On this page
- Key Takeaways
- Everyone Can Rent the Same Model
- The Most Valuable Data Sits Between the Firm and the Portfolio
- Reporting Packages Were Built to Be Read Once
- Definitions Come Before Dashboards
- Four Layers Make Up the Asset
- Ownership and Access Decide Whether This Works
- Start With the Data You Already Collect
- FAQ
Key Takeaways
- Every fund already runs a data pipeline. Monthly financials, KPI tabs, board decks and commentary move between the firm and each portfolio company. Most of it lands in inboxes and shared drives where nobody can query it.
- Anyone can rent the same model. Nobody else holds years of side-by-side operating data from your portfolio. That data is the part of an AI strategy a competitor cannot buy.
- GPs say the gap is real. In S&P Global Market Intelligence's 2026 survey, 72% of GPs picked operational improvements as their top value creation lever, and 75% rated AI ineffective for portfolio monitoring.
- Definitions come before dashboards. If five portfolio companies define EBITDA five ways, no agent can compare them, and it will try anyway.
- Ownership and access are the hard part. Portfolio companies are separate businesses with their own boards. Decide who sees what before you build anything.
Every month, each company in a private equity portfolio sends its fund some version of the same thing. A P&L. A KPI tab. A few paragraphs on what went well and what didn't. Board decks come quarterly. The value creation plan gets revisited. Operating partners take notes on all of it.
Add that up over the life of a fund and you have one of the richest operating datasets anywhere: years of monthly performance from dozens of businesses, each with a written explanation of why the numbers moved. Nobody else has it. Not the model vendors, not the data providers, not even the portfolio companies, who only ever see their own slice.
And at most firms, it lives in email attachments.
I think that is the biggest missed AI opportunity in private equity right now. Firms are spending real time choosing models and piloting copilots, and the model is the one piece every competitor can buy on the same terms. The data that moves between the firm and its portfolio is the piece only you can build on.
The survey data points at the same gap. In S&P Global Market Intelligence's 2026 Private Equity and Venture Capital Outlook, 72% of GPs chose operational improvements, the top-ranked value creation lever. In the same survey, 75% rated AI ineffective for portfolio monitoring, and fragmented or unstructured data came up as a key barrier.
Read those two numbers together. The lever most GPs are pulling runs on portfolio data. The portfolio data isn't ready for AI to read.
That is fixable. It starts with treating the flow of data between the firm and its portfolio as an asset you build on purpose.
Everyone Can Rent the Same Model
Claude, GPT, Gemini. Any firm can sign up for any of them, on roughly the same terms, this afternoon. The models keep getting better and cheaper, which is great for everyone building with AI. It also means the model cannot be your edge.
What makes an AI answer useful is context: what the model knows about your situation at the moment it answers. Ask a model "which of our companies are most likely to miss plan this quarter?" and the quality of the answer depends almost entirely on what you hand it. Hand it a folder of inconsistent spreadsheets and you get a confident guess. Hand it three years of clean, comparable monthly data with management's commentary attached and you get something an operating partner can act on.
Same model. Very different answers.
This is the part of the AI conversation I think private equity underrates. Model access is becoming table stakes. The firms that win with AI over the next few years will be the ones whose context layer is richest.
Ask what your competitors can't copy
Make two lists. On the first, put everything in your AI plans that another firm could buy next week: model licenses, copilots, prompt libraries, a vendor's portfolio monitoring tool. On the second, put what only your firm has: your portfolio's operating history, your operating partners' judgment written down, the playbooks you've run across companies. The second list is where your AI advantage lives. Most firms spend nearly all of their AI budget on the first.
The Most Valuable Data Sits Between the Firm and the Portfolio
Each portfolio company owns its own systems. The ERP, the CRM, the billing platform. The fund has its own systems too: fund accounting, LP reporting, the deal pipeline. Neither of those is where the cross-portfolio view lives.
That view lives in what crosses the line between them. Monthly financials and KPIs going up. Board materials and management commentary going up. Value creation plans, reporting templates, playbooks and policies coming down. It is the only place where fifteen different businesses get looked at through the same lens, by the same people, every month.
A portfolio company CFO sees one business in great detail. An operating partner sees every business at a lower resolution, and over time starts to notice patterns: which kinds of companies stall after an add-on, which pricing changes stick, which leading indicators show up two quarters before a miss. Today most of that pattern recognition lives in people's heads. The raw material for it sits in the reporting flow.
Map what crosses the line every month
Before anyone talks about architecture, write down every recurring thing that moves between the firm and its companies. For each one, note what format it arrives in, who touches it, and where it ends up. Most firms that do this find the same thing: a lot of valuable data, arriving reliably, and then going nowhere a machine can read it.
Reporting Packages Were Built to Be Read Once
A monthly reporting package has one job. Someone reads it, asks a few questions, and files it.
That design works fine for a human reviewing one company. It falls apart the moment you ask a question across twenty companies and thirty-six months. Templates drift. A portfolio company renames a tab. A metric gets quietly redefined after an add-on. Commentary sits in a PDF that never gets connected to the numbers it explains.
Firms know this. Allvue's 2026 GP Outlook survey of 102 senior leaders at private equity, private credit and venture capital firms found that 62% expect prominent AI investment in the next year, while only 8% report highly mature data environments. (Allvue sells software for exactly this problem, so read it with that in mind. The direction matches the S&P numbers, which also found 37% of GPs dissatisfied with the quality and availability of non-public operational metrics.)
Big AI ambition, thin data foundations. That combination produces pilots that demo well and then stall.
Ask one cross-portfolio question and watch what breaks
Pick a question you'd want answered across the whole portfolio. Something like: which companies grew gross margin over the last eight quarters while adding headcount slower than revenue? Then time how long it takes your team to answer it. Count the spreadsheets, the emails to portfolio company CFOs, the judgment calls about which numbers are comparable. That number of hours is your baseline. It tells you more about your AI readiness than any maturity survey.
Definitions Come Before Dashboards
Here's a thing that scares me more than bad data: inconsistent data that looks clean.
If one company reports adjusted EBITDA with management fees added back and another doesn't, the numbers sit side by side in a table and look perfectly comparable. A person who knows the companies catches it. An agent won't. It will compare them, rank them, and write a tidy summary of the wrong conclusion.
Private equity has plenty of company here. Gartner found that 63% of organizations either do not have or are unsure if they have the right data management practices for AI, and predicts that through 2026, organizations will abandon 60% of AI projects unsupported by AI-ready data. That survey covers data leaders broadly, not PE firms. But a portfolio is a harder version of the same problem, because the data comes from companies with different systems, different finance teams and different histories.
Agree on a short metric dictionary first
Pick the ten to fifteen metrics you actually compare across companies. For each one, write the definition, the source system, the owner at each company, and how adjustments are handled. Get the portfolio CFOs to sign off. It is the least exciting document in your AI program and probably the most valuable one.
Four Layers Make Up the Asset
When I think about the data between a firm and its portfolio, I sort it into four layers. Each one is useful on its own. Together they become the context that makes AI worth running across a portfolio.
| Layer | What's in it | Where it usually lives today | What AI can do with it once it's organized |
|---|---|---|---|
| Financials | Monthly P&L, balance sheet, cash, budget vs. actual | Excel packages in email and shared drives | Explain variances across companies, flag likely plan misses early |
| Operating KPIs | Pipeline, churn, utilization, headcount, unit economics | Custom tabs that differ by company | Benchmark similar businesses, show which levers moved results |
| Narrative | Management commentary, board decks, value creation plan updates, operating partner notes | PDFs and slide decks | Answer "when did this risk first come up, and what did we do?" |
| Playbooks | What the firm tried across companies and what happened | People's heads, a few memos | Bring what worked at one company to the next acquisition |
The first two layers are what most people picture when they hear "portfolio data." The third and fourth are what I think firms underrate.
Numbers tell you what happened. Commentary tells you why. Five years ago the narrative layer was basically unusable at scale, because no system could read three hundred board decks and connect them to the financials. Language models can. That changes what the reporting flow is worth.
Start with financials and narrative
If you can only organize two layers this year, pick financials and narrative. The financials give you a comparable spine. The narrative gives the model the reasons behind the numbers. KPIs and playbooks get easier once those two are in place, because you'll know which definitions matter and where the patterns are.
Ownership and Access Decide Whether This Works
This is the part nobody puts in the pitch deck.
Portfolio companies are separate businesses. They have their own boards, sometimes co-investors, and management teams who reasonably worry about how their data gets used. A portfolio company CEO should not be able to see a peer company's numbers unless the comparison has been anonymized and everyone agreed to it. LPs care how portfolio data is handled. And when a company exits, the question of what data the firm keeps, and what it can keep using, gets real fast.
I don't have a clean answer for every one of these. The exit question in particular varies by deal and by agreement, and I'd rather say that plainly than pretend there's a standard playbook. What I'm sure of is that these questions are much easier to answer before you build than after.
Write the access rules down before the first file lands
Decide who can see what at the firm level, at the company level, and inside each AI tool. Decide where the data physically lives. (My strong preference is the firm's own cloud account, on open formats, so the firm controls it.) Log access. Then build. Our AI Security, Compliance & Governance work exists because this step gets skipped so often.
Start With the Data You Already Collect
The good news is that you don't need a new data request to your portfolio companies. You already collect most of what matters. It just isn't organized for a machine to read.
The work looks roughly like this. One governed place for the data, in your own account. A metric dictionary the portfolio CFOs agree to. Historical reporting packages loaded and reconciled. Commentary and board materials indexed next to the numbers they describe. That is what our Data Lake Build Out team builds: structured and unstructured data side by side, on open table formats, with access controls set up before anything lands. "Fragmented across portcos" is one of the four places we see portfolio AI programs stall across the 50+ investment firms PressW has worked with, and it is usually the first one worth fixing.
I'll be honest about what I don't know yet. I don't know how far cross-portfolio agents will go. Whether they'll spot a struggling company before an operating partner does, or mostly save the analyst team a week a month. Probably both, depending on the firm. But I'm confident the firms that organize this layer now will be the ones in a position to find out.
If you want a quick test, pick one question you'd want answered across every company in your portfolio and time how long it takes today. If the answer is measured in days, it's worth a conversation. Book a Discovery Call and we can compare notes.
FAQ
Because nobody else has it. AI models are available to every firm on similar terms, but years of comparable operating data from your own portfolio companies is unique to your firm. The more of that data a model can read, the more useful its answers about your portfolio become.
Typically monthly financials, operating KPIs, board materials and management commentary flow up to the firm. Value creation plans, reporting templates, playbooks and policies flow down. Together, this firm-portfolio data is the only place every company gets measured through the same lens.
You need one governed place where portfolio data is stored, defined consistently and access-controlled. For most firms, a data lake or lakehouse in the firm's own cloud account is the practical way to get there, because it holds both spreadsheets and documents like board decks and commentary.
Start with a short metric dictionary. Define the ten to fifteen metrics you compare across companies, document how each one is calculated and adjusted, name an owner at each portfolio company, and get sign-off from the portfolio CFOs. Without that, cross-portfolio AI analysis will compare numbers that only look comparable.
Ownership and access should be agreed before anything is built. Each portfolio company owns its operating data. The firm needs clear rights to use what it receives for monitoring and benchmarking, clear rules on who can see what, and a plan for what happens to the data when a company exits.
Map every recurring report that moves between the firm and its companies, then ask one real cross-portfolio question and time how long it takes to answer. That gives you a baseline and shows which data, definitions and access rules need work first.
