Buy legacy cash flow. Recompile the organization. Compound the operating system.
Explanation
This article presents the author's methodological framework, not a case report of a specific transaction; the companies, amounts and personnel mentioned are illustrative figures within the framework, do not constitute verifiable factual records, and do not constitute investment advice. Refer to other “case cards” with clickable external sources for applicable data; this article does not use that scope.
Fact
- This is an AI-native organizational transformation framework for control-oriented acquisitions, targeting traditional enterprises whose business models have been validated, have high organizational debt, and are intensive in cognitive labor.
- Value creation targets cover workflow, decision‑making authority, information flow, incentives and organizational structure, with the core infrastructure defined as the AI Transformation OS.
- The proposal suggests first validating replicability with an AI Buyout Studio / AI Transformation HoldCo and a single‑purpose SPV, then raising a traditional Blind Pool Fund.
Judgment
The core arbitrage opportunity of AI Buyout comes from Organizational Debt: buying proven cash flows and, through control, recompiling human coordination into a management system that is machine‑executable, observable, learnable, and replicable.
By combining Dan Weijian, 3 G Capital, and the AI‑native organizational methodology, the “AI PE Fund” concept can be further advanced:
The core is not to set up a PE that invests in AI, nor to acquire traditional companies and then deploy AI. The core is to establish an “AI‑native enterprise transformation company” with capital capability, dedicated to purchasing control and recompiling pre‑AI era companies into AI Native Companies. The fund is merely a financing vehicle; the real product is the AI Transformation OS.
This methodology has already clarified the underlying reason: after AI amplifies individual capability, the new bottleneck shifts from “professional execution ability” to judgment, goal definition, acceptance, responsibility, and the handoffs, waiting, and coordination between people.
Therefore, the true arbitrage in this business is:
Organize debt
I. Redefine first: What exactly does AI PE buy
Traditional PE often looks for:
Good assets + cheap price.
The Dan Weijian-style Buyout further seeks:
Good assets + governance problems.
3G is looking for:
Good brand + inefficient organization.
What AI Buyout should be looking for most is:
Good business + bad organization + a large amount of automatable work.
Here, “bad organization” does not refer to poor employees, but to the company's operating methods that contain many legacy issues:
- The apparent real asset opportunity
- 300 people do a lot of Excel work · tasks are not yet software‑enabled
- A dozen departments approve each other · Decision Rights design is lagging
- Many managers are responsible for pushing progress · Coordination cost is extremely high
- Sales can only manage 100 customers · Human attention becomes the revenue ceiling
- Customer service, procurement and finance involve a lot of manual labor · can be automated with agents
- Veteran employees hold a wealth of experience in their heads · Tacit knowledge has not yet been capitalized
- SOPs are scattered across documents · Workflows have not yet been automated
- The company runs on meetings · No Organizational Harness
- Each merger requires re‑integration · No shared Operating OS
Therefore the best target may not be:
“the company that uses AI the least.”
But rather:
The customers, brands, products and channels are already established, but the operating system still runs on software and organizational structures from the 2010 generation.
This is a very important distinction.
Acquire durable businesses with high organizational debt, obtain control, and convert human coordination into machine-executable workflows to create structurally higher revenue per employee, faster growth, and greater operating leverage.
The Chinese can be a bit harsher:
Acquire cash flows whose business model has already been proven but have not yet been AI-ified, using control weight to allocate work, authority, information and profit, recompile traditional enterprises into AI-native enterprises.
This aligns with the core of the AI-native organization methodology:
Strategy → Objective → Workflow → Roles → Organization
Instead of using the old Department / Headcount for further optimization.
This is the most important point of the entire implementation plan.
The first phase is preferably not "AI PE Fund I".
It should be:
The structure could be:
- Hierarchical Function
- Management Co · Deal, Fundraising, Governance
- AI Transformation Co · AI/FDE/Organizational Transformation Team
- HoldCo · Holding Platform Enterprises
- Deal SPV · Independent Fundraising for Each Transaction
- Portfolio Company · Acquired Enterprises
The reason is simple.
The biggest uncertainty right now is not at all:
Whether there is money to buy a company?
But rather:
Can this AI Native Transformation Playbook actually be replicated?
So first prove:
1 company can succeed in transformation → The 2 company can replicate → The 3 company's deployment costs drop noticeably.
Once this fact is established, raising USD 500,000,000 and USD 1,000,000,000 funds will make the story completely different.
Because what you are raising is no longer:
"Believe that AI will change enterprises".
But instead:
We have already demonstrated that after a company adopts this OS, 12 months can bring these quantifiable changes.
This is GP's true track record.
This point also requires learning 3G, rather than traditional large PE.
The most dangerous model is:
Buy whichever company has the greatest AI potential.
This leads to having to relearn the industry each time.
The truly compounding model should be:
Sector Playbook × AI Transformation Playbook × M&A Playbook。
In other words, in phase 1, one should first penetrate one to two industries.
I will prioritize selecting such industries:
- Indicator ideal state
- Business model · validated
- Industry growth · does not need to be especially high
- EBITDA · Positive
- Customers · Fragmented
- Processes · High‑frequency, repetitive
- White‑collar labor · Relatively high
- Data · Already partially digitized
- Work content · High proportion of information, judgment, communication, coordination
- Industry · Highly fragmented
- M&A · Can be rolled up
- AI penetration rate · Still low
- Founders · Have exit/succession needs
Thus, what's truly interesting is not a pure AI company.
Rather, it could be:
B2B services + distribution + supply chain + vertical professional services.
For example, companies that distribute food ingredients/seasonings are more worthy of study than pure manufacturing plants:
Supplier management, inquiry, quotation, CRM, client development, formula matching, contracts, orders, procurement, inventory, accounts receivable, customer service, sales support—most of these are information-intensive tasks.
There is certainly AI value on the production side, but the first-generation AI buyout is best to look for first:
Companies with a high proportion of cognitive labor rather than a high proportion of purely physical labor.
In the past:
Financial DD + Commercial DD。
Add later:
This is the part of the entire fund's future that should most develop proprietary technology.
Traditional org charts can even be placed second.
First, draw:
That is:
How many tasks does this company actually have?
For example, a 300‑person enterprise, looking beyond the 300 individuals, can be broken down into 3000–10000 real tasks:
Find clients Client research Quote Quote approval Preliminary contract review Place order Expedite delivery Procurement Supplier selection Logistics exception Invoicing Reconciliation Payment follow‑up After‑sales Recruitment Training Weekly report Management meeting…
Then form:
Record:
Waiting, handover, approval, duplicate entry, synchronization, reporting, confirmation, information search.
This corresponds exactly to the AI-native implementation path in the document: first identify the bottlenecks after AI enters real work, then systematically tag waiting, handover, scheduling, meetings, approvals and duplicate confirmations.
The final result:
Not just a simple write:
Customer service can be AI-enabled.
Rather, it should be precise to the workflow.
For example:
- Workflow Current FTE AI Potential Revenue Impact Risk Priority
- Customer research · 15 · 90% · High · Low · A
- Quote preparation · 12 · 80% · High · Medium · A
- Contract initial review · 6 · 80% · Medium · Medium · A
- Procurement inquiry · 20 · 70% · High · Low · A
- Customer negotiation · 30 · 30% · Very high · High · B
- Financial bookkeeping · 10 · 80% · Low · Medium · B
- Strategic customer relationship · 10 · 10% · Very high · High · C
This table must ultimately be submitted to the IC.
I will call:
i.e.:
How large is the gap between this company's “current organizational model” and the “AI Native theoretical organizational model”?
A 100-point model can be established:
- Dimension weight
- Fundamental business quality · 30
- Workflow Agent potential · 15
- Human Friction · 10
- Data accessibility · 10
- Implicit knowledge codability · 10
- Revenue Amplification · 10
- Control/governance changeability · 10
- Roll-up reuse value · 5
The most interesting target is not:
AI maturity 90 points.
Instead, it should be:
Business quality 90 points + AI Native maturity 20 points.
Because the middle 70 points represent your Value Creation space.
After an acquisition, absolutely do not:
Buy ChatGPT for employees Train prompts Deploy Copilot Each department submits AI use cases.
This is called:
AI Adoption。
Not:
AI Transformation。
A unified one should truly be established:
Input:
Legacy Company
Process:
Work Graph → Human Friction Map → Objectives → Circles → Roles → Agents → Harness → Ledger → Incentives
Output:
AI Native Company。
The AI-native organizational methodology has already described the underlying infrastructure:
Context → Agent → Tool → Execution → Evidence → Evaluation → Memory。
These seven modules should be productized directly, becoming shared infrastructure for all Portfolio Companies.
This is the real IP of AI PE.
Traditional 3G may:
Day 1 personnel change Day 30 budget cuts.
AI Buyout should:
Day 0—30：Instrument
Do not restructure the organization on a large scale for now.
The first thing is to make the company "observable".
Obtain:
ERP CRM Email Feishu/Slack/WeChat enterprise data Contracts Knowledge base Financial system Customer service system Ticketing Quotations Procurement Sales activities.
Establish:
Context Layer + Work Graph + Baseline Metrics。
Then let the Agent enter real work.
The reason is: let AI enter real work first, so we can identify the true new bottlenecks.
Day 30—60：Rebuild Workflow
Select 3–5 truly P&L‑impacting workflows.
For example, do not:
"HR writes recruitment JD".
Should do:
Lead → Qualification → Quote → Contract → Order
The entire chain.
Or:
Demand → Supplier → Quote → PO → Delivery → Reconciliation。
Then establish:
Human \+ Agent \+ Tool \+ Evaluator。
The core KPI should not be:
Number of AI calls.
Instead:
How much does lead time decrease? How much does revenue capacity increase? How much does error rate decrease? How many customers can each FTE serve?
Day 60—100：Zero-Based Work
Only then does the organization truly start moving.
Ask each item:
Why is this position needed?
Why is this approval needed?
Why is this department needed?
Why must A hand over to B after completion?
Why must this decision be made by the manager?
Why does customer information need to be manually entered into the CRM?
Then:
This is the AI version of 3G ZBB:
Zero-Based Budgeting → Zero-Based Work。
And not:
300 employees reduced to 180.
Instead, start again from the goal:
Strategy ↓ Objective ↓ Circle ↓ Role ↓ Human + Agent。
This methodology provides a very important organizational principle:
Circle manages outcomes, Community manages capabilities.
Therefore, the Portfolio Company may ultimately go from:
CEO → VP → Director → Manager → Team
to:
CEO ↓ Growth Circle Key Accounts Circle Supply Circle Product Circle New Business Circle ↓ Humans + Agents。
The professional systems such as engineering, sales, and finance still exist, but they become more:
Capability Community。
rather than a power center.
This is a critical cut that determines whether AI PE reform succeeds or fails.
If it is only:
You are responsible for growth.
But the budget is still approved by the CFO,
Hiring is still approved by HR,
Bonuses are still decided by the CEO,
Performance is still evaluated by the department head,
Then essentially nothing has changed.
A true Circle Leader must become:
Ownership:
Goal rights + Resource rights + Evaluation rights + Profit distribution rights.
This corresponds to the core of “divide circles, divide activities, divide money”: only by tying responsibility, resource allocation rights, and profit distribution rights together can a Circle become a real operating unit.
This actually combines:
3G's Ownership Culture
with:
AI Native Circle
merged together.
I even think this could become one of the core proprietary software of AI private equity.
Traditional post‑investment management relies on:
CEO reports CFO reports Board materials Monthly reports.
But in the future it should be seen directly:
Who
- Which Agent
- For which Objective
- Did what
- Generated what Evidence
- What was the result?
In other words:
Git commits, quotations, experiments, sales results, customer feedback, Agent trajectories, and financial results are all automatically recorded.
Thus, GP may finally achieve for the first time:
Real‑time observation of how the company creates value.
Instead of hearing a CEO report once per quarter.
This could even form a new Portfolio Management System.
Traditional PE focuses on:
Revenue EBITDA Cash Flow。
Of course, continue to look at it.
But AI PE must add a set:
For example:
- Metric meaning
- Revenue / Human FTE – labor productivity
- Gross Profit / Human FTE – even more important labor productivity
- Revenue / Knowledge Worker – white‑collar leverage
- Customer / Sales FTE – sales capacity
- Workflow Lead Time – organizational speed
- Human Handoffs / Workflow – human friction
- Agent Completion Rate · Agent Autonomy Rate
- Human Exception Rate · Human Intervention Ratio
- Cost / Outcome · Final Task Cost
- Revenue / Circle · Startup Unit Efficiency
- Agent Spend / Gross Profit · AI Investment Efficiency
Ultimately, the most critical point is:
For the same one-dollar revenue, how much Human Cognition is required?
These are the core economic metrics of an AI-native company.
Traditional Buyout:
Purchase Price - EBITDA Growth - Deleveraging - Multiple Expansion.
An AI Buyout should be broken down into five layers:
① Cost Compression
Agent reduces repetitive cognitive labor.
The simplest, and also the easiest to imitate.
② Organizational Compression
Reduce:
Handover Hierarchy Approval Management Synchronization.
This is often more valuable than “fewer employees”.
③ Revenue Capacity
Same sales:
100 customers → 300 customers.
Same procurement:
50 suppliers → 500 suppliers.
This is where AI truly begins to generate compounding returns.
④ Knowledge Capitalization
Human experience:
Memory → Skill → Agent → Evaluation → Asset。
Knowledge from:
Employee Asset
Turned into:
Company Asset。
⑤ Acquisition Learning Curve
This is the truly massive part of PE.
First company:
USD 1,000,000 AI transformation cost.
Second company:
USD 700,000。
Fifth company:
USD 300,000。
Tenth company:
Many agents, skills, and evaluators are deployed directly.
Thus:
The larger the portfolio, the easier it is to modify the next acquisition.
AI may bring the first scale economies to traditional roll-ups similar to a software platform.
Models will inevitably be commercialized.
The real barrier should be:
Continuous accumulation:
Workflow Graphs Agent Skills Evaluation Rubrics Industry Context Transformation Benchmarks Role Designs Circle Designs Decision Rights Compensation Models Integration Playbooks。
For example, after acquiring the 10th food supply chain company, you already know:
What the inquiry Agent should do;
What context the sales Agent needs;
Which quotes require manual approval;
How the Procurement Evaluator judges;
Which positions are most likely to become bottlenecks;
Which KPIs are false;
Which employees are most likely to become Mini CEOs.
This knowledge does not belong to GPT-6.
It belongs to GP.
This is:
A typical AI Buyout platform, I would design the core team as:
- Team Role
- Investment Partners · Sourcing deals, financing, deal structuring
- Operating Partners · CEO, strategy, governance
- AI Transformation Partners · AI-native organization restructuring
- FDE / AI Engineers · Actually get the workflow running
- Data / Platform · Harness、Ledger、Context
- Talent Partner · Find Mini CEOs and core talent
- Sector Operators · Industry capabilities
The biggest change is:
AI/FDE should not be the post‑investment “IT department.”
They should join the Deal Team from the first meeting with the target.
Even within the Investment Committee:
Deal Partner
- Operating Partner
- AI Transformation Partner
All three roles should sign.
Because a project may:
Financially very attractive,
but it fundamentally lacks AI transformation space.
Then it is not an asset that AI private equity should buy.
It is not about sending 50 consulting advisors.
For example:
1 Operating Partner + 1 AI Transformation Lead + 2 FDE + 1 Data/Process Engineer + 1 Talent/Org Partner。
On-site for 100 days.
However, they are not conducting business on behalf of the company.
The task is:
Replace the company's Operating System.
Leave after the OS runs smoothly, then go to the next company.
This is similar to 3G continuously developing its own CEO.
AI private equity should continuously cultivate:
These talents may become even scarcer than Deal Makers in the future.
Otherwise, the organization will eventually re‑financialize.
If all carry is given to the Deal Team:
They naturally optimize:
What to buy, at what price to buy, and at what price to sell.
But the real bulk of AI PE alpha comes from:
How to adjust after the purchase.
So at a minimum, it should allow:
Operating Team + AI Transformation Team
Directly share a proportionate amount of carry.
Portfolio CEOs and Circle Leaders should also share value creation through Management Equity / Phantom Equity / Profit Pool.
Only then can we combine:
Dan Weijian's Control + 3G's Ownership + AI Native's Circle
to truly bring them together.
If I were to design the first deal, I would not choose:
"the sexiest AI opportunity".
Instead, I would choose:
An ideal first‑deal profile might be:
Annual revenue 2–1,000,000,000 RMB
EBITDA 20,000,000–100,000,000 RMB
50–500 employees
Has been operating for more than 10 years;
Stable customers;
Stable product;
Industry is fragmented;
ERP/CRM exists but usage is average;
Large numbers of sales, procurement, customer service, finance, and operations staff;
Founders are willing to sell control;
Valuation is not that of a tech company;
There is also a large amount of knowledge work.
The first deal may not even require a 100% acquisition;
But it must:
Have real control.
The board should not say:
"AI can be used, but don't change the organization."
If it cannot:
Change the CEO Change the organization Change incentives Change the software Change decision rights,
Then this deal should not be done.
Do not write:
After the acquisition, AI saves 30% of labor, so the price can be increased by 30%.
Instead, it should be stipulated:
Base Case
Do not assume multiple expansion.
Debt Case
Do not count unverified AI revenue growth.
AI Efficiency
Only workflows that have already been proven by the pilot can enter the model, and the discount is accounted for.
Headcount Savings
The theoretical agent replacement rate cannot be directly included in EBITDA.
It must truly be completed:
Workflow redesign → SLA does not decline → Customer experience does not decline → Reconfirm staff redundancy.
This can prevent the “AI story” from becoming a new justification for paying an excessive acquisition price.
This is exactly the 3G/Kraft Heinz negative lesson for future generations.
I will launch it this way.
- Core timeline tasks
- 0—3 months · Build a prototype of the AI Transformation OS; identify an industry
- 3—6 months · Screen 50 companies, conduct in‑depth research on 10 of them, and complete 1 acquisition.
- 6–9 months · Complete Work Graph + 3–5 core Workflow transformations
- 9–12 months · Zero-Based Work + Circle + Incentive reform
- 12–15 months · Demonstrate improvements in Revenue/FTE, EBITDA, Cycle Time
- 15–18 months · Acquire the first bolt‑on and replicate the same operating system
- 18 months later · Raise Fund I with a real transformation track record
Therefore, the most attractive slide in the fundraising deck should ultimately not be:
AI will change how many trillion dollars of global GDP.
It should be:
Company A
At acquisition:
240 people Revenue 420,000,000 EBITDA 38,000,000.
12 months later:
190 people Revenue 480,000,000 EBITDA 62,000,000.
Meanwhile:
Quote time：48h → 4h Customers/Sales：87 → 241 Workflow Human Handoffs：7.3 → 2.1 Agent autonomy：0 → 68%。
Then:
Company B
Deploying the same Sales / Procurement / Finance stack took only half the time of Company A.
If this is achieved,
The AI PE asset class is essentially established.
Following this methodology, AI PE ultimately is not even “AI + PE”.
More precisely, it is:
Traditional PE：
Capital Allocation。
Dan Weijian:
Capital Allocation + Governance。
3G：
Capital Allocation + Governance + Management OS。
AI Native Buyout：
Capital Allocation - Governance - Organization Design - Machine Execution。
The final outcome proposed by this methodology aligns closely with PE: a company should gradually become an adaptive system that continuously discovers targets, assembles people and agents, allocates capital, validates results, and eliminates wrong directions.
Therefore, the one‑sentence definition of this project can be written as:
AI PE is not about using AI to improve the efficiency of portfolio companies, but about acquiring control of a pre‑AI‑era company and completely redesigning the “work‑power‑information‑benefit” chain, turning the company itself into an operating system that can be executed by machines, observed, learned and replicated.
Condensed into a single capital‑market phrase:
The truly urgent next step is not to write a Fund Deck, but to first produce an “AI Buyout Investment & Transformation Playbook v0.1” that includes a target screening sheet, AI‑Native DD, Work Graph template, 100‑day plan, Organizational AI Delta score, post‑investment KPI and IC template. Only when this is created does it move from “concept” to a “tradeable, executable, fund‑raising” stage.
Counterevidence
- The cross‑company reusability of the AI Transformation Playbook still needs to be validated with a second and third portfolio company; a single success case may stem from management changes, industry cycles, or transaction pricing.
- Organizational downsizing may damage customer experience, tacit knowledge, and internal trust, and the theoretical Agent substitution rate cannot be directly included in EBITDA or acquisition pricing.
- Agentification shifts bottlenecks to target definition, data permissions, exception handling, evaluation, and accountability; successful technology deployment does not equate to realized operating results.
- Without genuine control over the CEO, organization, incentives, systems, and decision rights, AI‑native transformation can easily degrade into partial tool adoption.