Picture this: an LP due diligence questionnaire lands on your desk with a new line item: describe your firm's AI strategy across the organization and portfolio companies. You pull up last quarter's ops deck. It mentions ChatGPT twice and some general ideas of agents that could be built. It does not mention a comprehensive strategy, deployment roadmap, or agent workflow initiatives.
That gap is common, and this kind of inquiry is quickly becoming industry standard. According to data from FTI Consulting 2026 Private Equity AI Radar, the AI adoption rate for private equity firms sits at an impressive 86-95%, but their portfolio companies are at a mere 36%. Moreover, only 7% of those portfolio companies achieve enterprise-scale deployment.
Context matters here, though, and most importantly, what's the true utilization and effectiveness of the AI being used? Most institutional capital operating partners have never actually placed their firm, or their portfolio companies, on a scale beyond "we use AI" and "we don't". Employees are individually spinning up instances of Claude, gobbling up tokens without a base level understanding of the types of models to use, and how to build out skills that actually increase their output and efficiency.
Below is our four-level AI Proficiency framework, which we use to benchmark our clients' current usage and adoption of AI within their organization. It's been adapted from Ramp's L0-L3 AI proficiency framework that Geoff Charles originally introduced internally for his AI-native company. (Our company actually just made the switch to Ramp for our business banking and has been pleasantly surprised by the overall experience.)
Key Takeaways In This Post:
- Understanding Level 1: Occasional User
- Understanding Level 2: Competent User
- Five Habits to Quickly Move From L2 to L3 & L4
- Understanding Level 3: Workflow Builder
- Understanding Level 4: Technical-grade Builder & Operator
- Where the real AI maturity gap is (and why most people & organizations get stuck at Level 2)
- How to benchmark your own portfolio and assess your level
- What to do next and how to improve your AI proficiency
Level 1: Occasional User
People use a chat assistant like Claude or ChatGPT on their own, for research, drafting, editing. No company data connected. Nothing shared with the rest of the team.
What it looks like: A firm rolls out enterprise seats across the deal team. Usage climbs fast. An associate drafts investment memos in half the time. An ops lead cleans up a board deck overnight instead of over a weekend. Ask what the firm's AI strategy is, and the honest answer is "everyone has a chat window open."
Why it stalls here: This is real, measurable productivity for the individual using it, but it's invisible at the firm level. One person's shortcut doesn't transfer to the next hire. Nothing connects to the systems that actually run the business, so the tool never touches a deal database, a portfolio company's financials, or a CRM. If a partner asks what this saves the fund in dollars, there's no answer to give them. This is where most firms actually sit when they tell an LP they "have an AI strategy." This is where a proper AI Readiness Assessment and AI Transformation conversation with an AI expert can start to prove incredibly valuable.
Level 2: Competent User
The assistant reaches into real company data (a database, a document repository, a data room) to answer questions. Still one person at a time, driving every step.
What it looks like: An ops lead at a portfolio company stops asking someone to pull a report and instead asks an AI tool, in plain language, to query the ERP directly. The answer that used to take a day now takes minutes. This is a genuine capability jump. It's not faster typing anymore. It's faster access to ground truth.
Why it stalls here: It's still one person, one query, no governance, and nothing that survives that person changing roles. If the ops lead who built this habit leaves for another job, the capability leaves with them. There's no standard version to hand to the next hire, and nothing a compliance team could review even if they wanted to.
How to make the jump from L2 to L3: Find and gather all the evidence to iterate effectively. Export the actual chat sessions where the workflow worked and where it broke, and share the prompt or skill files you've been using. Then write a short, plain-English list of what's missing. A user who shows up with real sessions, data, & context becomes a builder ready to fix and improve. A user who explains the tool from memory gives them a guessing game. The sharpest users we work with act like product owners of their own workflows: they build the rough first version, document where it fails, and let a builder harden it into something the whole team can run.
Five habits that help competent users more quickly become workflow builders:
- Keep instructions lean. State the goal and the general context. Bloated instructions compete with each other and make results worse.
- Give AI its own workspace. Work from copies of key documents in a dedicated folder, so source files stay protected and version-controlled.
- Process heavy files once. Convert scanned PDFs to clean text after the first pass, so you don't pay to re-read them in every session.
- Match the model to the phase. Use your strongest reasoning model to design the approach, and a faster one to do the analysis.
- Start from proven methods. Point the AI at established frameworks and references instead of asking it to invent a methodology.
Level 3: Workflow Builder
Prompts, skills, and workflows get codified and distributed by role. The whole team gets the same repeatable capability, with some ownership and governance attached, instead of one person's private trick.
What it looks like: The diligence workflow one sharp analyst built gets turned into a template that every analyst on the team uses the same way. A skill for drafting LP updates gets built once, owned by someone, and reused by the whole IR function. This is the point where "we use AI" actually becomes "we have a system."
Why it stalls here: This stage is durable, but it's still bounded by human sequencing. A person still opens the tool, runs the workflow, and checks the output at every step. It scales what the team is capable of. It doesn't yet scale the team's time.
Level 4: Technical-grade Builder & Operator
Agents use tools to act across multiple systems on multi-step tasks, embedded in the actual workflow, with a human reviewing and approving before anything ships.
What it looks like: Instead of an analyst running the quarterly portfolio review by hand, an agent pulls from the CRM, checks it against the data room, drafts the update, and flags exceptions for a human to sign off before it goes anywhere. The human's job shifts from doing the work to approving it. Agents run end-to-end processes on their own. People step in only on exceptions. AI reshapes how the organization is structured and what roles actually do.
Why it's a long-term, realistic goal for most firms: This level demands real integration work, systems that actually talk to each other, permissions that are properly scoped, and governance strong enough to make the review step meaningful rather than a formality. But it's a totally achievable long-term goal for most firms with the right partner, training, and strategy.
Where the real AI maturity gap is (and why most people and organizations stall there)
The jump that matters most isn't Level 3 to 4. It's Level 2 to 3. That's where an individual habit becomes an organizational asset, and where governance starts to exist at all.
Most portfolio companies get stuck exactly there because the easy part (buying seats, connecting one tool to one database) is cheap and fast, and the hard part (codifying a workflow, assigning an owner, building in review and audit) requires actual organizational work. It's the difference between a pilot and a production system.
It's also, not coincidentally, the exact line an LP due diligence questionnaire is now probing for. "Do you have AI tools" is a Level 1 question and table stakes at this point.
"Do you have a governed AI capability that survives a headcount change" is a Level 3 question, and it's increasingly the one being asked.
This is also where a firm's Managed AI Services typically starts: not by adding more tools to Level 1 or 2, but by turning what's already working for one person into something the whole team, and eventually the whole portfolio, can rely on.
How to benchmark your own firm's team and portfolio companies:
Ask these questions at the firm level, then again for each portfolio company:
- If the person running your best AI workflow left tomorrow, would the capability leave with them? If yes, you're at Level 1 or 2.
- Is there a single AI workflow that at least three people use the same way, with someone accountable for it? If yes, you're at Level 3.
- Does any agent take an action across more than one system before a human reviews it? If yes, you're at Level 4.
- Could you answer an LP's AI due diligence question with a specific process, an owner, and a number, rather than a list of tools? If yes across the portfolio, you're closer to Level 3 or 4 than most of the market.
What to do next
Most operators can place themselves after reading this, and most land a level lower than they expected going in. That's useful information, not a bad grade. It tells you exactly where the next dollar of AI investment should go: not more seats, but the workflow, ownership, and governance work that turns Level 2 habits into a Level 3 system.
PressW's Methodology does this mapping directly for a firm and for it's portfolio companies, benchmarked against the four levels above and with a specific plan for closing the gap. We've run this across 50+ clients in 7 industries, with zero churn once the plan is underway.
If this sounds like a logical next step in your AI maturation as a firm or individual, book a call with our team to understand what AI level your organization is actually at.