White Paper

Passive Intelligence

How agentic AI captures the financial intelligence your portfolio companies already generate, before it dissipates.

John Leese, CPACo-Founder & Operating CFOApril 2026  ·  8 min read
Authored by
JL
John Leese, CPA
Co-Founder, Operating CFO  ·  Portfolio Operating Group
Published April 2026v1.0  ·  8 min read

The Problem Nobody Budgets For

Every PE-backed portfolio company generates more financial intelligence in a single week than its finance team can capture, retain, and act on. The intelligence exists. It just dissipates before it reaches the people who need it most.

Consider a typical mid-market holding company ($150M to $300M in consolidated revenue, ten business units, project-based work recognized under percentage-of-completion). Every week, project managers sit in WIP forecast meetings and discuss margin trends, change order activity, completion estimate shifts, and the gap between estimated and actual costs. Controllers and analysts analyze it in monthly variance reports. BU presidents flag it in emails. The CFO and CEO discuss strategic implications in conversations that rarely get formally documented.

The information is generated constantly. The problem is that too much of it dissipates after the meeting ends, after the email is read, after the call hangs up. Meeting discussions produce insights that disappear. Variance analyses arrive as point-in-time snapshots with limited pattern recognition across months or business units. Commitments made by project managers in WIP meetings go untracked unless someone manually follows up. Executive decisions that should inform forecast assumptions live in the CFO's memory rather than in the model.

The Core Problem

The people responsible for financial planning, analysis, and project oversight spend a disproportionate share of their time chasing information rather than acting on it. Quarterly re-forecasts take three to four weeks. Board decks are assembled manually from inputs that may already be stale. Contract risk (margin erosion, change order exposure, completion estimate drift) surfaces late, sometimes after it has already hit the P&L.

The result is a finance function that is simultaneously drowning in information and starved for intelligence. And the cost is not abstract.

~60%
of analytical team capacity spent on information assembly, not analysis
3–4 wks
typical quarterly re-forecast cycle before the team can act on it
$1–2M
annual margin impact from delayed risk visibility at $200M revenue, 15–20% gross margin

At $200M in revenue with gross margins of 15 to 20 percent, even a half-point improvement in margin visibility through earlier risk detection represents seven figures of annual impact. The labor currently dedicated to information assembly rather than analysis represents over $100K per year in capacity allocated to work that a well-designed system can eliminate.

A fade identified at month two of a six-month slide gives operations four months to act. The same fade identified at month six gives nothing but a write-down to explain.

From the seat  ·  Operating CFO perspective

The Approach: Capture What Already Exists

The solution is not another dashboard, another reporting tool, or another process layered on top of people who are already stretched. It is a system that passively captures the information already flowing through the organization, making it queryable on demand.

No new data collection processes are imposed on the people generating the information. Project managers, controllers, analysts, and BU presidents continue doing exactly what they do today. The system captures their existing outputs (meeting transcripts, emails, executive conversations) and ensures that what they say and write is retained, organized, and available when it matters.

Three Design Principles

Capture where information already exists rather than creating new collection workflows. The transcripts, emails, and conversations are already happening. The system structures what currently dissipates.

Use deterministic rules where possible and AI only where human-like judgment is required.Orchestration logic, routing rules, and threshold-based alerts are code. AI earns its place in tasks that require interpreting unstructured language: reading a project manager's hedged language about a completion estimate, recognizing that a controller's variance explanation echoes a pattern from three months ago, or translating a CEO's strategic direction into a forecast assumption.

Keep the human in the loop at every decision point that affects reported numbers. The system surfaces and recommends. The financial analyst, controller, and CFO review and decide.

Methodology

The architecture uses a model-routing strategy: lightweight AI handles the high-volume work (transcript summarization, email classification, data extraction) at the lowest cost tier, covering roughly 80–90% of processing. A more capable reasoning model handles the tasks requiring deeper judgment: evaluating risk signals, synthesizing forecast assumptions, and identifying cross-period variance patterns. The routing between models is rule-based, not a judgment call made by the AI.

Four Dimensions of Impact

Earlier risk visibility. The system surfaces project-level risk signals weeks earlier than the current reporting cycle allows. Margin erosion, change order exposure, and completion estimate drift are flagged as they emerge in weekly discussions, not discovered during quarterly close.

Objective accountability. An unbiased, longitudinal record of commitments and follow-through at the project and individual level, derived from meeting transcripts and documented communications rather than subjective recall. This provides a defensible foundation for performance and resource allocation discussions.

Compressed analysis cycles.The time required for variance analysis assembly drops dramatically. Where a controller currently spends roughly a full day per business unit gathering information before true analytical work can begin, a pre-assembled knowledge layer compresses that to one to two hours per BU. Across ten business units, that is the difference between two full weeks of the team's time and two to three days.

Sourced forecast assumptions.The financial analyst starts each forecast cycle with structured, AI-enabled recommendations rather than spending weeks gathering information about what has changed. The analyst's judgment authority over every input is preserved: the system recommends, the human decides.

MetricBeforeTarget
Time on information assembly~60% of team capacityUnder 20%
Quarterly re-forecast cycle3–4 weeks1 week or less
Variance analysis per BU~1 day1–2 hours
Board materials prep~2 weeks manual2 days review
Risk flag false positive rateN/A (new capability)Under 5%

Implementation: Crawl, Walk, Run

This is not a big-bang deployment. The system follows a gated, 18-month roadmap where each phase has defined scope, specific deliverables, and measurable criteria that must be met before advancing. There is no automatic advancement based on timeline alone.

Phase 1
Crawl
Months 1–4

Single business unit. Single data source. 100% human review of every output. The goal is not to save time. It is to establish baseline accuracy and build confidence that the extraction is reliable.

Phase 2
Walk
Months 5–9

All business units. Email ingestion and executive conversation capture come online. Remaining use cases activate. Human review shifts from 100% to sampling-based.

Phase 3
Run
Months 10–18

Full operational cadence. The team's role has shifted from assembly to oversight and analysis. Cross-BU pattern recognition begins surfacing insights no single unit's data would reveal.

Gate Reviews Matter

Gate reviews at Month 4 and Month 9 require documented achievement of advancement criteria. If extraction accuracy doesn't exceed 90% as validated by the reviewing team, the system stays in the current phase until it does. This discipline is what separates a responsible implementation from a tech demo.

Cost Profile and Governance

The cost structure reflects an internal build led by the CFO: modest capital expenditure, predictable operating costs, and a payback period driven primarily by the redeployment of existing labor from assembly work to analysis and oversight.

$24–36K
Year 1 all-in cost (CapEx + OpEx base)
<$18K
Annual run rate (Year 2+ OpEx ceiling)
320 hrs
Total internal time investment across 18-month build

Capital expenditure is concentrated in the first four months: external consulting for architecture, security review, and production hardening. Operating expenses are dominated by the AI platform licensing, chosen at the enterprise tier specifically for audit logging, SSO, and compliance API access. When the system processes meeting transcripts containing project-level margin discussions and board-level financial content, the absence of audit logging creates a governance gap that a PE sponsor's due diligence would reasonably question.

Governance: CFO Ownership

Ownership of the system sits with the CFO. This is not a shared-governance model with IT or an AI steering committee. The CFO owns the data, the outputs, the risk, and the decision to continue, pause, or shut down the system. This clear ownership structure avoids the diffusion of accountability that stalls AI initiatives in larger organizations, and it reflects the reality that the system's outputs feed financial reporting processes that already fall under the CFO's authority.

Because the underlying financial systems are read-from and not written-to, pausing the AI layer returns the team to their prior workflow with no disruption to ongoing operations. The system prepares. The team reviews. The review findings feed back into the system's configuration. It is the same preparer/reviewer framework already embedded in every well-run finance function.

Risk Mitigation

If the system produces a material error, the CFO suspends automated outputs and reverts to full manual review until a root cause analysis is completed. The architecture uses portable technologies with no platform-specific lock-in. Migration to a different provider is straightforward if the PE sponsor mandates it or the system outgrows the platform.

Who This Is For

This approach is designed for PE-backed portfolio companies where the finance function is lean, the information volume is high, and the cost of late visibility is measured in margin points, not convenience. It works best where:

  • The business is project-based with percentage-of-completion revenue recognition, where margin health lives in the details of individual contracts.
  • The holding company spans multiple business units, each generating their own operational intelligence that rarely cross-pollinates.
  • The finance team is small enough that a two-week information assembly cycle represents a material fraction of their capacity.
  • The executive team makes decisions that should inform forecast assumptions but currently rely on memory rather than documentation.

The bottleneck was never entering a number into a cell. It was figuring out what the number should be.

From the seat  ·  Operating CFO perspective

The information your portfolio companies need to make better decisions faster is already being generated every week. The question is whether it reaches the right people before it dissipates. Or after.

Ready to discuss what this looks like for your portfolio?

We've built and operated these systems from the CFO seat. Let's talk about what passive intelligence could mean for your companies.