ZenChAIne
記事一覧に戻る
Buying Instead of Building: Applying the AI Roll-up to Business Succession in Japan

Buying Instead of Building: Applying the AI Roll-up to Business Succession in Japan

ZenChAIne·
AI AgentsBusiness SuccessionSmall BusinessAI Roll-ups

Introduction

If you want to start an AI business, one option is to take over a company that already has customers and revenue, then change how its work gets done. Entrepreneur Greg Isenberg describes this approach in “$5T opportunity: AI Roll Ups”. It is a practical piece that goes from finding candidates to a scorecard and ready-to-use prompts, and I enjoyed reading it. I could see why it drew so much attention.

Greg is writing about the United States, though. In Japan, about half of companies have no identified successor, and professional qualifications and licenses are closely tied to how many businesses operate. This article is my attempt to work out what the approach would look like here. Whether to build a company or buy one is for each reader to decide, so I set out the steps and checks that can inform that decision, keeping Greg's proposal separate from my own assessment. It also continues my previous article, in Japanese about connecting AI with industry knowledge and real work, this time from the perspective of acquiring an existing business.

Key points

  • An AI roll-up acquires existing businesses and improves their operations with AI. In Japan, it offers one approach to taking over companies that lack successors.
  • Assess candidates through conversations with owners, a consistent scorecard and actual work records. In the first 100 days after acquisition, introduce AI into a limited part of the business and expand as the results justify it.
  • In Japan, check the conditions attached to professional qualifications and business licenses, and make sure someone can connect experienced employees with AI specialists. Public support for business succession may also be available.

What Is an AI Roll-up?

An AI roll-up acquires and combines existing businesses, using AI to improve operations and increase business value. A roll-up brings together several companies in similar lines of business, but it can begin with one acquisition: improve that company's operations, then apply what you learn to the next.

Greg names accounting and bookkeeping, property management, insurance agencies and IT operations as potential sectors. They share recurring work, results that can be checked and ongoing customer relationships.

The sourcing approach, scorecard, workflow samples and 100-day plan described here draw on Greg's proposal. The sector-specific applications, worked example and approach to working with employees are my own assessment of how this could be put into practice in Japan.

Which Sectors Are Worth Considering in Japan?

In Teikoku Databank's 2025 nationwide survey, 50.1% of approximately 270,000 companies had no identified successor. That was an improvement of 2.0 percentage points from the previous year, but half of the companies surveyed still had no successor in place. By sector, the rates were 57.3% in construction, 55.2% in retail, 52.9% in services, 51.1% in real estate, 46.8% in wholesale, 45.7% in transportation and communications, and 42.4% in manufacturing.

A survey published in March 2026 by Japan's Organization for Small & Medium Enterprises and Regional Innovation, known as SME Support Japan, found that 20.4% of the 1,647 companies that gave valid responses on adoption had introduced AI across the company or in some operations. The survey covered image recognition, demand forecasting and other AI applications as well as generative AI. The two surveys cover different populations and cannot simply be overlaid, but there are likely to be many companies that have neither a successor nor AI in their operations. I see room there to change how the work gets done after a handover.

Even so, a sector's name alone tells you little about what AI could improve. Look at recurring tasks, such as invoice entry and monthly reporting in rental property management, or inquiry classification in IT maintenance.

Sector or businessWork to investigate firstDecisions and relationships that remain with people
Rental property managementRepair requests, invoice entry and monthly reports to property ownersEmergency response, approval of repairs and important negotiations with property owners
Bookkeeping and payroll servicesCollecting documents, transferring data, reconciliation and checking calculation resultsProfessional judgment, work requiring qualified professionals and explanations to clients
IT maintenance and operationsClassifying inquiries, organizing logs, drafting responses and preparing regular reportsAccountability for incidents and approval of production changes
Wholesale, construction and manufacturingOrganizing orders, preparing estimates, invoicing and maintaining work recordsCredit decisions, price negotiations and safety decisions in construction or manufacturing

In the near term, much of the opportunity lies in this information-based work. For work on site, physical AI, which operates in the real world through robots, is another possibility, but equipment costs and safety need to be part of the assessment. The Japanese government's March 2026 roadmap also distinguishes the timelines for tasks such as patrols and transport from those requiring dexterity and situational judgment. For someone taking over a construction company, I think the realistic sequence is to reduce administrative work first, freeing up skilled workers' time, then pursue automation of complex work on site over the medium to long term.

Find Acquisition Candidates in Person, Too

Let AI help assemble a list of candidates, then meet the owners and see the companies for yourself. Greg recommends finding prospects through public professional and licensing registers, industry association directories and local information, while handling outreach to owners personally.

In Japan, you can narrow the search to a region and a sector, then work through industry associations, chambers of commerce, regional financial institutions and Business Succession and Handover Support Centers. For example, you might list rental property management companies within traveling distance, then use introductions and visits to establish whether an owner is open to discussing a handover.

Which customers have you worked hardest to serve? Who keeps the day-to-day business running? What would you want preserved after you retire? A company list produced by AI cannot answer those questions, so I would want the first conversation to take place in person. Visiting is also worthwhile because it lets you build relationships with owners before their companies appear widely on the market.

Greg lists business-for-sale marketplaces among his sources too, but he does not rate them highly. Most small firms that would sell are not listed anywhere, and their owners have not decided yet, so in his view the buyer's job is to find them before a broker does. Once a company reaches a marketplace, much of the advantage is gone.

In Japan, M&A intermediaries and business-for-sale platforms can also help with sourcing and negotiations, but the companies listed there have already decided to sell. Even with an introduced opportunity, I would still want to meet the owner and the people doing the work and form my own understanding of the business.

For each candidate, record the location, work performed, employee roles, recurring contracts and the owner's intentions for succession. Separate assumptions based on public information from facts the owner has confirmed, and keep a list of questions for the next meeting. The research you had AI do then becomes preparation for that meeting.

How Can a Scorecard Help Compare Candidates?

Once a company looks promising, compare it with others using the same scorecard. Greg's criteria and weights translate into the following questions for a Japanese acquisition.

CriterionOriginal weightWhat to check in meetings and documents
How repetitive is the work?20%Volumes of monthly reports, incoming requests, invoices and similar tasks
Can the results be verified?20%What billing amounts, contracts and processing results can be checked against
Is there recurring revenue?15%Recurring contract terms, renewal rates and reasons for cancellation
Does the seller want to support a handover?15%Desired timing, how long the owner can stay to teach others and non-negotiable conditions
Do customers have durable relationships with the company?15%Whether relationships continue through employees other than the owner
Is the data organized?10%Whether past cases can be found and contracts connected to billing records
Is the market made up of many small operators?5%Whether operating methods developed in one company could be applied to others

Keep the supporting documents and verification date for every score. For recurring revenue, that means checking contracts and renewal records. For customer relationships, it means examining whether customers also work with people other than the outgoing owner.

Greg says to score every target the same way so that your excitement doesn't make the decision, and he attaches rough thresholds to the total. A firm scoring under 60% is usually “a job, not an acquisition,” and one scoring above 80% deserves a letter of intent. That is Greg's view, and the accompanying Excel workbook does not judge candidates automatically against these thresholds.

The scorecard and interview questions are included in the accompanying Excel workbook. A plain-text copy of the prompt is also available. Both downloads are in Japanese.

Use the workbook to compare companies while recording the evidence behind each assessment. To adapt it to a different sector or region, give the following English prompt to Claude or Codex.

A Prompt for Creating the Scorecard

text
No external website access is needed. Using only the conditions below, create an Excel workbook for comparing candidates for small-business succession in Japan, together with a set of interview questions.
Target sector: [Enter the sector]
Target region: [Enter the region]
Number of candidates: [Enter the number of companies to compare]
 
Use Greg Isenberg's AI roll-up assessment criteria and keep these weights unchanged:
Repetitive work 20%; verifiable results 20%; recurring revenue 15%; seller's willingness to support a handover 15%; durable customer relationships with the company 15%; data organization 10%; a market with many small operators 5%.
 
For each criterion, provide at least three questions to ask the owner or responsible employees, and list the documents to inspect. For recurring revenue, for example, ask about the scope of recurring contracts, renewal rates and reasons for cancellation, and examine contracts and renewal records. Label any questions or scoring guidance added beyond the seven criteria and their weights as "Supplementary suggestions for Japan."
 
Create these worksheets: "Instructions," "Candidate Comparison," "Evidence Log," "Interview Questions" and "Separate Checks." The Evidence Log should record verification status, rating, evidence, document name, verification date, respondent and unresolved questions. Link its entries to Candidate Comparison. Use a 1–5 rating scale, with concrete descriptions of what scores of 1, 3 and 5 mean for each criterion.
 
Leave unverified entries blank. Do not invent scores. Show the number of verified criteria, and do not show a total score until all criteria have been verified and supported by evidence. Use different colors for input cells and calculated cells, and provide a way to check that the weights add up correctly.
 
Keep the acquisition price, loan repayments, working capital, off-balance-sheet liabilities, permits and licenses, and retention of qualified professionals and key employees separate from the score. Do not create an automatic rule for deciding whether to acquire a company. Make the workbook useful for identifying what to investigate in the next meeting. Do not invent information about real companies or figures that have not been provided.
 
Before saving the file, test the calculations with no inputs, with all criteria completed and with some criteria still unverified. Also check the workbook for clipped text.

The questions and scoring guidance are my additions, intended to make Greg's seven criteria usable in Japanese business succession. If you give company documents to an AI tool, agree with the seller on how they may be used, and remove personal and confidential information first.

Examine the acquisition price, loan repayments, working capital and whether the required qualified professionals will stay separately from the total score. The scorecard shows which companies deserve further investigation and which weaknesses need checking. Whether to buy is a judgment that combines the total with these separate checks.

Build an Automation Map from Actual Work

Use records of completed work to test where improvements may be possible. Greg proposes collecting 20–50 anonymized examples of core workflows, then dividing tasks into four groups: automate immediately, use with human review, provide assistance only, or keep with people.

The automation map is a list of each task's inputs, steps, outputs, volume, time required and consequences of errors. Agree with the seller on how the records may be used, and include cases with missing documents or rework alongside routine cases.

For a rental property management company, you could compare inquiry histories, invoices and the resulting data entries, and monthly reports to property owners. The table below illustrates that assessment.

TaskAI's roleHuman review and judgment
Classifying documents and standardizing file namesOrganize files according to agreed rulesValidate the initial results and return unreadable documents to a person
Invoice entry and monthly reportingExtract amounts and draft reportsReconcile against original documents and accounting data, then approve sending
Repair inquiriesOrganize the request and identify missing information and possible responsesDecide urgency, the scope of the work and who bears the cost
Important negotiations with property ownersSummarize the historyExplain the situation, negotiate terms and reach agreement

Adding volumes and processing times helps set testing priorities. In a hypothetical calculation, entering 300 invoices a month at 12 minutes each takes 60 hours. If human review and correction after AI processing still take an average of five minutes per invoice, that falls to 25 hours, a difference of 35 hours.

Decide whether those 35 hours would reduce overtime or allow the company to manage additional properties, add AI usage fees, initial setup and maintenance costs, and compare the business before and after the change on the same basis. That gives you an estimate of the business effect.

Separate the AI Preparer from the AI Reviewer

Greg describes a design with separate roles for preparing drafts and checking them, with neither allowed to send anything to a customer on its own. In invoice processing, one AI could prepare the proposed entries, a second could look for discrepancies against contracts and original documents, and the responsible employee could give final approval.

Even when the reviewer reports no problems, both AIs may have missed the same error. Use conventional calculations and reconciliation for mechanically checkable results, such as matching amounts, and assign responsibility for professional judgments and sending the output.

Record Experienced Employees' Judgment as Working Rules

Review sample cases with experienced employees and ask where new hires tend to make mistakes, which conditions always require checking and whom to consult. This is a way to document the tacit knowledge Greg emphasizes during everyday work.

For example, a rule might say: if a repair exceeds the approval threshold in the contract with the property owner, return it to the responsible employee without placing an order. Record the underlying contract, exceptions and responsible person as well. When someone corrects an AI draft, keep the input documents, the correction and the reason, then add recurring corrections to the rules after that person has checked them.

After changing a rule, run previously mishandled cases again and confirm that the same mistake no longer occurs. Make that recheck part of the operating routine.

Someone Needs to Connect Experienced Staff with AI Specialists

I believe a person who can work between the technical team and the people running the business will be particularly important when applying this approach in Japan. Employees who have relied on manual processes for years may need more than a new system handed to them before they can use it with confidence.

My own career has been entirely in IT, so my example comes from there. Production deployments are sometimes carried out by two people checking each step together, even where automation would seem to let one person do the job. Ask why that check is repeated every time and you may uncover a past incident or a promise to a customer. Understand that history, then decide together which checks to preserve and which tasks to change.

The role also involves listening when people find a tool difficult to use and bringing that experience back into the AI design. It is close to the work of an FDE, or forward-deployed engineer, who works closely with a client's team to understand its operations and carry out development and implementation.

The owner either takes responsibility for these conversations or brings in a partner who is good at them. If the buyer's own strengths are primarily in AI and technology, I would also use interviews to establish whether someone at the target company can connect experienced employees with technical specialists.

In a family-run business, pay attention to work that never appears in a job title. Understand the bookkeeping and customer support handled by family members, along with the small adjustments experienced staff make to keep things running, then agree on who will take over and how long the handover will last.

What Should the First 100 Days Establish?

Greg's 100-day plan starts with observation that limits disruption to customers, moves into a small live deployment, and expands after the results have been checked. Applied to a small Japanese company, it could look like this.

PeriodWhat to doWhat to check before moving on
Days 1–30Meet employees individually. Visit customers with the former owner. Test AI alongside the work people already doHave you identified the responsible employees, exceptions and conditions in which AI makes mistakes?
Days 31–60Put one high-volume, low-risk workflow into live use, with people reviewing every AI outputHave you measured total preparation and review time, corrections and the burden on staff?
Days 61–100Expand to other tasks where the results support it. Review correction records with experienced employeesIs quality holding up, without putting undue pressure on customers or key employees?

For the first 30 days, avoid abruptly changing customer contacts or commitments, and test against both past cases and current work. If invoice entry moves into live use over the next 30 days, record the amounts and fields people correct, and why, as well as processing speed. Once those results are stable, choose the next one or two workflows, such as monthly reporting.

Decide which measures to review each week. A shared view of transaction volumes, human preparation and review time per case, correction rates and reasons, profitability including AI costs, and retention of customers and key employees can reveal changes such as faster processing accompanied by a heavier review burden.

Define the correction rate as the number of cases corrected by a person divided by the number processed by AI so that weekly comparisons are consistent. Separate minor wording edits from incorrect invoice amounts, and keep sight of serious errors even when they are rare. Human time should include time spent instructing the AI, checking its work and handling rework.

By day 100, you should know which tasks the system can handle reliably, how much human involvement they require and what has improved for the company. Establishing that operating method in one business also prepares you to take over another.

Dividing Work Between Qualified Professionals and AI Specialists

In Japan, tax accountants and labor and social insurance consultants have certain duties reserved to qualified professionals. Those professionals may have little experience improving workflows with AI, which leaves room for a partnership between someone with the professional knowledge and someone who can build the AI-enabled workflow.

In a tax accounting practice, for example, an AI specialist could improve document collection and reconciliation while the tax accountant handles tax judgments and explanations to clients. Subject to statutory exceptions, only a tax accountant may carry out tax representation or prepare tax documents as a business, so an AI specialist cannot run the practice alone. The business has to be designed starting from which tasks are shared with the qualified professional.

Profession or businessMain work tied to qualifications or licensing in JapanDivision of responsibility with an AI specialist
Tax accountant, or zeirishiTax representation, preparation of tax documents and tax consultationsSupport document collection and reconciliation, while tax judgments and client-facing professional work are performed under a legally compliant arrangement with qualified professionals
Labor and social insurance consultant, or sharoshiPreparing applications under labor and social insurance laws, submitting them on another's behalf, acting as an administrative representative and preparing statutory books and documentsSupport data entry and checking, with qualified professionals handling reserved activities
Attorney, or bengoshiLegal opinions, representation, settlements and similar handling of legal mattersSupport research, organization and drafting, while acceptance of legal engagements, legal judgments and handling of matters take place under a legally compliant arrangement
Real estate transaction specialist, or takkenshi, and licensed real estate businessThe individual specialist explains important matters and places their name on the important-matters document and statutory document issued when a contract is concluded, known as the Article 37 document. The business needs a license to conduct activities such as real estate sales and brokerage as a businessSupport preparation of property information and draft contracts. Check the individual's qualification separately from the business license and requirements for dedicated real estate transaction specialists

These activities are reserved when a tax accountant's work is carried out as a business at another person's request, and when an attorney's work is handled as a business for the purpose of compensation. Both are subject to statutory exceptions. For labor and social insurance consultants, the restrictions generally apply when the activities are performed as a business for compensation at another person's request, and not all HR and labor advice is reserved work. Check the provisions cited below for the exact scope.

Sources: National Tax Agency, Labor and Social Insurance Consultant Act, Attorney Act, and Ministry of Land, Infrastructure, Transport and Tourism guidance on real estate business licenses.

In a partnership, establish which person or entity accepts professional engagements and who is responsible for the work, with the AI specialist supporting the qualified professional's work. An arrangement that merely borrows the professional's name is not acceptable. Professional corporations also have restrictions on matters such as who may invest in them, so an acquisition or partnership needs a structure that complies with the rules for that profession.

Reduce the Time and Cost of Starting from Scratch

Taking over an existing business gives you customer and supplier relationships, employees who know the work, and an operating setup that would otherwise need to be built from scratch. In a licensed business, it may also reduce some of the application and preparation time involved in starting a new operation.

For example, a new real estate business needs to arrange its license, dedicated real estate transaction specialists, and a business security deposit or membership in a guarantee association. Acquiring the shares of the existing legal entity may allow the business to continue with those arrangements in place, although required change procedures still apply.

Similarly, taking over the same legal entity and operating setup may reduce the effort involved in obtaining new permissions for a company with a liquor sales license or a cosmetics manufacturing and marketing business license, known as a seizō-hanbai license. How much startup preparation can be avoided is one factor in assessing the acquisition.

After establishing what preparation can be avoided and what business operations can be carried forward, compare the purchase price, debt and additional funding needs with the cost of starting afresh. A real estate business license does not simply transfer when only the business is acquired, while a share acquisition also leaves off-balance-sheet liabilities and other obligations within the acquired company. What you take on, both assets and burdens, depends on how the deal is structured.

Public Support for Business Succession in Japan

Japan's Business Succession and M&A Subsidy may cover eligible small and medium-sized businesses' professional fees for intermediation, advice and pre-acquisition due diligence, as well as professional fees and capital investment associated with post-acquisition integration. Consider this support separately from the price paid for shares or the business itself. Some expenses become ineligible if work begins before the grant decision, so check the application guidelines for eligible costs and permitted start dates before proceeding with an acquisition or engaging advisers.

As of November 10, 2026, applications for the 16th round had closed on November 6, so check the official website for future rounds. Business Succession and Handover Support Centers are available for consultation, and Japan Finance Corporation's business succession financing is another option for funding that must be repaid.

Summary

Having written this far, I asked myself what ZenChAIne could do for someone who wants to carry out a roll-up like this. One answer is to work on site as an FDE or fractional CTO, inventory the acquired company's workflows together with its people, and stay alongside them while AI tools are built and put into use.

The other is to help the incoming owner raise their own AI literacy through Claude Code courses, so that they can use AI for development themselves. I see both as ways of supporting, from the technical side, the people who change a business while making use of the experience of qualified professionals and long-serving employees.

References