Before close: the nine levers that set the price, the structure and the ceiling

Why this chapter is longer than its software equivalent

In the software edition, diligence had eight levers and most of them were about reading things faster: code, contracts, customer calls, ledgers. The asset was legible. A services firm is not. Its method lives in the heads of a dozen people, its revenue is a mix of hours, projects and retainers that the billing system may not distinguish, its relationships are personal, and a share of its work may be reserved by law to a licensed human. The conversion ceiling, the number that bounds every lever in the rest of Act I, is set here or not at all.

Nine levers act before or at close. Seven move the price, the structure and the ceiling. Two (P8 and P9) are executed in the first hundred days but decided in diligence, and they are here because a buyer who has not planned them before signing will not get to them in time.

The five conditions apply from the first conversation. Every item below carries a P&L line or a model line, a baseline the firm's own data can supply, an owner on the deal team, and a date by which it is answered or the deal is repriced.

P1. Work decomposition audit

What it is. Every service the firm sells, broken into the units of work that deliver it, and every unit tagged one of three ways: codifiable (the inputs, rules and outputs can be specified and the work done by the platform with a human reviewing), judgment (the work depends on experience the firm cannot yet write down), or reserved (a licensed or accredited human must perform or sign it, by law or by accreditation body). A supplier-compliance assessment, for example, decomposes into data requests to suppliers, chasing and validation of responses, substance screening against regulatory lists, gap analysis, report drafting, client review and change monitoring. Most of those units are codifiable; the gap analysis and the client conversation are judgment; nothing is reserved. A statutory audit decomposes differently and ends in a signature that is.

Line it moves. Price, and the conversion ceiling: the share of delivered hours that can move to the platform at all. Every D-group lever is bounded by it.

Typical range and evidence. Not independently measured. The figures in circulation are investors' targets: General Catalyst speaks of automating 30 to 50 percent of a services firm's tasks, and its IT-services vehicle reported pilots finding 38 percent of a managed-service provider's tasks automatable.¹ Both are the acquirer's own figures. Treat them as the hypothesis the audit tests, not as a benchmark. Grade: Weak (investor-reported); Strong as a method.

How to do it in diligence. Take the firm's time records for the trailing twelve months (if it bills hourly or tracks utilization, it has them; if it does not, that is a P2 finding). Map recorded hours to service lines, then to units. Sample twenty engagements across the three largest service lines and have the platform team, not the deal team, tag each unit. The output is a table: service line, unit, hours per engagement, tag, and the platform's current capability against it (which feeds P6). The codifiable share of hours, weighted by cost, is the ceiling.

Conditions. The firm's time data must be usable at the unit level, or the audit has to be rebuilt from engagement files and interviews, which doubles its cost and halves its precision. The tagging must be done by people who will build the conversion, so they own the estimate.

Kill criteria. Codifiable share of delivery cost below 40 percent: the firm is a judgment business and belongs in a conventional services thesis, not this one. Reserved share above 30 percent: see P5; the pyramid cannot be reshaped and the ceiling is a legal one.

Metrics. At close: codifiable, judgment and reserved shares of delivery cost, by service line. Year one: share of codifiable units actually on-platform (this becomes D1's numerator). Year three: judgment share re-tagged, because the platform learns and the line moves.

Horizon and stage. Deploy; diligence.

Across the five firms. A (compliance consultancy): codifiable share high, reserved near zero; the audit mostly confirms. B (accounting): codifiable high in preparation, reserved at the signature; the audit must separate the two or the ceiling is overstated. C (legal process): the line between process and advice is a jurisdictional question as much as a work one; tag with counsel in the room. D (technical services): physical and accredited units are reserved by nature; the audit decides whether to bid. E (revenue cycle): nearly all codifiable; the audit is about error cost, not share.

P2. Revenue quality and pricing structure

What it is. The revenue, taken apart by how it is earned. Hourly, fixed-fee, retainer, subscription, contingent or success-based; by client, with concentration; by renewal history, with the firm's own definition of a renewal tested against what the invoices show; by realization (billed as a share of recorded hours) and utilization (billable as a share of available hours), the two numbers a services firm manages to and the two that disappear after conversion.

Line it moves. Price, and the representations: a firm that calls project revenue "recurring" because the client comes back every year is describing a habit, not a contract.

Typical range and evidence. The pricing mix varies by sector more than by firm. In small-firm accounting the conversion away from hourly billing has largely happened: a 2025 benchmark of US accounting and tax firms by a billing-platform vendor found fixed fees the dominant model at 54 percent of firms and hourly billing under 4 percent.² In compliance consulting and legal process the hourly share remains high enough that the authors of the strongest case against this thesis made it their second reason: under hourly billing, efficiency cuts revenue, so firms revert to maximizing hours.³ The public market now prices the sector accordingly; a banker's professional-services index traded at 1.0x revenue and 6.1x EBITDA on June 30, 2026, from 2.2x and 13.4x a year earlier.⁴ Grade: Medium (sector benchmarks); the firm's own mix is Strong, because it is in the invoices.

How to do it in diligence. Rebuild revenue from invoices, not from the management accounts. Classify every invoice line by pricing basis. Build the client cohort table by first-invoice year. Compute realization and utilization from the time system. Then ask the question the rest of the playbook depends on: for each service line, what would the client have paid for the outcome if no hours had been recorded? The answer is the fixed-fee price P9 will test.

Conditions. Invoice-level data for at least three years. A billing system that records pricing basis, or an analyst to infer it.

Kill criteria. Top client above 25 percent of revenue with a contract shorter than the conversion period. Realization below 80 percent (the firm is already discounting hours it cannot defend, and a fixed fee will land lower than the model assumes).

Metrics. At close: recurring share by the contractual definition; hourly share; realization; utilization; top-ten concentration. Year one: recurring share (this is the scorecard's first line). Year three: realization and utilization retired as management metrics (D8).

Horizon and stage. Deploy; diligence.

Across the five firms. A: retainers often exist but are annual and cancellable; count them as recurring only if renewal history says so. B: mostly done; P2 is about tiering headroom. C: hourly and per-matter; P2 sets the P9 price. D: project-heavy with renewal permits as the recurring seed. E: already outcome-priced (percentage of collections); P2 is about payer concentration.

P3. Relationship and key-person map

What it is. For each client above one percent of revenue, the person at the firm who holds the relationship; for each such person, their compensation structure, their equity or partnership interest, their non-compete and its enforceability in their jurisdiction, their stated intentions, and the share of their time that P1 tagged codifiable. The last column is the one that matters: it is the list of people whose billable role is about to be automated and whose relationships the firm cannot afford to lose.

Line it moves. Price, and structure: who rolls equity, who gets an earn-out, who is paid to stay and in what role.

Typical range and evidence. There is no independent evidence on partner departure or client attrition after a change of ownership in professional-services firms. The first study of it, funded in May 2026 by the Foundation for Auditing Research and covering partner turnover and client outcomes at PE-backed audit firms, has not reported.⁵ Industry commentary speaks of partner turnover accelerating under private equity ownership without data. The nearest number is from a different population: an executive-search firm cites AlixPartners finding 58 percent of PE-backed company CEOs replaced within two years.⁶ A relationship business cannot absorb that. Grade: Weak (no sector data); this lever runs on the firm's own map.

How to do it in diligence. Interviews with every partner and every account lead, separately, before the letter of intent is public if the structure allows. Pull the client list against the CRM and the email system to confirm who actually talks to whom, because the official owner and the real one differ. Score each relationship holder on three things: revenue held, codifiable share of their own work, and flight risk. The people in the top-right of that chart are the structure.

Conditions. Access to partners before signing. Honest answers, which requires that the conversion story be told to them as a role change with a future, not as an efficiency program.

Kill criteria. More than 40 percent of revenue held by people whose non-competes are unenforceable and who have not committed to roll. Any relationship holder above 10 percent of revenue who declines to meet.

Metrics. At close: revenue held by each named person; rollover committed; roles defined. Year one: named-person retention; client-level revenue against the close baseline (R4). Year three: share of revenue held by people hired after close, which is the measure of whether the book has been institutionalised.

Horizon and stage. Deploy; diligence.

Across the five firms. A: relationships are the whole asset; the map is the deal. B: partners hold books but clients are stickier (switching accountants is painful); the map sets rollover. C: lawyers move and clients follow; the map decides the earn-out. D: engineers with stamps are reserved people as well as relationship holders. E: relationships are institutional (hospital CFOs), less personal; the map is shorter.

P4. Client contract and data-rights review

What it is. Every client agreement, engagement letter and master services agreement read for four things: assignment and change-of-control; confidentiality and the right to process client data on a third-party platform; any clause restricting the use of automation or AI in delivery, or requiring disclosure of it; and the rights to the work product and to the data derived from it. In the software edition this lever decided what the revenue levers were allowed to do. Here it decides whether the conversion is permitted at all: a services firm whose engagement letters say the work will be performed by named professionals has a conversion problem before it has a technology one.

Line it moves. Price; representations and warranties; and what Convert is allowed to do (R2, R10, X3 all depend on it).

Typical range and evidence. The reading itself has become cheap, and that evidence carries from the software edition: a legal-AI vendor's named customer reported vendor-contract review falling from two days to two hours, and a document-intelligence vendor cites 20 to 30 hours saved per deal in diligence review.⁷ The content of services contracts is less studied. The strongest public claim is from the thesis's critics, who say services contracts have shortened from around ten years to three or less, which is why there is little lock-in to recoup an AI investment.³ That is an argument, not a dataset, and the firm's own contracts are the only evidence that matters. Grade: Medium (review tooling); the firm's own position is Strong once read.

How to do it in diligence. Every agreement, not a sample. Tag each on the four questions. Produce two lists: contracts that block conversion (named-professional clauses, no-automation clauses, data-use prohibitions) and contracts that need a consent or an amendment at renewal. The second list is P9's calendar.

Conditions. The full contract population in machine-readable form. Counsel in the jurisdictions concerned for the data-processing question, especially where client data crosses borders.

Kill criteria. Contracts covering more than 30 percent of revenue prohibit third-party processing or automated delivery and are not renewable within the conversion period.

Metrics. At close: share of revenue under contracts that permit platform delivery; share needing amendment; share blocking. Year one: amendments achieved at renewal. Year three: share of revenue under contracts that grant data-derivative rights (the R10 base).

Horizon and stage. Deploy; diligence.

Across the five firms. A: engagement letters are often light and permissive; the risk is client data-protection terms. B: engagement letters are standardised and the professional body's rules apply; read those too. C: client outside-counsel guidelines frequently restrict AI use and require disclosure; this is the decisive review for archetype C. D: accreditation-body rules may require named qualified persons. E: business-associate agreements and payer rules govern data; the review is regulatory as much as contractual.

P5. Regulatory and licensing constraints

What it is. The legal ceiling. For each reserved unit in P1, the rule that reserves it: who may perform it, who may sign it, who may own the firm that does it. In US accounting, state rules generally require CPA firms to be majority-owned by licensed CPAs, which is why private-equity investment uses an alternative practice structure, with attest work in a CPA-owned firm and advisory in a separately owned one.⁸ In US law, most states prohibit non-lawyer ownership of law firms under versions of the professional-conduct rules, with limited exceptions; legal process work that is not the practice of law can be owned and delivered differently, and the line between them is drawn by state.⁸ Engineering stamps, accredited laboratory results and notified-body steps have their own rules.

Line it moves. Price, and the ceiling: reserved units cannot move to the platform, reserved signatures cannot be removed from the pyramid, and ownership rules may dictate the deal's legal structure before its commercial one.

Typical range and evidence. Statutes and professional rules, which is the strongest evidence in the book. Grade: Strong, once each rule is cited from its text in the chapter's appendix.

How to do it in diligence. Counsel in each jurisdiction answers three questions per reserved unit: who may perform it, who may sign it, and what the ownership rule for the firm is. Then the deal team answers a fourth: does the alternative structure the rule forces (an APS, a managed-services company alongside a professional firm) still let the platform deliver the codifiable units? Usually yes; the reserved signature sits on top of platform-delivered preparation. But the structure costs money and governance, and it belongs in the model.

Conditions. The P1 tagging is complete, so counsel is answering about units rather than about "the business."

Kill criteria. Reserved share of delivery cost above 30 percent with no route to platform-delivered preparation beneath the signature. An ownership rule that prevents the platform company from controlling the delivery entity's operations.

Metrics. At close: reserved share; required structure; compliance plan. Year one: zero regulatory findings on the conversion. Year three: the same.

Horizon and stage. Deploy; diligence.

Across the five firms. A: low. B: high; the APS is standard and the signature stays human. C: medium; the practice-of-law line is the whole question. D: high in places; stamps and accreditation. E: low to medium; payer rules, not licensure.

P6. Platform-fit assessment

What it is. The gap between what P1 says is codifiable and what the platform does today. For each codifiable unit: does the platform already perform it, perform it with configuration, require a build, or not address it. The build list, costed and sequenced, is the capex reserve in the model and the critical path for R2, the twelve-month service line.

Line it moves. The capex reserve and the sequencing of the whole conversion. A fit assessment that is wrong by a quarter delays the first platform-delivered line past condition three.

Typical range and evidence. No independent data. The pattern is visible in both directions in compliance: an AI compliance platform acquired a forty-year-old broker-dealer compliance consultancy in 2025, and a private-equity-backed cybersecurity assessor acquired an AI compliance platform in 2026, each buying the half it lacked.⁹ Neither has published what the fit was or what it cost to close. Grade: Weak (no published outcomes); Strong as a method if the platform team does it.

How to do it in diligence. The platform team takes P1's unit table and the twenty sampled engagements and runs them, or attempts to, on the platform as it stands. Not a demo: the firm's actual inputs, with the firm's own reviewers judging the output. The result is a four-way split of codifiable units (ready, configure, build, no) and an error rate on the ready units against the firm's own work, which is D3's first baseline.

Conditions. The platform team's time in diligence, which deal teams resist paying for and which is the cheapest money in the deal. Client data for the sample, which P4 must permit.

Kill criteria. "Build" share above half of codifiable cost, or a build list whose critical path exceeds nine months, because R2 then cannot land inside twelve.

Metrics. At close: ready, configure, build and no shares; build cost and schedule; error rate on ready units. Year one: build list delivered against schedule; R2 live. Year three: "no" share re-assessed.

Horizon and stage. Deploy; diligence.

Across the five firms. A: fit is usually high where the platform was built for the domain; the assessment is confirmation plus a gap list. B: tax and bookkeeping platforms exist; the question is which the firm converts onto. C: contract and matter platforms exist; the gap is the firm's templates and precedents (D4). D: fit is the bid decision; much of the work is physical. E: payer-specific rules make configuration the bulk of the work.

P7. Conversion underwriting

What it is. The approved levers in the model, each with its baseline from P1 to P6, its range, its kill date and its owner; the foundation cost from P8; what is in the base case and what is upside; and the structure that follows (rollover, earn-out, seller note) tied to the scorecard's metrics rather than to billings.

Line it moves. Price and structure.

Typical range and evidence. The adoption evidence carries from the software edition: a majority of PE firms surveyed use AI in diligence and valuation, while 39 percent of general partners expected no material portfolio benefit from AI in 2026.¹⁰ In this category specifically, there is no outcome data to underwrite against: a market map that graded roughly twenty AI roll-up vehicles on public evidence found every EBITDA or margin figure company-reported, investor-reported or projected.¹¹ Grade: Strong (adoption) / Weak (outcomes). The underwriting rule that follows is the software edition's: the base case contains only levers with Strong or Medium evidence and a firm-specific baseline; everything else is upside, priced at zero.

How to do it. Build the model as the register will later be run: one line per lever, baseline, target, date, owner. The base case for a services conversion is P9 and R1 (the pricing conversion, from the firm's own realization data), D1 on the "ready" units only (from P6's error rate and hours), D5 and D6 (from the firm's own data), and the O-group levers that carry Medium or better evidence. R2 is in the base case only if P6's build list lands inside nine months. R6, R9 and R10 are upside. The multiple at exit is the entry multiple unless the scorecard thresholds are met, and the re-rating is a separate line that the committee sees as a separate bet.

Earn-out design. A conventional professional-services earn-out pays on revenue or client retention over two to four years; a worked structure presented to accounting-firm partners in 2025 used 50 percent cash at close, 50 percent deferred over five years and 40 percent rollover.¹² That structure protects the buyer against the book walking, which is condition five, and it does nothing for conditions two and three. A seller paid on billings has no reason to convert pricing; a seller paid on hours delivered has every reason to resist the platform. The conversion earn-out pays on the scorecard: recurring share, on-platform share, gross margin, and book retention together, so the partner who keeps clients while refusing to convert them earns less than the one who does both. Rollover equity does the rest: the re-rating, if it comes, is where a rolled partner's money is.

Conditions. The scorecard metrics are instrumented from day one (P8), or an earn-out on them is unenforceable.

Kill criteria. Base-case EBITDA at the entry multiple does not clear the fund's hurdle without the re-rating. That is the committee's own rule and it is the one most often broken in this category.

Metrics. At close: the register with every line populated. Year one and three: lever outcomes against the register, reported to the board in the same format.

Horizon and stage. Deploy; diligence.

P8. The hundred-day foundation

What it is. Everything the Convert levers need and none of them can build for themselves: client files and engagement history migrated onto the platform; the model gateway with logging, access control and the firm's own confidentiality rules enforced; the legal basis for processing each client's data confirmed against P4; baselines for every scorecard metric from the firm's own data as of the close date; the initiative register with kill dates; and the error-rate sampling method agreed with the firm's quality lead before any unit goes live.

Line it moves. One-time operating expense. It enables every lever after it and is the reason the conversion either has baselines or has anecdotes.

Typical range and evidence. The mechanism carries from the software edition, where the foundation chapter rests on security standards and regulators' guidance rather than on vendor claims. In a services firm the content differs: the asset being migrated is client files and the firm's own templates, not a codebase. Grade: Strong (mechanism).

Conditions. Decided before close, funded in the model, led by someone whose only job it is.

Kill criteria. Baselines not instrumented by day 100: nothing after this can be measured, and the earn-out in P7 is void in practice.

Metrics. Day 100: client files on-platform; scorecard baselines published; register live; sampling method agreed. Year one: the register still in use at every board meeting.

Horizon and stage. Deploy; close to day 100.

P9. Pricing conversion at close

What it is. Condition two, executed. In the first renewal cycle after close, every client on hourly or time-and-materials terms is offered a fixed fee or a subscription for the same scope, priced from P2's answer to "what would they have paid for the outcome," while the work is still delivered by people. The automation comes afterwards and lands in margin.

Line it moves. Revenue, and the recurring share on the scorecard. More importantly it moves where the saving from every D-group lever lands: in the firm's margin, or in the client's.

Typical range and evidence. Sector benchmarks say the conversion is achievable because a whole sector has done it: fixed fees are the dominant model in small-firm US accounting and hourly billing has fallen under 4 percent.² The largest firms are moving the same way at the top of the market; the US head of one of the Big Four told the Financial Times in March 2026 that it will sell AI tax and consulting tools by subscription without an hourly-billing consultant.¹³ What no public source measures is the retention cost of the conversion at a given firm. Grade: Medium (sector); the firm's own conversion is measured in P2 and R1.

How to do it. P4's calendar gives the renewal dates. P2 gives the price. The offer is made by the relationship holder from P3, not by the new owner, and it is framed as predictability for the client. Clients who insist on hourly terms keep them at a rate above the fixed-fee equivalent, which nudges without forcing. The first cycle converts what renews; R1 carries the rest over the following year.

Conditions. P2 complete, so the fixed price is set from the firm's own realization data and not from a hope. P3 complete, so the person making the offer is the one the client trusts. Delivery still manual, so the client sees no change in who does the work while the price basis changes.

Kill criteria. Renewal price per engagement falling faster than cost per engagement at the first cycle: the conversion has handed the saving to the client, and the D-group must pause until pricing is fixed. Client-level churn at the first cycle above the trailing three-year average.

Metrics. Day 100: share of renewing revenue converted. Year one: recurring share; price per unit against the hourly-equivalent baseline. Year three: hourly share of revenue (target: immaterial).

Horizon and stage. Deploy; close to day 100.

Across the five firms. A: the whole game; retainers exist but are priced on hours in disguise. B: largely done; P9 is tiering. C: the hardest, because the hourly culture is in the partners as much as the contracts (O7). D: project fees convert to permit-and-renewal subscriptions where the work recurs. E: already outcome-priced; P9 does not apply.

What goes in the model

The software edition's rule was that the model contains only what diligence measured. For a services firm the measured inputs are these, and a model that uses a benchmark where one of them should be is a model of a different firm.

Two lines deserve a sentence each. The re-rating is a separate bet and is modelled as one: the committee should see the return at the entry multiple and the return with the re-rating, and should be able to say which it is underwriting. And the error rate from P6 is the only number in the model that can make a lever negative: a platform that is wrong more often than the people it replaces does not save cost, it moves cost into X1.