BOTTOM LINE UPFRONT
In this episode of AI & PE: The Future of Value Creation, Kyle Roemer welcomes Ash Pembroke from Accordion’s AI Strategy and Value Creation team to unpack Accordion’s latest PE AI adoption benchmark report, which outlines how PE firms and portfolio company finance functions are adopting AI today.
Kyle and Ash broke down seven takeaways that shape how PE-backed companies should think about scaling AI in the finance function:
1. The ask has shifted from ideas to owners
Fewer than 30% of companies work from a formal AI playbook or have a center of excellence in place, and that shows up in what clients ask for. Last year, they wanted help shaping use cases and spotting where the value could be. But those early engagements didn’t always deliver, so this year the ask has grown: bigger, more formal programs with real training and adoption plans built in.
The playbook isn’t sitting on a shelf. It’s being written in real time, pulled straight from documented business processes and hands-on work with real use cases.
2. Governance has to run at two levels at once
Prolific, easy-to-access AI tools raise the governance question fast. Two guardrails need to run simultaneously:
- Boilerplate technology governance: enterprise security standards and monitoring baked into every tool rollout
- Ongoing use governance: cost federation, so a function leader knows what their team is consuming and why
Both decisions can be made once and pushed into every implementation, but they have to be made.
3. CFOs are cautious owners, and that caution is the barrier
Only 17% of PE-backed CFOs have formally claimed AI transformation as a strategic mandate. Makes sense: finance exists to be the brake, and generative AI’s non-deterministic behavior doesn’t sit well next to a need for auditable outcomes.
A human-in-the-loop workflow with traceable outputs takes real engineering, not just switching on enterprise Claude or ChatGPT. Most companies still haven’t done that unsexy foundational work.
4. The three-legged stool replaces the wunderkind engineer
Early forward-deployed AI work assumed a single savant engineer could drop into a messy environment and deliver ROI alone. In reality, the model that works involves three roles:
- The expert: the product owner of the process, closest to the business problem
- The architect: ensures what’s built is sustainable, secure, and future-proofed
- The engineer: builds it, sometimes one person, sometimes a small team
It functions a lot like a scrum team, by design.
5. Data readiness is the top blocker, and the deeper issue is process
71% of operating partners cite data readiness as the biggest barrier to AI adoption. But clean data isn’t really the problem; LLMs handle scanning and formatting fast now. The real issue is the process behind it, seventeen versions of the same spreadsheet, no naming convention, no version control. Blaming the data is easier than making the decision it demands.
6. AI-enabled finance is becoming a deal underwriting factor
86% of operating partners expect buyers to pay a premium for AI-enabled finance capability within a few years, and that premium isn’t just about efficiency gains. It reflects what transformation actually delivers: consolidated decision rights, sharper forecasting, more transparency across the business.
That’s exactly what lets investors underwrite a turnaround in six months instead of two and a half years, even inheriting a messy setup. Speed to value is starting to show up in deal pricing.
7. Move fast, but move with an owner
Guidance for operating partners on moving from pilots to real programs:
- Pick one accountable owner per initiative to avoid decisions getting spread across a committee
- Train teams through real delivered use cases alongside the people doing the work
- Get something in front of the business early, even if it’s not fully built, and use that to gather the next round of requirements
Staying academic for too long is its own kind of risk.
Subscribe to AI & PE: The Future of Value Creation:
Watch the full episode: