Running it: the register, the cadence, the objection, and what goes wrong

The register

One line per lever, from the day the model is built (P7) to exit. Each line carries: the lever ID; the P&L line it moves; the baseline, from the firm's own data as of close, with the date; the target at year one and year three; the owner, who is a function lead and never a transformation office; the kill criterion and the kill date; the horizon and stage; the current value; and the status (not started, live, at statistical sampling, killed, closed). The register is the board pack. It replaces the utilization report in a services firm on day 100 (D8), and it is the thing a buyer reads at exit before the financial model, because it shows whether the numbers were earned or asserted.

A register where nothing is ever killed is a wish list. The first lever killed on schedule is the proof the system is real.

The initiative card

Every lever on the register has a card, and every card has the same fields, carried from the software edition and extended with two the conversion needs:

• What it is, in a sentence that states the mechanism.
• The line it moves.
• The baseline, with its source and date.
• The typical range, with its evidence grade and the two or three anchor figures behind it, and, where the evidence is the firm's own, a statement that no external range is used.
• The conditions that must hold for it to work.
• The kill criteria, each as a number and a date.
• The metrics at close, year one and year three.
• The horizon and the stage at which it opens.
• New: the units it touches (from P1), so that a delivery lever's scope is a list and not a percentage.
• New: the client-facing change, if any, and the renewal cycle in which the client sees it, so that no client is surprised by a conversion they were not sold.

The cadence

Weekly. The kill review. Every live lever's current value against its baseline and its kill criterion. Levers at their kill date are killed or explicitly extended by the chief executive with a written reason. D1's on-platform share and D3's error rate are published together every week; one without the other is a number nobody should trust.

Monthly. The scorecard to the board: the ten lines, against the close baseline, with the register summary (live, killed, closed). Named-person retention is reported by name. Client-level book retention is reported by client above one percent of revenue.

Quarterly. The exit scorecard (X6), from year two: audited metrics, organic growth separated from acquired, the claims log, the data-consent share, and the implied value under the three buyer cases. The committee sees which case the firm is in.

At each renewal cycle. The pricing conversion report: revenue converted, price per unit against hourly-equivalent, churn at conversion, clients retaining hourly terms.

At each add-on. The integration scorecard (B2): days to first platform unit, days to first pricing conversion, integration cost, and the pacing rule's status (B5).

The objection, worked through

The strongest case against this playbook was made in June 2025 by Nathan Benaich and Nikola Mrkšić, who called the AI roll-up thesis "a fundamental category error": confusing operational improvement with business-model transformation.¹ The hub states their four reasons. This section works each through the register, because a sponsor who raises the objection deserves a lever-level answer rather than a slogan.

Reason one: the acquirer rents labor on the client's terms. The client controls the stack, the process and the approvals; the services firm does what it is told, with whatever efficiency it can find, and the client captures the gain. The answer on the register: P4 reads every contract for whether the firm may deliver through its own platform and on its own process; X3 builds the governance that lets a client say yes; R2 puts the firm's platform in front of the client as the thing being bought. The metric is the share of revenue under contracts permitting platform delivery (P4) and the share delivered through platform lines (R2's year-three metric). If those are low at year three, reason one holds and the firm is renting labor with better tools.

Reason two: hourly billing means efficiency cuts revenue, so firms revert to maximizing hours. The answer: condition two, P9 and R1, executed before any automation; D8, which removes hours from the management system; O7, which removes hours from the compensation system. The metric is recurring share by contract and price per unit against hourly-equivalent. If recurring share is not rising while on-platform share is, the saving is going to the client and reason two holds.

Reason three: contracts have shortened to three years or less, leaving no lock-in to recoup the investment. The answer: R8, converting episodic engagements to continuous monitoring subscriptions where switching means losing history; R3, scope expansion that makes the platform hold more of the client's compliance, tax or billing estate; and the scorecard's retention metrics. The metric is renewal rate on monitoring subscriptions against episodic clients (R8) and NRR. If a converted client churns like a project client, reason three holds.

Reason four: inefficiency is the product; clients pay for flexibility, customization and someone to hold accountable. The answer: the playbook agrees, and prices it. R11 makes the named senior, the response time and the accountability a tier above the platform tier rather than a cost bundled into hours. D3 gives the client a documented answer to "how do you know the platform is right." The metric is premium-tier attach. If clients will not pay for the human tier, reason four holds in an unexpected direction: they valued the person less than the firm believed, and the firm is more of a software business than its partners thought.

The verdict the register gives. Four reasons, four sets of metrics, all on the scorecard. A conversion that answers all four has a software-like case to make to software buyers, at software multiples as they are now, which the case study shows are a few turns above a services multiple with a premium rather than a different asset class. A conversion that answers two has made a services firm materially more profitable and more durable and should be sold as one. The authors' prediction, that AI roll-ups will behave as "tech-enabled private equity" rather than scale like software, is the second outcome, and the playbook treats it as a respectable result rather than a failure. The failure is underwriting the first at entry and discovering the second at exit, which P7 prevents by modelling both from the start.

What goes wrong

Drawn from the five firms and from the evidence base's one consistent feature.

The saving goes to the client. Automation before pricing conversion. The kill signal is P9's: price per engagement falling faster than cost per engagement. The fix is to pause the D-group until pricing is fixed, which is painful because the D-group is where the visible progress is.

The book walks. A relationship holder leaves and clients follow, or partners disengage because the compensation formula changed without explanation or the role they were offered had no price on it. The kill signal is named-person retention below 90 percent in any quarter. The fix is in P3, O7, R11 and X5, and it is mostly done before close or not at all.

The error rate is not measured, so it is not managed. The firm never measured rework (D6), so D3 has no baseline, so the first client claim is the first data point. The fix is D6 before D1, every time; the regulator's finding that independent accuracy evidence is scarce is the reason the firm has to produce its own.²

The platform runs a generic method. D4 was skipped because it looked like documentation. The platform delivers something that is correct by its own standards and wrong by the firm's, and the client notices. The fix is senior time paid for explicitly (O7), before the first unit moves.

The hiring freeze outruns the automation. D2 froze production hiring; D1 moved fewer units than planned; capacity ran short; the firm rehired at a premium or missed deadlines. The fix is to tie the freeze to D1's actual on-platform share, unit type by unit type.

The second deal starts before the first is finished. The one public return study of buy-and-build by add-on count says this is where the return goes.³ The fix is B5's pacing rule, written before the second deal.

The numbers are the vendor's. Every figure in the board pack traces to a platform vendor's case study or an investor's projection, because the firm's own baselines were never instrumented (P8). A buyer at exit asks for the firm's own series and there is none. The fix is P8 by day 100 and the register thereafter; there is no later fix.

The re-rating is underwritten at entry. The committee priced the deal on the software exit and the firm delivered the services exit. The fix is P7's rule: the re-rating is a separate line and a separate bet, visible from the first model.

The one thing the evidence is consistent about

Across the vehicles, the vendors and the investors in this category, the people reporting the results are the people selling them, and no one has published audited margins.⁴ That is not an accusation; it is the stage the category is at. It means the firm's own register is the only evidence that will exist for this firm, and that a board that runs the register honestly, kills what fails and records what it killed, will own something at exit that nobody else in the category has: a measured conversion. That is the point of running it.