Why Decentralized AI Ownership Destroys Portfolio Cost Visibility

Forty percent of private equity portfolio companies' AI spend is invisible to the chief investment office. This is not a guess. FTI Consulting's 2025 study of 150 PE-backed mid-market companies found that 4 out of 10 dollars spent on AI—training, inference, tools, people—sits in business-unit budgets with no rollup to portfolio leadership. The cause is structural: when you give each operating subsidiary the autonomy to pick their own AI vendors, tools, and execution models, you inevitably create a federated mess that finance cannot see or govern. The solution is not to centralize AI execution. It is to centralize AI governance while allowing decentralized implementation.

The 40% Invisible AI Spend Problem

Every PE operating partner has lived this version of the story. You fund the CFO's cost-attribution initiative in Q1. By Q2, you have dashboards showing AI spend across the portfolio. By Q4, the CFO's assistant discovers that the largest portco is running a custom LLM training program on AWS that no one knew existed—$340K per month buried in cloud infrastructure. The claims processing team is using a third-party vendor that charges $8K per month. The customer success team is piloting three different AI agents, and the spend is split across three corporate credit cards and one contractor's cloud account.

This is shadow AI at scale. According to FTI, it accounts for roughly $3 out of every $100 in portfolio AI spend going ungoverned. Multiply that across a $3B portfolio with $24M in annual AI spend, and you're looking at $720K per year of unaccounted infrastructure.

But the real cost is not the hidden dollars. It is the decisions you cannot make. You cannot optimize vendor contracts if 40% of your spend is invisible. You cannot benchmark cost-per-outcome across portcos if some portcos are not even tracking it. You cannot avoid duplicate vendor spend or identify licensing arbitrage opportunities. You cannot answer the board's question: "Is our portfolio AI spend concentrated in 2–3 vendors, or is it diffuse?" (Answer: probably both, and you don't know which.)

Why Decentralization Happens

Decentralization is not a failure of planning. It is a rational response to uncertainty and speed. When the CEO of your healthcare portfolio company does not know whether they will use GPT-4 or Claude for their claims workflow, they do not wait for a 90-day RFP. They stand up a pilot on both platforms and charge it to OpEx. When the COO of your staffing company finds an AI agent that cuts contact-center handle time by 18%, they sign the contract and ask permission later.

Each decision is locally rational. In aggregate, they destroy visibility.

The structural problem is that AI governance did not exist three years ago. Your portfolio companies' finance and ops teams have no playbook for "how to budget for AI," no procurement standard for "which vendor category requires a five-year contract," and no shared vocabulary for "what does it cost to run this workflow?" So every company invents its own answer. The result is a portfolio that sits at stage 1 or 2 of the 5-Stage AI Cost Maturity Curve—invisible or tracked, but not allocated, optimized, or governed.

The Portfolio-Level Blindness

Decentralized ownership creates three specific blindness zones for the chief investment office:

First, you cannot see vendor concentration. The CIO asks: "What percentage of our portfolio AI spend goes to OpenAI?" Finance answers: "About 60% of what we can measure." You know that's wrong, but you cannot prove it. The portcos are using OpenAI, Anthropic, Azure OpenAI, and APIs bundled into third-party products that themselves use OpenAI under the hood. You have no contract leverage. You cannot negotiate volume discounts. You cannot even model what happens if OpenAI's pricing rises 20%.

Second, you cannot benchmark cost-per-outcome across portcos. You ask: "Is our healthcare claims operation more efficient than our insurance underwriting operation?" Finance answers: "The claims team says cost per adjudicated claim is $18; the insurance team doesn't track it at all." Now you have no idea whether to double down on claims AI or fix underwriting first. You cannot allocate capital to the portcos with the best ROI, because you cannot see ROI.

Third, you cannot identify duplicate spend or licensing waste. It happens constantly. One portco signs a $2K per month contract with a workflow automation vendor that another portco could have used for a different use case, saving 60% on a duplicate tool purchase. You never find out, because the vendors are set up on separate credit cards and neither CFO knows the other is buying.

Centralized Governance, Decentralized Execution

The solution is not to tell every portco "you must use OpenAI and only OpenAI" or "all AI decisions must go through the CIO for approval." That kills innovation speed and ignores the reality that different portcos have genuinely different AI needs. A healthcare provider's claims workflow is not an insurance adjudication workflow.

Instead, implement centralized governance with decentralized execution. This means:

Layer 1: Portfolio-wide AI governance framework. Establish a standard across all portfolio companies: (1) all AI spend above $5K per month must be logged to a portfolio cost-tracking system; (2) all AI vendors must be evaluated against a standard rubric (cost transparency, lock-in risk, contract terms, integration depth); (3) each portco must track at least one cost-per-outcome KPI for their primary AI workflow; (4) quarterly business reviews include an AI economics deep dive.

This is not authoritarian. It's the finance discipline that should exist around any technology spend.

Layer 2: Vendor relationship management at the portfolio level. The CIO or an AI operating partner (see article #87 on the emerging AI Operating Partner role) owns the relationships with the top 8–10 vendors: OpenAI, Anthropic, Google, Azure, Hugging Face, Zapier, Make, and 2–3 specialized vendors (insurance-specific, healthcare-specific). This team negotiates volume discounts, aggregates spend data, and flags redundancy.

Layer 3: Execution autonomy. Each portco's operations team picks which vendors and tools best serve their business. But they operate within the governance framework and must log their choices and spend to the portfolio system. This creates a forced discipline without innovation drag.

Implementation: The Three-Month Build

Month 1: Visibility. Audit all portcos for hidden AI spend. This requires digging into cloud bills, contractor invoices, software subscriptions, and credit card expenses. FTI found that portcos average 8–12 point sources of AI spend when audited carefully. Once you have visibility, classify the spend: APIs, platform licenses, people, infrastructure, training data, observability. Get to stage 2 of the maturity curve (AI spend tracked but not allocated).

Month 2: Governance framework. Draft the portfolio AI governance standard. It should be 2–3 pages: what must be logged, who approves what, which KPIs are mandatory, how often you report. Get buy-in from 2–3 of your largest portcos first, then roll out to the rest. This is not a request; it's a requirement of portfolio financial health, like expense reporting.

Month 3: Vendor rationalization. Identify all unique vendors across the portfolio. Look for consolidation opportunities (can two portcos merge onto a single OpenAI account?) and negotiate volume discounts with the top vendors. Establish a "preferred vendor" list that portcos can deviate from if they justify it. This creates a gating function without killing autonomy.

The Operating Partner's Role

This is precisely the problem that the emerging AI Operating Partner role is designed to solve. Rather than burying AI governance under the CIO or CFO, assign an operating partner (or a dedicated ops leader) the mandate: "Portfolio AI spend must be visible, governed, and optimized by Q3." This person owns the vendor relationships, the framework, and the quarterly review conversations with each portco. They report to the partner-in-charge and sit in investment committee meetings where AI economics are discussed.

This role is trending hard. Korn Ferry and Heidrick & Struggles both released research in late 2024 on the rise of the "AI Operating Partner" as a distinct function within PE firms. This is why.

Why This Matters at Exit

When you exit a portco, the buyer's diligence team asks: "What's your true AI spend, and is it sustainable?" If you cannot answer with specificity because 40% is invisible, you have a problem. The buyer's advisors will uncover the hidden spend during IT diligence, use it to negotiate down purchase price, and flag it as a control gap. You want to walk into that conversation with full transparency. "Our portfolio AI spend is tracked to [system], visible to all business units and the CIO, governed under a standard framework, and benchmarked against peer portcos. Here's the data." That removes a source of buyer uncertainty and accelerates the deal.

The path to portfolio AI cost visibility is straightforward: centralize governance, decentralize execution, and audit quarterly. Operating partners running this analysis across a portfolio can request the PE Operating Partner Field Guide and the cross-portfolio rollup template to implement this framework.