A buyer typically expects the practice's accounts, an anonymised client and fee schedule, work-in-progress and debtor records, staff and software details, engagement terms, and selected files later in diligence. AI can index those documents, reconcile schedules, flag missing evidence and draft a question log. Keep identifiable client material behind a controlled disclosure stage reviewed by advisers.
The pack should answer a buyer's practical question: what revenue will transfer, what work and liabilities come with it, and how much depends on the seller? A folder full of exports rarely answers that. Build a numbered index, freeze each source file, and make every AI summary point back to a specific record and date.
Start with a staged index, not a shared drive dump
Create three access stages. Stage one is an anonymous overview with service mix, fee bands and financial history. Stage two, after appropriate confidentiality and professional checks, contains more granular but still controlled client and operational schedules. Stage three is selected identifiable files and introductions when the legal and professional basis permits. An NDA, or confidentiality agreement, helps govern the buyer's conduct but is not a universal permission to expose client working papers.
One professional body's guide to selling a practice lists what buyers want once a non-disclosure agreement is signed: a detailed client list showing recurring fees and work done, staff roles with salary, benefits and hours, a sample of working papers, the software used for each task and three years of the practice's own accounts. It also says client names are not shown until the sale is agreed. The same body's helpsheet on buying and selling fees says client authority would usually be needed before disclosing confidential client information, that the client's consent to buyer access to files should be obtained, and that ownership of and access to files for any later claim should stay with the seller. The precise duties vary by engagement and professional body, so have a qualified adviser review the release plan rather than treating the guide as a legal transfer rule.
Give every pack item an ID, owner, source period, disclosure stage, access group and status. For example, “F03 — recurring fee schedule — last completed year — stage one — redacted — owner review pending”. AI can convert a rough file list into an index and identify missing IDs. It must not decide that an item is safe to release from its filename. A folder called “client list anonymised” might still contain addresses in a hidden spreadsheet column.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowReconcile the practice's income story
Provide the buyer with practice accounts for the requested periods, plus a bridge from invoiced fees to recurring, one-off and non-recurring income. Export fees by client code, service, invoice date, period served and amount. An invoice paid this year for prior-year work should not automatically count as this year's recurring run rate. If the practice uses fixed monthly fees, show how many months were actually billed and whether any catch-up or write-off changed the figure.
For an illustrative six-person practice, total fees in the last complete year are $720,000. The ledger shows $510,000 of recurring accounts, tax and payroll work; $150,000 of project work; and $60,000 of unusually large one-off assignments. The buyer's first-page schedule can show those three buckets, not a single claim that the firm has “$720,000 recurring revenue”. AI can classify invoice descriptions, but a manager must verify a sample and all large items against engagement letters and invoice history.
Suppose the top ten clients generated $260,000. That is about 36% of total fees. Show the concentration using coded clients, plus the services, tenure and contract terms. Do not infer that every dollar will renew. A buyer will ask whether the seller personally maintains those relationships, whether fees were discounted, and whether the clients have consented to any transition. The schedule can identify “client C014, $42,000, accounts and advisory, relationship owner: seller” without revealing the client's name at the first stage.
Use AI to create a variance table: client code, prior-year fee, current-year fee, change, reason to ask. If C014 rises from $28,000 to $42,000, the $14,000 increase could reflect an expanded recurring scope or a one-off transaction. The tool should flag it, not explain it without evidence. Send the question to the relationship owner. Link the signed scope change and invoice run before describing the increase to a buyer. For data cleaning before any analysis, see the messy-data tutorial.
Show WIP, debtors and deferred work separately
Work in progress, or WIP, is effort or jobs not yet billed. It can include recoverable work, unapproved overrun, forgotten time and tasks that will be written off. Export the WIP ageing by job and client code, then compare it with subsequent invoices and cash receipts. Keep the source date visible. A buyer comparing WIP at two different dates may mistake a normal billing cycle for a sharp deterioration.
In the illustrative practice, gross WIP is $82,000. After reviewing job status, $48,000 is ready to invoice, $19,000 awaits a client milestone, and $15,000 is old or disputed. AI can sort lines by age and spot missing job codes, but it cannot declare the $15,000 collectible. The owner records whether each item is covered by a signed engagement, whether the client has accepted the work and whether an invoice was raised after the cut-off. The buyer sees both gross WIP and a transparent, justified adjustment.
Debtors need their own ageing, credits and subsequent-receipt schedule. A $9,000 invoice that is 120 days old might be routine for a slow-paying client, or it might be disputed because the scope changed. AI can match invoice numbers to payments but should not invent the reason for delay. Keep credit notes and bad-debt provisions visible. If a project was billed in advance, include the deferred obligation: the buyer may inherit delivery work even though the cash was received by the seller.
A concrete error illustrates the risk. The AI summary says “$12,000 of aged WIP was collected after year-end” because it matches a receipt from the same coded client. The receipt actually paid a different invoice. The reviewer must match job, invoice and payment reference, not client code alone. A false match can inflate both the claimed recoverability of WIP and the perceived quality of debtors.
Describe clients without revealing them too soon
A buyer needs service mix, fee level, tenure, concentration, delivery effort and retention risk. An anonymised client schedule can supply these without names at the opening stage. Use stable client codes so a buyer can trace the same account across fee, WIP and debtor schedules. Remove free-text notes, email domains and uncommon industry descriptions that identify a business in context. A client with a rare service and a very large fee may be recognisable even without a name; band or suppress that row until the disclosure gate permits more detail.
For example, rather than “sole manufacturer of a named specialist device, $67,420”, stage one might show “specialist manufacturing client, fee band $60,000–$70,000, accounts and advisory”. The exact fee and name can remain in a restricted reconciliation table. AI can propose redactions, but a person who knows the practice must test whether the combination of details is still recognisable. The client-data anonymisation tutorial is useful for that check.
Before a buyer opens any client files, determine who owns the records, what the engagement terms say, whether client authority is needed and which people may view the files. Professional guidance expects the client's consent before a buyer is given access to files, so do not assume the firm can simply volunteer working papers. A counsel-reviewed access log should record the reason, scope, viewer, date and permitted use. If the buyer only needs to verify quality of work, a representative set of permissioned files may answer the question without opening the entire archive.
Include staff, systems and obligations
Staff records tell the buyer whether the service can continue. Prepare an anonymised schedule of role, hours, tenure, salary and benefits, qualifications, key responsibilities and current vacancies. Add a capacity view: which staff handle the top client codes and which jobs only the seller can perform. Avoid handing over personnel files or private notes in the first pack. An adviser should review employment and transfer obligations where relevant.
In the six-person illustration, the seller handles 30% of advisory fees and signs off complex jobs; a senior manager handles 70% of recurring accounts work. That matters more than a generic statement that “the team is experienced”. AI can map assignments from a practice-management export, but check stale owners and shared jobs. If the manager is leaving in two months, the capacity story changes. Present the resignation as a dated fact and the replacement plan as a plan, not a signed certainty.
Inventory the software that runs the practice: accounts production, tax, payroll, client portal, document storage, billing and workflow. Record subscription owner, renewal date, data-export path, transferability and integrations. Do not claim the buyer can take over a licence until the vendor confirms it. Include an access map showing where client records sit, with credentials excluded. The accountant engagement-letter tutorial can help organise scope fields, but a human must check each client's actual signed terms.
Ask AI for a missing-document matrix
Give the assistant an approved index and redacted list of available files. Ask it to return “request”, “present”, “period mismatch” or “human review”, citing the file ID for each conclusion. Do not upload a whole client drive. The strongest output is a short question list with evidence pointers, not a paragraph that says the practice is “diligence ready”.
Compare this buyer request list with the approved pack index.
Use file IDs only. Do not infer a document exists from a similar title.
Return: request item | matching file ID | period | status | owner question.
Flag client-identifiable items for restricted review.
If an item cannot be matched, write MISSING.
Buyer requests: [approved list]
Pack index: [redacted file IDs, titles and periods]One illustrative output says “three years of accounts — F01, F02, F03 — present”. A reviewer opens F02 and finds it is a draft with no final adjustment for partner drawings. Mark the row “human review”, replace the file only after checking the signed accounts, and preserve the draft in a restricted source folder. The file ID alone did not prove quality. Another output flags a missing insurance record; the owner asks the broker what period and cover evidence the buyer actually needs rather than uploading a binder indiscriminately.
A second illustrative output identifies 34 engagement letters for 280 active clients. That may indicate missing documents, or simply that letters are stored in another system. Search by client code and service, not by filename alone. Record “located”, “unsigned”, “outdated scope”, “renewed under master terms” or “missing”. Do not manufacture a completed-letter count from the AI's search results. The fact that the system could not see a file is a request to investigate, not proof the firm failed to obtain it.
Control the buyer's questions and pack revisions
Use a question log with request ID, buyer question, source file IDs, responsible person, response status, disclosure stage, adviser approval and release date. This prevents contradictory answers from the seller, broker and manager. If a buyer asks why recurring fees fell by $18,000, the answer should point to the coded client movement schedule and an approved explanation. A free-form AI response is too easy to send before anyone notices that it contains a client name.
For a filled example, Q17 asks about $18,000 of lost recurring fees. The schedule shows $11,000 from one client that ceased trading and $7,000 from a change in service scope for another. The manager verifies the dates and invoices. The buyer-facing answer uses client codes, states the two documented reasons and notes whether replacement fees are included elsewhere. It does not say “revenue has recovered” unless the later invoices support that claim. AI can draft the sentence from approved facts; a reviewer checks it against the register.
When an updated schedule changes a number, revise its version and the pack index. Suppose WIP moves from $82,000 to $77,000 after a $5,000 duplicate is removed. The buyer's Q&A, summary and data-room copy must all reflect $77,000, with the reason visible. Keep the previous version for an audit trail. If several people can edit one live spreadsheet without a release owner, the pack will contradict itself even when every source document is correct.
Run a cold-reader test before release
Ask a colleague who did not build the pack to answer five questions using only its permitted-stage contents: What is the recurring fee total? Which clients or service lines create concentration? How much WIP is old or disputed? Which work depends on the seller? Which documents remain restricted? If they cannot find the answer and its source within a few minutes, improve the index and explanation before giving access to a buyer.
Then test confidentiality. Search the export for client names, email addresses, hidden columns, comments, document properties and embedded links. Open a sample of supposedly redacted files as the buyer's account would see them. Remove access when a buyer drops out. A practice sale is a transaction involving people who have entrusted the firm with their affairs; the buyer's legitimate need for evidence does not make every client record immediately shareable. A controlled, traceable pack makes diligence faster because each release is deliberate.
Make the release checklist specific to each file
Before an item changes from draft to available, the release owner should answer five questions on its index row. Does the file match the requested period? Does its total agree with the pack's summary? Does it contain information beyond the buyer's question? Does the current disclosure stage allow that information? Has a second person opened the exact released version? A blanket tick beside the folder cannot answer those questions when one spreadsheet has several sheets and a hidden comments column.
Consider an illustrative fee workbook with a visible sheet showing coded clients and $510,000 recurring fees. A hidden tab contains names, personal email addresses and notes about payment disputes. Exporting the visible sheet as a values-only file and rechecking its total is safer than sharing the original workbook and relying on the hidden setting. The manager then checks that the new export has no formulas pointing to the restricted source. The released file gets its own ID and date so it cannot be confused with the working copy.
A sample working paper needs a different check. A buyer asks to see how review points are closed. Choose a file for which access is authorised, remove unnecessary personal details only if the redaction does not distort the evidence, and include the actual review sign-off trail. If the only available file is unusually tidy or was reworked for the sale, disclose that it is a sample and do not imply it represents the whole practice. Professional guidance on working-paper access makes this a question for an appropriate adviser and client-authority process, not an automatic data-room upload.
Separate buyer evidence from the seller's claims
The seller may believe the practice has exceptionally loyal clients, low rework and transferable relationships. Put each assertion in an assumptions log with the supporting measure. For loyalty, show retention by coded client over several periods and explain exits. For rework, use recorded review corrections if the system captures them; do not invent a quality score. For transferability, list who manages each material relationship and whether introductions are planned. AI can organise the claims and spot missing evidence, but it should not convert a seller opinion into a factual statement.
As an illustrative check, the seller says that 95% of clients renew annually. The fee schedule shows 280 active clients last year and 264 of those still active now. That is about 94% by client count, but the 16 departures may represent much more or less than 6% of fees. Calculate both count and fee retention from the same defined cohort. Exclude newly acquired clients from the retained cohort, and check whether dormant accounts were counted as active. A buyer will notice when a retention percentage changes definition between pages.
Likewise, if the seller says “no single employee is critical”, compare assignments with the staff schedule. If one payroll specialist handles 85 of 90 payroll clients and no documented substitute exists, the evidence contradicts the claim. The pack should show that dependency and any cross-cover plan. The purpose is to make the buyer's questions answerable, not to guarantee the sale. A candid weakness with a source and plan is more credible than an AI-polished assurance that cannot survive a file review.
Buyer-pack questions before disclosure
Should the first buyer pack contain client names?
Usually start with an anonymised fee and service schedule that shows concentration, recurring work and trends. Professional guidance on practice sales says client names are not shown until the sale is agreed. Obtain legal and professional advice about confidentiality, client authority and the staged release of identifiable files.
Can AI decide whether WIP will be collected?
No. AI can reconcile work-in-progress entries to jobs, invoices and subsequent receipts and flag old or unbilled items. A practice owner must judge whether the work is complete, billable under the engagement, disputed or likely to be written off. Keep the original ledger and a dated adjustment log.
Is a confidentiality agreement enough to share client working papers?
A buyer confidentiality agreement is a useful control, but it does not by itself settle the practice's duty to clients or document ownership. Professional guidance says client authority would usually be needed before disclosing confidential client information, and the client's consent should be obtained before a buyer sees the files. Check your professional rules, engagement terms and legal advice before access.
Further reads
- AI Readiness Checklist for Accountants, Solicitors, Consultants — Assess whether the hand-over processes are documented.
- How to Automate Client Onboarding in a Small Accounting Practice — Show the buyer how new work is accepted.
- Seller Discretionary Earnings With AI: Recast and Find Add-Backs — Separate supported adjustments from seller claims.
- 7 AI Mistakes That Put an Accounting Firm's Client Trust at Risk — Review confidentiality risks during the sale.
- HubSpot for Accounting Firms: Lead Tracking and Onboarding Pipelines — Document pipeline hand-offs if the practice uses a CRM.
- Confidential Information Memorandum AI: A Business Broker's Workflow — A broker's workflow for drafting a CIM with AI: outline, financials into narrative, a line-by-line fact-check and confidentiality rules, plus Flippa's builder.
- Due Diligence Checklist for Selling a Business: Build One With AI — A seller-side due diligence checklist by category, plus how a broker uses AI for the data room index, the gap tracker and the buyer question log.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: ICAEW, The accountant's guide to selling your practice; ICAEW, Buying and selling fees helpsheet. Checked 4 November 2026.