Having spent decades in private equity and technology, we believe most firms are still thinking about AI too narrowly. The current conversation often assumes the highest strategic leverage comes from implementing AI at the general partner level—building dashboards, centralized monitoring systems, portfolio-wide intelligence layers, and synthetic operating capabilities sitting above the portfolio. 

That is important work. But it is not where the economic leverage begins. 

Why AI Should Start Inside Portfolio Companies 

The first and most important implementation of agentic AI should occur inside the portfolio companies themselves. 

The reason is simple economics. 

A private equity firm typically operates on roughly 2% of committed capital through management fees for operational expenses. The portfolio companies themselves represent the deployment and performance of the remaining up to 98% of capital. If AI materially improves operational performance, margin structure, growth rates, working capital efficiency, pricing intelligence, or exit metrics, the largest economic impact necessarily sits inside the portfolio companies, not inside the management company. 

That does not mean the GP-level layer lacks importance. It means sequencing matters. 

The Right AI Implementation Sequence for PE Firms 

The firms that win the next decade will likely follow a progression: 

• Implement AI deeply inside individual portcos  
• Standardize and propagate successful implementations across the portfolio  
• Build GP-level agentic infrastructure on top of increasingly instrumented companies  
• Eventually create intelligence flows between portcos that compound learning across the entire fund  

AI in the Portco: Where Value Creation Happens 

Most “AI strategy” conversations inside firms today conflate two fundamentally different programs. There is AI in the portco—agents living inside operational workflows: 

• Sales and account agents that research prospects, draft outreach, and update CRM after every call  
• Pricing agents that read competitor moves, inventory, and demand signals to recommend the right price customer by customer  
• Finance agents that close the books, reconcile accounts, and flag anomalies before the controller does  
• Customer service agents that resolve routine tickets autonomously and route the rest with full context  

This is where the first-order value creation happens. 

Revenue acceleration, SG&A compression, pricing optimization, inventory efficiency, working capital improvements, customer retention, procurement leverage, and institutional knowledge retention all occur at the operating company level. These are not theoretical efficiencies. They directly impact EBITDA, cash flow generation, lender confidence, valuation quality, and ultimately exit multiples. 

The portco operates and benefits; the firm advises. 

AI on the Portco: Portfolio-Level Intelligence 

Then there is AI on the portco—agents the firm runs from the outside, looking down across the portfolio: 

• KPI anomaly detection that flags drift before the next board pack  
• Board materials auto-generated from portco source systems  
• An institutional memory layer queryable by every deal team  
• Cross-portfolio benchmarking that makes one portco’s pricing experiment instantly visible to its peers  

The firm operates and benefits; the portco is the data source. 

These are different programs, not points on a spectrum. And the sequencing matters more than most firms appreciate. 

Why Most GP-Level AI Efforts Fail Early 

In practice, firms attempting to build sophisticated “on-portco” infrastructure before their portfolio companies are properly instrumented often discover the same problem repeatedly: fragmented data, inconsistent KPI definitions, uneven management engagement, incompatible ERP systems, and insufficient operational telemetry. 

The centralized intelligence layer becomes constrained by the immaturity of the underlying operating companies. 

Ironically, the fastest path to building a powerful GP-level AI layer is first building successful AI-enabled portcos. 

What Happens When Portcos Become Agentic 

Once portcos begin operating agentically: 

• Data quality improves  
• Workflows become measurable  
• Operational signals become standardized  
• Teams become comfortable interacting with agents  
• Benchmarking across companies becomes materially easier  

The Structural and Economic Differences 

The structural differences run deep. “In” investments sit in portco P&L, are owned by the portco CEO, must demonstrate measurable impact within the hold period, and transfer to the buyer at exit—often as a multiple lever. 

“On” investments are funded at the firm level, amortize across the fund’s life, and stay with the GP forever. 

The data plumbing is different, too. “In” keeps data inside the portco perimeter, while “on” requires extracting and normalizing data from heterogeneous portco systems—a problem most firms have not solved well. 

Key Economic and Governance Questions 

There are also economic alignment questions the industry has not fully addressed: 

• When the GP builds centralized AI capabilities, who pays?  
• Management fees?  
• Portco P&L?  
• Shared services allocations?  
• Who owns the resulting IP at exit?  
• Does the buyer inherit the infrastructure?  
• Does the GP retain the intelligence layer?  

These questions become increasingly important as agentic systems move closer to core operational workflows. 

The Next Frontier: AI Between Portcos 

The real transformation ultimately comes when the portfolio itself becomes agentically interconnected. 

A third category is now emerging that earlier architectures could not reach—agents between portcos: 

• Procurement agents pooling spend across multiple companies  
• Pricing benchmarks flowing from one portco to another with appropriate consent walls  
• Talent agents surfacing executive candidates the moment any portco opens a CFO seat  
• Working capital optimization patterns propagating across the portfolio  
• Commercial playbooks transferring from one operating environment to another almost in real time  

This may ultimately become the highest strategic leverage point in private equity. 

Building a Synthetic Operating Layer 

A single agent can monitor dozens of portcos overnight for: 

• Margin deterioration  
• Pricing drift  
• Covenant pressure  
• Working capital anomalies  
• Customer churn risk  
• Pipeline weakness  
• Procurement inefficiencies  
• Operational variance against historical patterns  

It can gather context from prior board materials, surface comparable situations from past deals, and place drafted interventions in front of operating partners before the morning standup. 

That is not a dashboard. It is a synthetic operating layer sitting above the portfolio. 

Technology Layers Required for Agentic PE 

Each layer has its own technology center of gravity: 

• In the portco: workflow-specific agent platforms connected to CRM, ERP, finance, supply chain, and operational systems  
• On the portco: GP-level orchestration frameworks, KPI systems, governance layers, and analytics  
• Between portcos: federated architectures, identity controls, and secure intelligence-sharing systems  

None of these technologies are exotic anymore. The challenge is sequencing them correctly. 

What Determines Success vs Failure 

The firms that fail will attempt to impose centralized AI architectures onto immature operating environments. 

The firms that succeed will build from the operating company upward: 

• AI-enabled workflows inside individual companies  
• Repeatable implementation patterns across the portfolio  
• Portfolio-wide operational intelligence  
• Long-term institutional learning advantages  

The Three Prerequisites for Success 

The technology itself is only one leg. Three prerequisites determine whether any of this produces durable advantage: 

• A normalized data architecture across portcos  
• A scalable governance framework  
• A culture that views AI as leverage, not surveillance  

The third may be the most important. 

Cultural Adoption and Risk Considerations 

If portco CEOs interpret AI as surveillance, they will curate data defensively. If they view it as a value driver, they become co-builders. 

Failure modes are real: 

• Data leakage  
• Compliance exposure  
• Hallucinations in workflows  
• Cyber vulnerabilities  
• Audit complications  
• Buyer resistance during diligence  

The Future of AI in Private Equity 

Buyers will increasingly distinguish between companies that use AI tools and companies whose operations are structurally agentic. 

The latter are more scalable, resilient, and likely to command higher valuations. 

The Bottom Line: Start with the 98% 

The durable prize is not a smarter GP. 

It is a portfolio of companies whose operating systems are more intelligent, adaptive, and valuable. 

The firms that win will compound intelligence across portcos, vintages, and time. 

The question is no longer whether agentic AI will reshape private equity. 

The question is which firms will begin where the economic leverage actually sits—inside the 98%—before attempting to optimize the 2%. 

Interested in how this could apply to your portfolio companies? Let’s connect. 

Written by Riaz Siddiqi,  Executive Chairman, and Prashant Mishra, Chief AI and Data Officer at AAXIS.