Founders ask Collective 54 this 8 times in our records, 4 of them in 2026. The word true is the giveaway: most firms measure effort and call it cost.
The single most useful move here is not a spreadsheet technique. It is deciding what you are measuring.
The unit of measure of profit for a healthy boutique firm is the project, and the financial performance of the firm is the sum of its projects. Cost follows the same rule. A firm-level cost number tells you what you spent. A project-level cost number tells you what a service costs to deliver, which is the only version that can inform a price, a staffing plan or a decision to retire an offering.
Capital 54 uses exactly this unit in diligence. On one commercial photography firm the measure was cash flow per project, built from four inputs: the fee, hours per staff member, fully loaded cost per staff member, and allocated overhead. Those four inputs are the answer to this question, and they are worth taking one at a time.
You need to know who touched the work and for how long, and you need it broken down far enough to be useful.
Project-level hours tell you whether an engagement was profitable. Task-level hours tell you why, and they are what let you act. The practical route is a breakdown of recent representative engagements to the task level, which is the same exercise that underpins certification and service engineering, so most firms are doing it once and using it three times.
One caution. Time that is not recorded is not free, it is hidden. Late time entry is one of the quiet ways profit disappears, and a cost model built on incomplete timesheets will systematically understate the cost of exactly the engagements that are hurting you, because those are the ones where people stopped recording.
The second input is what an hour of each person actually costs, and salary divided by hours is not it.
Fully loaded means the total cost of employing someone spread across the hours they are available to work. Once you have that, every hour carries a burdened cost and every task has an economic footprint. An analyst spending 25 hours on a task stops being a utilization statistic and becomes a 2,500 dollar delivery cost.
That single translation, from hours to dollars, is what turns a costing exercise into an operating one. Once the cost is visible the decisions follow. Should this task be automated? Shifted to AI? Done offshore? Handled by a more junior role? Or is this exactly where senior expertise belongs? None of those questions can be asked usefully until the number exists.
The denominator matters too. The standard assumption is a forty-hour week across forty-eight weeks, which is 1,920 hours per employee. Utilization sits on top of that: at 75 percent, three quarters of those hours are available to charge to work, and the cost of the other quarter still has to land somewhere.
In Era 3 there is a third component that did not exist in the old model, and leaving it out is now the most common error in a cost build.
Cost to serve has to be forecast across human, AI and tooling components. If part of your delivery is performed by AI, that portion has a real cost, it is not zero, and it does not behave like labor. It does not scale linearly with volume, and it can change materially between one quarter and the next as models, pricing and usage shift.
This creates a specific trap. A service that looks dramatically cheaper to deliver because AI absorbed part of the work has not necessarily improved its margin, because the gain leaks unless someone captures it. Efficiency that is not governed gets passed to clients unintentionally, eroded through discounting, absorbed by scope expansion or hidden inside packaging drift. A cost model that shows the drop without anyone watching what happens next simply documents where the money went.
The fourth input is the share of non-billable cost the engagement has to carry.
Overhead is everything that does not directly contribute to selling or delivering the work, which in our structure means operations, marketing, finance, IT, legal and HR. Sales sits in its own block after that, and what survives both is EBITDA.
For a costing exercise the question is how much of that load each project absorbs. Allocate it consistently rather than precisely, because a defensible consistent method beats an elegant one nobody repeats. What matters is that the same rule is applied across engagement types, so that comparisons between them are real.
The published benchmarks give you a sanity check on the total. A firm running above 75 percent gross margin and 40 percent EBITDA is carrying roughly 35 points of revenue across overhead and sales combined. If your allocation implies far more than that, the cost model is not wrong, the cost structure is.
A cost figure that only informs a price is underused. Once activity-based costing exists it informs how services are designed, how work is priced, how deals are scoped, how teams are staffed, how capacity is planned and how cash is managed. Waste that was previously invisible becomes obvious, and tradeoffs that were emotional become economic and defensible.
Two uses are worth naming specifically.
The first is variance. Take the same service delivered five times and compare the cost each time. Wide variation means the delivery model is not standardized, which is the finding that disqualified the photography firm in diligence, because a model that is not standardized is not scalable.
The second is modeling before commitment rather than discovery after. Expected and acceptable margins should be modeled ahead of a deal, with cost to serve forecast and sensitivity tested against discounting and scope creep. A cost model used only in hindsight is a report. Used before signature it is a constraint.
If every engagement is genuinely one of a kind, a precise standard cost is not achievable and chasing it wastes effort. Build a range instead, with an honest worst case, and price to the range rather than to the midpoint.
If you are very small, a full activity-based costing build is premature. Start with the four inputs on your three largest engagement types and stop there. The precision you gain from going further will not change a decision at that size.
And if your timesheets are unreliable, fix that before building anything on top of them. A sophisticated model sitting on bad hours produces confident wrong answers, which is worse than no answer, because people act on it.
Measure it per project rather than per firm, because the project is the unit of profit in a boutique firm and the financial performance of the firm is the sum of its projects. Build the number from four inputs, the same four Capital 54 uses in diligence: the fee, hours per staff member, fully loaded cost per staff member, and allocated overhead. Take hours to task level rather than project level, since task level is what tells you why rather than whether, and treat unrecorded time as hidden rather than free. Use fully loaded cost rather than salary, so every hour carries a burdened cost and 25 analyst hours reads as a 2,500 dollar delivery cost, which is the translation that turns costing into an operating decision about whether to automate, shift to AI, move offshore, push to a more junior role, or keep senior. Add the line most firms omit, which is AI and tooling, because cost to serve in Era 3 spans human, AI and tooling components and the AI portion is neither zero nor linear. Allocate overhead consistently rather than precisely, and sanity-check the total against the roughly 35 points of revenue implied by the 75 percent gross margin and 40 percent EBITDA benchmarks. Then use it before signature rather than after, and watch the variance across repeat deliveries, because wide variation means the delivery model is not standardized and therefore not scalable.
Four, and they are the same four Capital 54 uses in diligence on a firm: the fee, hours per staff member, fully loaded cost per staff member, and allocated overhead. The unit matters as much as the inputs, because the project is the unit of profit in a boutique firm and the financial performance of the firm is the sum of its projects. A firm-level cost number tells you what you spent. A project-level number tells you what a service costs to deliver, which is the only version that can inform a price, a staffing plan or a decision to retire an offering.
Because it converts effort into economics. Most boutique firms manage in units of effort rather than units of economics, tracking hours, utilization and capacity without tracking what those hours cost or returned. Once every hour carries a burdened cost, an analyst spending 25 hours on a task stops being a utilization statistic and becomes a 2,500 dollar delivery cost. That single translation is what makes the operating decisions askable: whether the task should be automated, shifted to AI, done offshore, handled by a more junior role, or is exactly where senior expertise belongs.
The AI and tooling line. Cost to serve in Era 3 has to be forecast across human, AI and tooling components, and the AI portion is neither zero nor linear, since it does not scale with volume the way labor does and can move materially between quarters. There is a trap attached: a service that looks much cheaper because AI absorbed part of the work has not necessarily improved its margin, because ungoverned efficiency leaks, passed to clients unintentionally, eroded through discounting, absorbed by scope expansion or hidden in packaging drift.
More than pricing. Activity-based costing informs how services are designed, priced, scoped and staffed, how capacity is planned and how cash is managed, making previously invisible waste obvious and turning emotional tradeoffs into economic ones. Two uses matter most. Variance across repeat deliveries of the same service, because wide variation means the delivery model is not standardized and therefore not scalable, which is the finding that disqualified one firm in diligence. And modeling expected and acceptable margins before commitment rather than discovering them afterwards, which turns a report into a constraint.
Sources: Greg Alexander, The AI-Native Boutique Firm (Advantage Books, January 2027), specifically The AI Finance Manager for the finding that boutique firms manage in units of effort rather than units of economics, tracking hours, utilization and capacity without tracking what those hours cost or returned, for the explanation that this was a technical rather than philosophical limit because attaching dollars to every unit of work was too slow and error-prone to be practical, for the translation of 25 analyst hours into a 2,500 dollar delivery cost and the sequence of operating questions that follows about automating, shifting to AI, offshoring, using a more junior role or keeping senior expertise, and for activity-based costing informing how services are designed, how work is priced, how deals are scoped, how teams are staffed, how capacity is planned and how cash is managed, making invisible waste obvious and turning emotional tradeoffs into economic ones; and The AI Pricing Manager for forecasting cost to serve across human, AI and tooling components, for modeling expected and acceptable margins before commitment and testing sensitivity to discounting and scope creep, and for the finding that ungoverned efficiency gains leak through unintentional pass-through, discounting, scope expansion and packaging drift. Greg Alexander, The Boutique: How to Start, Scale, and Sell a Professional Services Firm (Advantage, 2020), chapter 12 for the Capital 54 diligence in which cash flow per project was built from fee, hours per staff member, fully loaded cost per staff member and allocated overhead, and for the project-level volatility that revealed a delivery model which was not standardized and therefore not scalable; chapter 16 for the project as the unit of profit measurement and the position that the financial performance of a firm is the sum of its projects, and for building task-level breakdowns of representative engagements through postmortems; chapter 14 for the standard assumption of a forty-hour week across forty-eight weeks producing 1,920 hours per employee and for utilization sitting on top of that; chapter 30 for the benchmark set of more than 75 percent gross margins and 40 percent EBITDA margins. Collective 54, The AI-Native Firm Map front matter, for overhead covering operations, marketing, finance, IT, legal and HR, with sales in its own block and EBITDA as what survives both. Note on scope: the roughly 35 points of revenue used here as a sanity check on total overhead and sales load is arithmetic derived from the published gross margin and EBITDA benchmarks rather than a published figure.
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.