Andrew RadosevichWhy the headcount substitution math almost never works at this size
The intuition is simple: if AI saves 20 percent of time on a workflow, the firm should be able to do the same work with 20 percent fewer people. At $5M to $15M firms the math breaks for three reasons. First, the saved time is rarely concentrated in one role; it is distributed across associates, paralegals, project managers, and partners, none of whom individually drop below a full FTE worth of work. Second, the saved time funds quality improvements that were previously skipped under deadline pressure, not idle capacity. Third, the institutional knowledge cost of removing a mid-career professional almost always exceeds the labor savings on the workflow they ran. BCG's 2025 GenAI in Professional Services work documented this pattern explicitly: firms capturing value did so through expanded output, not reduced headcount [1].
What capacity reallocation actually looks like
A $9M consulting firm installs AI on proposal drafting and meeting note synthesis. Sprint 1 baseline: senior associates spend 6 to 8 hours per week on proposals and 4 to 5 hours per week on meeting note follow-ups. Sprint 2 install: AI drafts proposals from the discovery transcript and generates structured meeting notes with action items pre-assigned. Sprint 3 result: 7 to 9 hours per week recovered per senior associate. The reallocation question is not whether to keep the associates, but where to put the recovered capacity. Three options dominate: more client work served per associate (revenue lift), new service line build (revenue expansion), or quality and oversight improvements that reduce realization losses (margin lift). Most firms pick a combination of two.
The McKinsey scaling problem and why it bites SMBs hardest
McKinsey's 2025 State of AI survey found that while AI adoption is broad, value capture concentrates in firms that scale beyond pilots into permanent operating workflows [2]. Smaller professional services firms hit this wall harder because the gap between pilot success and operational scale is usually one person's time. The firm that runs three AI pilots successfully and then fails to install any of them at scale typically does so because no internal owner was named in advance to inherit operation. The capacity-reallocation framing solves this directly: the owner is the role whose capacity is being reallocated, and they have a structural incentive to make the install hold.
The four workflows that scale best at this revenue band
First, document drafting (engagement letters, proposals, memos, reports), where the AI produces a first draft and the senior reviews. Second, intake and discovery synthesis, where AI converts client conversations and submitted documents into a structured matter or engagement record. Third, internal knowledge retrieval, where AI surfaces relevant prior work product, precedent, and templates from the firm's document store on demand. Fourth, scheduling and meeting administration, where AI handles transcription, summarization, action item assignment, and follow-up drafting. All four reallocate capacity from junior to senior work, which is the direction the P&L benefits from in a knowledge-work business.
How to communicate the install to the team
The communication frame matters as much as the install itself. The St. Louis Fed productivity research found that sustained productivity gains depend on operator confidence in the tool [3]. Operators who believe the install is the prelude to layoff actively underuse the tool and document its failure modes. Operators who believe the install funds growth and reduces drudgery use it aggressively and surface improvements. The owner should communicate the install plan, the reallocation thesis, and the no-layoff commitment in writing, in the same conversation. A verbal reassurance that is not in writing tends to be assumed to be temporary.
The revenue path: turning recovered hours into measurable growth
Recovered capacity is not revenue. It becomes revenue through one of three mechanisms. One: increased throughput at the existing service line (more matters, more engagements, more projects per associate). Two: new service line launch using existing seniors who now have time to staff it. Three: tighter client account expansion, where senior time previously absorbed by drafting is redirected to client relationship and account growth. Most $5M to $15M firms see the first measurable revenue impact at month 6 to 9 after the install, with the largest gains in years two and three as the operating cadence absorbs the recovered capacity into deliberate growth motions.
Common questions
What if our firm genuinely is overstaffed today?
Then the headcount adjustment should happen independent of the AI install, on its own performance and capacity rationale, with its own timeline. Bundling the two creates the worst of both: operators perceive AI as a layoff vehicle, and the firm loses both the headcount and the install. The right sequence is to right-size first if needed, communicate it clearly, then install AI on the stabilized team.
How do we know if the recovered capacity is real or imaginary?
Sprint 1 of the install captures a baseline (hours per week on the target workflow). Sprint 3 measures the same metric. The delta is real. The trap is measuring perception rather than time: if the operator says they feel more productive but the metric did not move, the install did not deliver value and the workflow should be re-scoped.
Does this apply equally to legal, accounting, and consulting?
Yes, with workflow differences. Legal firms get the largest lift on intake and document review. Accounting firms get it on intake classification and document extraction. Consulting firms get it on proposal drafting and meeting synthesis. The capacity-reallocation logic is identical across the three; the specific workflows where the math closes inside 90 days differ by practice area.
Sources
- New GenAI Tools Offer an Edge. Why Aren't More Professional Services Firms Using Them?. BCG. Accessed June 18, 2026.
- The State of AI in 2025: Agents, Innovation, and Transformation. McKinsey & Company. Accessed June 18, 2026.
- The Impact of Generative AI on Work Productivity. Federal Reserve Bank of St. Louis. Accessed June 18, 2026.