In S&P Global Market Intelligence's 2026 private equity survey, published in April, 37% of respondents said they were dissatisfied with the quality and availability of operational metrics from portfolio companies. In the same survey, 75% rated AI as ineffective for portfolio monitoring.
Taken together, those two numbers describe a frustration most operating partners will recognize. The data that reaches the fund arrives late and is often incomplete. This article looks at why portfolio problems continue to reach the fund only after impacting the business, and what changes when the fund reads each portfolio company's own systems instead of waiting for what those companies report.
Key Takeaways
- If your view of a portfolio company comes from the quarterly board pack, your value creation plan is being steered on lagging indicators.
- The earliest warning signs usually appear only when two systems are read together, such as survey scores next to billing tickets.
- AI that summarizes submitted reports inherits their delay. Tapping directly into ERP, CRM, ticketing, and network systems speeds up the flow of information.
- An earlier signal is only useful with an owner: an AI agent can flag it, but there needs to be a named person to review and decidesd the review is logged.
What Does Value Creation in Private Equity Depend On Now?
Value creation in private equity is the increase in a portfolio company's worth between acquisition and exit, through revenue growth, margin improvement, and better use of capital. With multiple expansion and cheap leverage doing less of the work than they did a decade ago, more of that increase has to come from running the business better.
Bain & Company's Global Private Equity Report 2026, released February 23, puts numbers on the shift. A typical deal now needs roughly 10% to 12% annual EBITDA growth to reach a 2.5x return over five years, compared with about 5% in the 2010s. And holding periods at exit now hover around seven years.
Seven years is about 28 quarters. So the board of a typical portfolio company will formally review it roughly 28 times between close and exit.
The value creation plan gets its course corrections at those checkpoint. Only, by that time, the issues will already have taken a bite out of the organization's profit. A problem that begins a week after one board meeting is at least a quarter old by the next, and often older, because the board pack delivered to the operating partners ahead of the meeting takes weeks to assemble after the quarter closes.
Why Does the Board Pack Arrive Too Late to Act On?
The board pack reports outcomes. Revenue, margin, churn, and EBITDA are lagging indicators, which means they confirm what has already happened. The causes show up earlier, in operational systems the fund never reads, and by the time they move the pack's numbers, the renewal or the margin is usually already gone.
At the portfolio company, the finance team exports from the ERP. Someone reconciles it against CRM figures that don't quite match. The KPIs go into the fund's template, in a format slightly different from the company next to it in the portfolio. The pack goes out weeks after the quarter it describes has closed.
A lagging indicator measures a result after it has happened. A leading indicator, while often noisy, moves before the result does, making it far more valuable. In our experience, most portfolio companies already hold their leading indicators. They sit in different systems, owned by different teams, and almost none of them reach the board pack.
| Area | What the board pack shows (lagging) | What moves first (leading) | Where it lives |
|---|---|---|---|
| Customer health | Churn, net revenue retention | Falling NPS or CSAT, negative survey comments, more complaint tickets | CRM, survey tools, ticketing |
| Service operations | Cost to serve, margin | Rising average handle time, repeat contacts, growing ticket backlog | Contact-center platform, ticketing |
| Network or production | Quarterly uptime, incident count | Maintenance backlog, recurring alarms, deferred work orders | NOC, maintenance systems |
| Finance | EBITDA, revenue against plan | Billing errors, delayed statements, days sales outstanding creeping up | ERP, billing |
| Contracts | Revenue leakage found at audit | Price adjustment dates approaching with no billing change | Contract register, ERP |
None of the signals in the third column are exotic. The problem is that the fund doesn't see those systems, and management at the portfolio company usually looks at them one at a time.
Why Can't Any Single System See the Warning Sign?
Because the most useful warning signs are relationships between systems. A dip in customer satisfaction is noise in the CRM until it lines up with a spike in billing tickets. Longer call handling times look like a staffing issue until they line up with a network maintenance backlog. Each system owner sees half the picture.
The customer team checks NPS each week and sees a three-point drop that sits within normal variation. The billing team sees ticket volume rise after a system migration and reads it as slow adoption. Both readings are plausible when taken individually.
That’s the data silos problem in practice. Each team reads its own system, and the warning sign only appears when you put the two together. In our experience, that’s where the most expensive portfolio surprises start.
Now multiply it. A fund with 20 portfolio companies, each running five or six core systems, would need someone comparing every pair of systems at every company, every week. In reality, owns that specific task, so the comparison happens once a quarter, in the board pack, after the outcome has arrived.
What Happens When an Agent Reads Every Portfolio Company's Systems?
The Squirro Portfolio Insight Agent is an AI agent built for this gap. It connects to each portfolio company's ERP, CRM, ticketing, network operations center (NOC), and financial systems, brings their metrics into a common layer, and flags deviations for review by the portfolio company or the fund, often weeks before they would reach a board pack.
Our public demo runs the agent on sample data for a fund with nine portfolio companies. One of them is a utilities business, and following its signal shows how the agent works:
- One system. The CRM survey feed shows NPS down 8 points over 60 days, the steepest decline in the portfolio. It says something is wrong without saying what.
- A second system. The agent lines the NPS series up against the ticketing system. Billing tickets climbed over the same window, and survey comments cluster around statement errors and delays.
- A third source. The billing migration log puts the new system's go-live about 60 days back. NPS was flat through the prior quarter and turned down within days of that date.
- The timing. In the demo, this signal fired roughly seven weeks before the next board pack would have shown it.
The same run flags a transport company where average handle time is up 22%, which the agent traces to unresolved maintenance tickets in the NOC. It also flags a waste-services company with three contracts nearing their price indexation window and no billing update started. That last case belongs to contract intelligence, which gets its own article in this series.
Each flag carries a tag naming the system it came from, with drill-through to the underlying records, so an operating partner can check the reasoning before raising it with management. The figures above are sample data, used to show the mechanism.
How Do You Compare Portfolio Companies That Report Differently?
By normalizing their metrics before comparing them. One company tracks NPS, another tracks CSAT, and each formats its operational data differently. The agent maps each company's metrics onto a common scale, so the fund-level view can rank companies as stable, watch, or alert, and point attention at the steepest move.
For an operating partner covering several companies, that ranking is where much of the value sits. "Which company needs me this month?" only has an answer when every company is measured the same way. The board pack tries to do this through a shared template, but it does it quarterly and by hand.
A limit that deserves being pointed out explicitly: the agent sees only what is connected. If a company runs its customer surveys in a spreadsheet that never leaves one person's laptop, that signal doesn't exist for the agent until someone moves it into a system. Connecting a new portfolio company is typically a matter of weeks.
Who Sees the Signal, and Who Decides What Happens Next?
The people who would see the underlying data anyway. Permissions in the agent mirror the access controls in each source system. Management at each portfolio company sees its own business and nothing else, and the fund-level roll-up is a separate, gated view. The agent flags. A named person reviews and decides, and that review is logged.
Oversight by design
The Portfolio Insight Agent is designed to surface and flag. Follow-up actions, such as a call with management, a change to the plan, or a fix at the portfolio company, all stay with human experts. Each flag and the review that follows it are recorded in the audit log, including who reviewed it and what they decided, so the fund can show later when it knew and what it did.
The audit trail answers the question an investment committee asks after a miss: when did we first know? With the agent in place, that question has a dated, checkable answer.
The bigger change is in the conversation. You get to ask the CEO about the billing migration in week three, while it's still a project issue. Without the signal, the same conversation happens in the board meeting, when it has become a churn number. And because management at the portfolio company sees the same signal at the same time, the discussion starts from shared evidence, which in our experience is what makes portfolio company teams willing to adopt the tool.
What Should a Fund Ask Before Rolling This Out Across the Portfolio?
Whether it works at portfolio scale without creating a new governance problem. A fund-level deployment touches many companies, system stacks, and permission models, so the questions worth putting to any vendor cover isolation, onboarding speed, hosting, certification, accuracy, and what happens at exit.
- Is each company's data isolated? In Squirro, data is ingested per portfolio company, with isolation enforced from day one. The fund roll-up is a separate view with its own access.
- Where does the data live? Squirro runs as SaaS or fully on-premises, which matters when a portfolio company has its own residency or sector requirements.
- What certifications back it? Squirro is ISO 27001:2022 certified and holds a SOC 2 Type 1 report.
- How accurate is it, and on whose content? Ask for accuracy a customer measured on its own data. In one manufacturing deployment, a customer measured accuracy above 90% on technical troubleshooting queries, where other systems had struggled to get past 70% on the same content. That figure belongs to its own use case. What carries over is the lesson that accuracy came from tuning retrieval, taxonomy, and guardrails together, which makes it a question about architecture.
Back to the 75% of survey respondents who rate AI as ineffective for portfolio monitoring. They are describing what they have used so far, and much of it sits on top of submitted reports, so it inherits the lag. An agent that reads the source systems still needs three things to work: the systems connected, thresholds tuned so that flags carry weight, and a named owner for every flag. Without that last one, it becomes another inbox.
For the broader case for running one AI platform across a portfolio, including contract enforcement and revenue recovery, see AI for Private Equity: Turning Portfolio Data into Operational Alpha.
See the Agent Run on a Sample Portfolio
See how the Squirro Portfolio Insight Agent reads the systems of every portfolio company, connects signals across them, and flags what needs attention before it reaches the board pack.
