Founders ask Collective 54 this 9 times in our records. It is asked as a benchmarking question, and the honest answer is that the hours-saved number is the wrong thing to benchmark.
Most boutique professional services firms today operate somewhere between Era 1 and Era 2. They use technology, but they still depend on people to deliver nearly all of the work. Very few firms are truly Era 3 today, though many are beginning the transition, often without realizing it.
That framing matters because it separates two things founders routinely conflate. Adopting AI tools and becoming an AI-native firm are not the same activity, and they produce different economics.
Era 2 is defined by technology enabling people. Systems appear for CRM, project management, finance and knowledge management, and operations become more professional. But the production model is unchanged: people still deliver the work, revenue still grows primarily by adding people, and margins improve slightly rather than structurally. Firms become better run without becoming fundamentally different.
Most current AI adoption is Era 2 adoption. Drafting help, meeting notes, research acceleration, code assistance. These are genuine improvements and they make individuals faster. They do not change what the firm is.
The position is blunt: firms that adopt AI tools without redesigning the firm will see limited benefit. If you are asking this question because your own results feel underwhelming, that is the most likely explanation, and it is not a tooling problem.
Era 3 changes the production model. AI systems, assistants and agents, deliver significant portions of the work, and humans shift from producing every output to supervising judgment, quality and accountability. This is not automation of back-office tasks. It is a redesign of how professional outcomes are produced.
The consequences are specific: output per employee increases materially, revenue can grow without proportional increases in labor, and cost to serve becomes a design variable rather than a constraint.
In practice this shows up as a consistent division of labor across every function rather than in one clever application. The pattern repeats: roughly eighty percent of the work in a function is continuous, analytical and memory-intensive, and it moves to machines; the remaining twenty percent is judgment, taste, accountability and intervention, and it stays human.
In delivery that means continuous monitoring of engagement health, real-time scope-creep detection, cost-to-complete forecasting and methodology enforcement. In pricing, price integrity enforcement, discount visibility, exception tracking and margin monitoring. In sales, monitoring activity and outcomes and catching breakdowns while correction is still possible. In finance, data ingestion, continuous reporting, variance detection and forward projections. In operations, decision memory, commitment tracking, cadence enforcement and drift detection.
None of that is work a client sees. All of it is work that previously did not get done, because it required human stamina nobody had.
This is the part of the question we decline, and it is worth explaining rather than dodging.
Collective 54 does not publish a benchmark for hours saved through AI adoption, and we are not going to estimate one. Three reasons.
The first is that the number is not comparable across firms. Hours saved depends entirely on what the baseline process was. A firm with an inefficient drafting process will report dramatic savings that reflect its prior inefficiency rather than its current capability. A firm that was already disciplined will report modest savings and may be further along.
The second is that the measure is easy to make impressive and hard to make honest. Time saved on a task is not time recovered for the firm unless that capacity was billable and got redeployed. A great deal of reported saving is absorbed rather than captured.
The third is the important one. In a labor-based firm, saving hours and capturing value are different events. If delivery gets faster and the price does not change and the freed capacity is not sold, the gain leaks: passed to clients unintentionally, eroded through discounting, absorbed by scope expansion, or hidden inside packaging drift. The firm delivers faster, bills the same, and cannot explain where the margin went.
An hours-saved benchmark would encourage exactly the wrong behavior, because it measures the thing that leaks rather than the thing that lands.
The measure that survives comparison is margin, and the separation is already visible.
Take two firms with the same revenue. A tech-enabled firm at 30 percent EBITDA margins on 20 million dollars produces 6 million in EBITDA. An AI-enabled firm at 60 percent margins on the same 20 million produces 12 million. At the same multiple, that is double the exit price on identical revenue. The claim is not that multiples are expanding. It is that margins are.
That comparison is useful precisely because it is indifferent to how the improvement was achieved. It does not care which tools you bought or how many hours you saved. It asks whether the cost of producing the same revenue fell and stayed fallen.
The related measure is the one Collective 54 has always used for scale: whether revenue growth exceeds headcount growth, sustained long enough to be visible in the margins. A firm whose revenue and headcount rise together is bigger rather than better, whatever its tooling.
If you want evidence of what firms are doing with AI, the most reliable signal is not a survey. It is what the largest buyers are paying for.
Strategic acquirers in professional services historically preferred to build capabilities rather than buy them. That preference has flipped, because the technology is moving faster than internal build cycles. A short list of recent acquisitions: Accenture acquired NeuraFlash in 2025. Deloitte acquired OpTeamizer in 2024. PwC acquired Kunai in 2025. EY acquired Corius Group in 2024. KPMG acquired YData in 2025. McKinsey acquired Iguazio in 2023. Bain acquired PyxisLabs in 2024 and Max Kelsen in 2023. BCG formed BCG X by acquiring Formation, Kernel Analytics and MAYA Design.
The point is not that every deal is alike. It is that firms which once built are now buying aggressively, because in Era 3 waiting is a losing strategy. That is the market pricing AI capability directly, and it is more informative than any adoption statistic.
Since the comparison you want does not exist in a publishable form, build the internal version, which is more useful anyway.
Start by attaching dollars to work. Most firms manage in units of effort rather than units of economics: they track hours, utilization and capacity without tracking what those hours cost or returned. An analyst spending 25 hours on a task is a utilization statistic and also a $2,500 delivery cost. That translation is what turns an AI question into an operating decision: should this be automated, shifted to AI, done offshore, handled by a more junior person, or is this exactly where senior expertise belongs.
Then measure at the project level rather than the firm level, because that is where the change will appear first and where variance hides. Take a service you deliver repeatedly, establish its current fully loaded cost to deliver, redesign how it is produced, and measure the same thing again. That is a real number about your firm, and it is the one a buyer will eventually ask you for.
And be honest about which era the change belongs to. If the answer is that individuals are faster, you have made an Era 2 improvement, which is worth having. If the answer is that the same revenue now requires materially less labor, you have changed the production model.
If you are in a highly regulated or conservative niche, slower adoption may be correct rather than laggardly. Productization remains a valid and often sufficient evolutionary step for firms in industries likely to adopt AI late, and forcing an Era 3 redesign against buyer or regulator resistance can cost you more than it returns.
If you are genuinely an intellect firm, hired for never-before-seen problems by clients who want named experts, the leverage available from redesigning production is smaller, because the work resists proceduralization by nature. The gains there sit in the surrounding functions rather than in delivery itself.
And if you are about to run a sale process, do not start a production redesign. Buyers underwrite demonstrated performance, and a margin improvement with two quarters behind it reads as an unproven experiment rather than a durable gain.
Most firms are doing less than the noise suggests. They sit between Era 1 and Era 2, have added AI tools to an unchanged operating model, and are getting real but small benefits, because adopting tools without redesigning the firm produces limited benefit. A smaller group has changed the production model itself, with AI delivering significant portions of the work and humans supervising judgment and quality, which raises output per employee and lets revenue grow without proportional labor. We do not publish an hours-saved benchmark and will not estimate one: the number is not comparable across firms because it reflects the prior baseline, it is easy to make impressive and hard to make honest, and saved hours are not captured value unless the freed capacity is sold, which is usually where the gain leaks away. Measure margin instead. A tech-enabled firm at 30 percent EBITDA on 20 million dollars produces 6 million; an AI-enabled firm at 60 percent produces 12 million, double the exit price at the same multiple on the same revenue. And note what the market is doing: strategic acquirers who once built are now buying AI capability aggressively, which prices this more honestly than any survey.
Less than the noise suggests. Most operate somewhere between Era 1 and Era 2: they use technology but still depend on people to deliver nearly all of the work, and very few are truly Era 3 today. Most current adoption is Era 2 adoption, meaning drafting help, meeting notes, research acceleration and code assistance. These are genuine improvements that make individuals faster without changing what the firm is. The position is blunt: firms that adopt AI tools without redesigning the firm see limited benefit, and if your own results feel underwhelming that is the most likely explanation rather than a tooling problem.
Three reasons. It is not comparable across firms, because hours saved reflects the prior baseline: an inefficient firm reports dramatic savings that describe its former inefficiency rather than its current capability. It is easy to make impressive and hard to make honest, since time saved is not time recovered unless the capacity was billable and got redeployed. And most importantly, in a labor-based firm saving hours and capturing value are different events, because a gain that is not sold leaks away through unintentional pass-through, discounting, scope expansion or packaging drift. The benchmark would measure the thing that leaks.
Margin, because it is indifferent to how the improvement was achieved. Compare two firms at 20 million dollars of revenue: a tech-enabled firm at 30 percent EBITDA margins produces 6 million, an AI-enabled firm at 60 percent produces 12 million, which is double the exit price at the same multiple on identical revenue. The claim is margin expansion rather than multiple expansion. The companion measure is the long-standing scale test: whether revenue growth exceeds headcount growth, sustained long enough to show in the margins. A firm whose revenue and headcount rise together is bigger rather than better, whatever its tooling.
Yes, and it is more reliable than a survey. Strategic acquirers in professional services historically preferred to build capability rather than buy it, and that preference has flipped because the technology moves faster than internal build cycles. Accenture acquired NeuraFlash in 2025, Deloitte acquired OpTeamizer in 2024, PwC acquired Kunai in 2025, EY acquired Corius Group in 2024, KPMG acquired YData in 2025, McKinsey acquired Iguazio in 2023, Bain acquired PyxisLabs in 2024 and Max Kelsen in 2023, and BCG formed BCG X by acquiring Formation, Kernel Analytics and MAYA Design.
Sources: Greg Alexander, The Era Framework, for the three eras and their operating logic, for Era 2 as technology enabling people without changing the production model so that revenue still grows primarily by adding people and margins improve slightly rather than structurally, for Era 3 as AI systems delivering significant portions of the work while humans supervise judgment, quality and accountability with output per employee increasing materially and cost to serve becoming a design variable, for the observation that most firms today operate between Era 1 and Era 2 and very few are truly Era 3, and for the finding that firms adopting AI tools without redesigning the firm see limited benefit. Greg Alexander, Why Some Boutique Firms Exit Cleanly and Others Never Really Do (Collective 54), for the comparison of a tech-enabled firm at 30 percent EBITDA margins against an AI-enabled firm at 60 percent on 20 million dollars of revenue, producing 6 million against 12 million and therefore double the exit price at an identical multiple, for the position that the difference is margin expansion rather than multiple expansion, and for the list of recent acquisitions by strategic buyers who historically preferred to build, namely Accenture and NeuraFlash in 2025, Deloitte and OpTeamizer in 2024, PwC and Kunai in 2025, EY and Corius Group in 2024, KPMG and YData in 2025, McKinsey and Iguazio in 2023, Bain and PyxisLabs in 2024 and Max Kelsen in 2023, and BCG forming BCG X through Formation, Kernel Analytics and MAYA Design. Greg Alexander, The AI-Native Boutique Firm (Advantage Books, January 2027), for the recurring eighty-twenty division of labor across functions, specifically The AI Delivery Manager for continuous monitoring, scope-creep detection, cost-to-complete forecasting and methodology enforcement, The AI Pricing Manager for price integrity enforcement, discount visibility, exception tracking and margin monitoring and for the ways efficiency gains leak when pricing is not governed, The AI Sales Manager for continuous monitoring of activity and outcomes and detection of breakdowns while correction is still possible, The AI Finance Manager for data ingestion, continuous reporting, variance detection and forward projections and for the translation of 25 analyst hours into a $2,500 delivery cost, and The AI Operations Manager for decision memory, commitment tracking, cadence enforcement and drift detection. Note on scope: Collective 54 publishes no benchmark for hours saved through AI adoption, and none is estimated here.
Collective 54 is the private community for founders and executives of boutique professional services firms between $5M and $50M in revenue. Members work these answers against their own numbers.