reaches an LP without you.
The data room gets a first reader that is not you.
The CIM lands Thursday night. By morning there is a first-pass summary with page citations, the MAC clause pulled, the working capital peg flagged. Your day goes to the sceptical reading, testing the management case and defending your lines in front of a partner. Folder work stops being the job.
You approve the answer once, not every quarter.
An LP asks about a capital call or a management fee offset. The answer already sits in the fund admin extract with the calculation shown, and you approve it once instead of rebuilding it every time. Nothing ever writes back into the fund admin system. That boundary is yours to quote in the next operational due diligence.
What stays human
- The ask
- The line check
- Partner sign-off
The steps the day below leaves to a person, by design.
IC Monday, as the deal team associate lives it
The feeds were read overnight.
Market feeds, filings and news on the target's space were ingested and indexed before anyone logged in. The reading is already done when the question arrives. That is what makes a same-morning brief possible.
The ask.
humanThe associate requests a committee brief on the target and names the deal team. The request runs inside that team's wall. No other team can see that the deal exists.
The brief assembles.
Fixed committee sections: market view, competitive landscape, risk flags, sources. Every line cites the data room page or the feed it came from. A claim that cannot be traced is dropped, not shipped.
The archive answers first.
What did the firm learn the last time it looked at this space. Past diligence findings surface, scoped to what this team is allowed to see. The deal starts from the firm's memory instead of from zero.
The line check.
humanThe associate verifies the lines the decision will turn on. A stale comp gets flagged, the section rebuilds from the source, and the citation trail shows the fix.
Partner sign-off.
humanThe deal partner reads, challenges, and approves the memo. Nothing reaches the investment committee unreviewed. This gate is the product, not a formality.
The findings file back.
The brief, the committee's questions and the answers land in the deal archive. The next deal in this space starts warmer than this one did.
The economics
Before and after economics
Line
- Intel prep per deal
Before
comps, filings and news pulled by hand before every committee
After
a cited committee brief, assembled from the indexed feeds and checked line by line
- Partner week
Before
assembly hours spent on briefs and letters
After
the same hours on founders, bankers and management teams
- Coverage
Before
rationed by whoever had research capacity that week
After
every deal, one standard of brief
- Assembly spend, modelled
Before
$8,640 a week across a six person deal team
After
measured against a baseline you sign before the build
The first three rows are qualitative and carry no figure on purpose. We publish a number for this sector when a client signs off on one, measured against a baseline agreed before the build. The modelled row multiplies this page's calculator defaults, a team of 6 at 12 manual hours a week each at a $120 loaded hourly cost.
What we build for private equity firms
The deal desk
Deal intel briefings on demand.
One ask, one committee-ready brief, every line cited to its source.
Due diligence document review.
First-pass reads of data rooms and CIMs, findings in one place instead of forty inboxes.
Deal sourcing and origination outreach.
The outbound engine pointed at proprietary deal flow, feeding the pipeline your DealCloud already tracks.
The fund office
LP and investor reporting.
Quarterly letters assembled from your systems. Partners review instead of write.
Portfolio company monitoring.
Metrics and risk signals from every portco, whether they arrive through iLevel, Chronograph or a PDF board pack.
LP DDQ and fundraise responses.
Answers assembled from the fund documents and the last fundraise's responses, each one cited, IR approves every send.
The firm's memory
Knowledge brain for past deals.
Every diligence finding the firm has ever produced, searchable, so deals stop re-learning.
Where the data comes from
Data room
The CIMs, QoE reports, management presentations, legal documents and diligence Q&A threads for each live deal, whether the room runs on Datasite or Intralinks. The system reads it first and returns summaries with page citations, so the team opens folders to check lines, not to find them.
Where it stops. Read only. The system quotes the data room and never writes to it, and one deal team's room is invisible to every other team by construction.
Fund admin
Capital accounts, calls and distributions, NAV and the LP register, whether that lives in Allvue or eFront or with an outsourced administrator like SS&C or Citco. Reporting drafts assemble from an extract of it, with every number traced back to its line.
Where it stops. We will not let an agent write to your fund admin system. Numbers flow out of it into drafts a person approves. Nothing flows the other way.
Market feeds
Filings, news, transaction comps and fund benchmarks from the services the desk already pays for, PitchBook, Preqin, Capital IQ. The feeds merge with the firm's own view into one cited account of what matters.
Where it stops. The merge runs one way. Public data flows in. Nothing proprietary flows out to a feed vendor, and nothing trains anyone's model.
How an answer is found in your deal archive
A question runs against your deal archive two ways at once: vector search for meaning and full-text search for exact wording. Reciprocal rank fusion merges both result sets, and the answer carries the source it came from.
- 01A question
- 02Vector search (meaning)
- 03Full-text search (exact wording)
- 04Rank fusion
- 05Answer, with its source
Built around your rules
| Regime | What it demands here | How the system complies |
|---|---|---|
| Confidentiality and NDAs | Deal data is seen only by the people under the NDA, per deal, for as long as the obligation runs. | Access scoped per deal team at the database layer, data stays in your accounts, and nothing is used to train models. |
| Material non-public information | Information walls between deal teams that hold even when a person makes a mistake, and a record of who saw what. | Walls enforced by row-level security by construction, not by policy, with audit logs on every access. |
| LP confidentiality and the LPA | Disclosure follows the LPA and the side letters. Nothing reaches an LP that the fund documents do not permit. | Reporting agents work from your LPA's disclosure rules, and a human approves every external document before it leaves. |
| PIPEDA | Personal information at a fund means the LP register and the deal team's own records, handled under Canadian privacy law with residency and retention answered up front. | The LP register stays inside your fund admin extract, deal-team records stay in your accounts, and the build is designed around PIPEDA from the start with data residency scoped per engagement. |
| NI 31-103 registrant obligations | Where the manager is registered with a Canadian securities regulator. Books and records kept and producible, client information held confidential. | Everything runs in your accounts, so the records the regime demands accumulate where your records already live, with audit logs attached. |
| Vendor posture | Vendors who can survive the operational due diligence questionnaire your LPs will send. | Our core vendors (Anthropic, Supabase, Vercel) publish SOC 2 reports on their trust pages. We design within that posture and claim no certification we do not hold. |
The regimes that govern private equity, what each demands, and how the system complies
Regime
- Confidentiality and NDAs
What it demands here
Deal data is seen only by the people under the NDA, per deal, for as long as the obligation runs.
How the system complies
Access scoped per deal team at the database layer, data stays in your accounts, and nothing is used to train models.
- Material non-public information
What it demands here
Information walls between deal teams that hold even when a person makes a mistake, and a record of who saw what.
How the system complies
Walls enforced by row-level security by construction, not by policy, with audit logs on every access.
- LP confidentiality and the LPA
What it demands here
Disclosure follows the LPA and the side letters. Nothing reaches an LP that the fund documents do not permit.
How the system complies
Reporting agents work from your LPA's disclosure rules, and a human approves every external document before it leaves.
- PIPEDA
What it demands here
Personal information at a fund means the LP register and the deal team's own records, handled under Canadian privacy law with residency and retention answered up front.
How the system complies
The LP register stays inside your fund admin extract, deal-team records stay in your accounts, and the build is designed around PIPEDA from the start with data residency scoped per engagement.
- NI 31-103 registrant obligations
What it demands here
Where the manager is registered with a Canadian securities regulator. Books and records kept and producible, client information held confidential.
How the system complies
Everything runs in your accounts, so the records the regime demands accumulate where your records already live, with audit logs attached.
- Vendor posture
What it demands here
Vendors who can survive the operational due diligence questionnaire your LPs will send.
How the system complies
Our core vendors (Anthropic, Supabase, Vercel) publish SOC 2 reports on their trust pages. We design within that posture and claim no certification we do not hold.
Deal data stays inside your accounts under no-training API terms, scoped per deal team with audit logs.
The objections
We have DealCloud and a research budget. What does this add?
DealCloud tracks who you know and where the deal stands. It does not read the data room and it does not write the IC brief. A deal desk built this way sits on top of the systems you keep. It is aimed at the assembly hours, not at the software.
Our archive is a mess. Ten years of memos across shared drives and inboxes.
That is the expected starting state, not a blocker. The first phase indexes the IC memos, diligence findings and QoE summaries wherever they sit, then asks the deal archive the questions a partner would ask. What did we conclude on this space in 2021. Why did we pass. You see the retrieval score on your own past deals before anything is built on top of it.
If the system writes the first draft, how do associates learn the craft?
They learn the part that was always the craft. Reading a QoE sceptically, testing a management case, defending a line in front of a partner. Nobody learned judgment by reformatting comps at midnight.
Our next fundraise will put AI in every DDQ. Does this create a disclosure problem?
It gives you the answer instead of the problem. The architecture fits on one page: data stays in your accounts, walls enforced at the database layer, no training on your data, a person approves everything external. That reads better in an ODD meeting than whatever your associates are quietly doing with public tools.
Portfolio companies own their own systems. How does monitoring work without reaching into them?
It reads what the portcos already send you. Board packs, monthly reporting packs, covenant certificates. The system watches those documents for movement and flags what changed. It never logs into a portfolio company system.
We are mid-fundraise. The timing feels wrong.
The first phase touches nothing an LP sees. It indexes your own archive and measures retrieval. Reporting work starts only after the brief holds, which is the same order the build sequence on this page commits to.
How we build it for private equity firms
Index the archive first.
Before any agent exists we load the diligence memos, CIMs and IC decks the firm has already produced and measure retrieval against questions your partners have answered before. If the archive cannot answer its own history, nothing built on top of it will.
Build the brief, not the decision.
The deal-desk agent assembles a committee-ready brief with every claim cited. It runs against a graded set of real IC questions before a partner ever sees output.
Widen to reporting once the brief holds.
LP reporting and portfolio monitoring reuse the same index and the same eval harness. We do not start here, because reporting is where a wrong number costs the most.
What we will not automate
The investment decision. The system does collection and first-pass analysis, and a partner who can defend every line signs the memo. We also will not let an agent write to your fund admin system.
How the system is built for private equity firms
See the full capability mapRetrieval
Every diligence memo, CIM and IC deck the firm has produced, chunked and indexed. Ask for the MAC clause and exact wording finds it. Ask what could kill this deal and meaning finds it. The brief cites the page.
- pgvector
- Full-text BM25
- Reciprocal rank fusion
Agents and orchestration
A deal-desk agent assembles the brief from your store plus public filings and feeds. It drafts. A partner approves. Autonomy stays capped below the threshold where anything reaches the committee unreviewed.
- agent-worker
- Autonomy guard
- Proposal queue
Evaluation
A graded set of real questions your IC has asked, with answers your partners already agreed on. Any change to prompts, retrieval or model reruns it before it ships. Accuracy drift shows up in the harness, not in a meeting.
- Eval graders
- Regression targeting
- quality-worker
Models
Frontier reasoning models for synthesis across hundreds of pages of deal documents. Embedding models for retrieval. Plain statistical forecasting for portfolio metric drift, where a large model is slower and no better.
- Frontier models, one gateway, routed per task
- Embedding models via the same gateway
- Time-series forecasting
Data boundary
Information walls are enforced at the database layer, not by policy. A deal team's data is invisible to every other team by construction, with audit logs on access.
- Supabase row-level security
- Access audit logs
- No-training API terms
Data room, Fund admin, Market feeds feed a hybrid index. The agent runtime works from that index, and every consequential action passes a human approval before it reaches Ask your brain, Ask your team, Approval queue.
Your systems
- 01Data room
- 02Fund admin
- 03Market feeds
The system
- 01Hybrid index
- 02Agent runtime
- 03Your approvalhuman
Where your team works
- 01Ask your brain
- 02Ask your team
- 03Approval queue
What that means in practice
Retrieval and RAG
Connecting AI to your actual documents so it answers from your knowledge, accurately and with citations, instead of making things up.
Where we stop. A dedicated vector database is justified by scale, not by default. Most of RAG quality is won or lost in chunking and indexing strategy, not in the model choice, and we have walked clients back from RAG to plain search when that was the honest answer.
Agents and orchestration
AI that does the work instead of just answering: looks things up, calls your systems, completes multi-step tasks, and knows when to hand off to a human.
Where we stop. Multi-agent swarms are oversold; most jobs need one well-guarded loop. If a cron job and a script solve it, that is what we build, because 90 percent per-step accuracy compounds to 59 percent over five chained steps and no framework changes that arithmetic.
Evaluation and observability
How we prove the AI actually works: measured, monitored and regression-tested like real software, not vibes.
Where we stop. There is no engagement where we skip this. The honest variable is depth: a document pipeline gets faithfulness and extraction suites, an outbound agent gets human review sampling, a classifier gets a held-out test set. We size the harness to the risk, never to zero.
Where your team works
Tour the platformAsk your brain
Ask a question. Get the answer and where it came from.
Every answer cites the document behind it, so you can check the source yourself. When the brain does not have the answer, it says so instead of guessing.
Ask your team
Chat with a team that already knows your business.
Every agent is briefed on your documents, your data, and your preferences. Ask for the number, the draft, or the plan and cite where it came from.
Approval queue
The AI proposes. You approve. Nothing sends itself.
Every consequential action arrives as a proposal with the risk, reversibility, and expiry spelled out before you say yes. Control stays in the room.
Cost per outcome
Every dollar of spend traced to the work behind it.
Outcomes delivered, cost per outcome, value attributed, return on spend. The same measurement the guarantee is settled against, live on one page.
By team
The same system, seen from the desk that runs it.
Run your numbers.
Your operations
Savings use the low end of our hours-reclaimed range. The math is conservative on purpose.
The math
Calculated at the low end of every range.
Every first build is covered in writing: 5x ROI in 30 days. Or we work for free.
The hard questions
Free · 3-5 days
Know your number in five days.
We map your operations, find the highest-ROI automations, and hand you a ranked plan with the payback math attached. Yours to keep, whoever builds it.
Prefer to talk first? Book 15 minutes with James. No pitch deck.