The Gap Between Work and Headcount
Every operator knows the shape of this problem even if they have never named it. A product manager resigns in the middle of a roadmap. A migration that touches three teams belongs to none of them and has been ninety percent finished for a quarter. The board asks for a report that requires two weeks of somebody's undivided attention, and there is no such person. In each case the work is real, urgent, and bounded - and the instrument the company reaches for is a permanent employment contract, which is none of those things.
The mismatch is not a failure of management. It is a structural artefact of how companies are allowed to buy capability. Headcount is approved annually, defended politically, and granted on the assumption that the need is permanent. Work does not respect any of those constraints. So the org either waits - carrying the cost of the gap silently - or gives the work to its best people, who then stop doing the thing they were actually hired for.
What follows is not a case for never hiring. It is a case for matching the instrument to the shape of the need, and for measuring whether you got it right.
Why Now - and Why Most of the Data Is Bad
Three things changed at once, and only one of them is about consultants.
The senior independent talent pool got deep enough to be reliable. This is the part with real evidence behind it. MBO Partners has run its State of Independence study for fifteen consecutive years, and the 2025 edition puts roughly 72.9 million Americans working independently, with a record 5.6 million independents earning more than $100,000 a year - up about 19 percent in a single year. That last number is the one that matters. It is not a measure of gig work; it is a measure of how much genuinely senior capability now sits outside permanent employment. Ten years ago, hiring a fractional operator meant accepting whoever was between jobs. It no longer does.
Tooling collapsed the output floor for one person. The same study found 74 percent of independents using generative AI, reporting savings of roughly nine hours a week. Whatever you make of self-reported time savings, the direction is not in dispute: the volume of work a single experienced operator can carry has risen sharply, which changes the arithmetic of whether that operator needs a team behind them.
Hiring got slower and more expensive at exactly the wrong moment. SHRM's 2025 benchmarking data puts the US average time to fill at 44 days across all roles, with cost per hire averaging $5,475 for non-executive roles and $35,879 for executive ones - money spent before anyone has produced anything.
Fractional work is a category with a large content-marketing industry attached to it, and its numbers should be read with that in mind. Published 2025 estimates of the global fractional executive market range from roughly $5.7 billion to $9.4 billion - for the same year. A spread that wide is not a measurement, it is a guess with a decimal point. I have seen the claim that unfilled positions drain American businesses of "$1.08 trillion every month" repeated across vendor blogs; that would be well over half of US GDP annually, which should have been the end of the sentence. Treat market-size and cost-of-vacancy figures in this space as directional. The hiring-cost and workforce numbers above come from SHRM and MBO Partners, which are methodologically published and worth more.
Not a Recruiter, Not a Freelancer, Not an Agency
The three models get bundled together because they all produce a person who is not on your payroll. Commercially they could not be more different, and the distinction is easiest to see by asking one question: what event triggers the fee?
A recruiter is paid when someone starts. A freelancer is paid when a defined task is delivered. An agency is paid for hours, marked up to cover a management layer you did not ask for. An embedded operator is paid when a problem is closed. Those four trigger points produce four completely different sets of incentives, and every downstream difference follows from them.
| Recruitment agency | Freelancer | Consulting agency | Embedded operator | |
|---|---|---|---|---|
| You are buying | A candidate | One skill | A team plus its overhead | Judgment across several disciplines |
| Fee triggered by | The person starting | The task shipping | Hours elapsed | The problem closing |
| Typical cost shape | 15-30% of first-year salary, then the salary itself | Day or project rate, per task | Marked-up hourly, ongoing | Per job, no retainer |
| Who does the work | Someone you have not met yet | The person you hired | Often more junior than who pitched | The person you spoke to |
| Context between jobs | Rebuilt from zero by the new hire | Re-briefed every time | Re-scoped every engagement | Retained - the relationship is the asset |
| When the work crosses a boundary | You hire again | You hire a second freelancer | A change order | Same person continues |
| Incentive if the problem persists | Neutral - fee already earned | More tasks, more revenue | More hours, more revenue | Negative - closing it earns the next job |
| If it does not work out | Severance and a second search | The contract lapses | The term runs out | The engagement simply ends |
The bottom row is the one worth sitting with. Every model in that table except the last is financially neutral or positively disposed toward the problem lasting longer. That is not a moral claim about the people involved - it is arithmetic. If you want an adviser whose interests point the same direction as yours, you have to look at how they get paid, not at what they say in the first meeting.
The Freelancer Ceiling: One Skill Versus the Whole Job
The comparison people most often get wrong is with freelancing, because superficially the arrangement looks identical: an experienced person, no payroll, paid per engagement. The difference is not seniority or price. It is breadth, and it shows up at exactly the moment a real problem stops being tidy.
Consider a migration that has been almost finished for three months. A freelance developer will do the engineering competently and stop at the edge of the codebase - which is precisely where the actual problem lives, because the reason the migration is stuck is usually that three teams each own a piece and none owns the outcome. Solving it requires writing code, running a cross-team delivery plan, renegotiating what "done" means with a product owner, and telling an executive something they will not enjoy. That is four disciplines. Under a freelance model it is four contracts, four briefings, and four people with no shared view of the problem - assuming you can find them, and assuming the seams between them do not become the new bottleneck.
This is the honest structural claim for the embedded model, and it is checkable rather than aspirational: most expensive organizational problems do not sit inside a discipline, they sit in the seams between disciplines. A model that buys one discipline at a time cannot, by construction, own a problem that spans several. A person who has been a developer, then a product and program manager, then done the organizational work, can - and, importantly, can tell you which of the four the real constraint actually is before anyone starts building.
Ask a prospective operator what happens when the problem turns out to be different from the brief. A freelancer's honest answer is "that is outside my scope." An agency's is "we will scope a second phase." The answer you want is a description of what they would do next - because a problem that changes shape mid-engagement is the normal case, not the exception.
The Money, With the Arithmetic Showing
The case for on-demand talent is usually made by comparing a day rate to a salary, which is the wrong comparison and flatters the wrong side. A salary is the smallest component of what a permanent hire costs. Here is the full stack, with sources, before anyone has done a day of work.
Put those together for a single senior hire at a EUR 100,000 salary and the pre-productivity commitment looks like this. The formulas are shown deliberately, because the inputs are yours and the output should change when they do:
| Component | Formula | Illustration |
|---|---|---|
| Recruiter fee | salary × 20% | EUR 100,000 × 0.20 = EUR 20,000 |
| Cost per hire | SHRM benchmark | roughly EUR 5,000, non-executive |
| Manager time in the loop | interview hrs × loaded hourly | 30 hrs × EUR 90 = EUR 2,700 |
| Vacancy carry | days open × daily contribution | 44 days, valued from your own revenue per head |
| Ramp subsidy | salary × months × (1 - productivity) | EUR 8,333/mo × 6 × 0.5 = EUR 25,000 |
| Committed before first output | sum of the above | EUR 50,000+ plus the vacancy carry |
Two honest caveats. The ramp subsidy assumes an average of 50 percent productivity over six months, which is a modelling convention rather than a measurement - your own figure may be materially better or worse. And none of this argues that permanent hires are a bad investment; for continuous work they are the cheapest instrument available, which is exactly why the comparison has to be made against the shape of the need rather than in the abstract.
There is one further cost that never appears on an invoice and is usually the largest: the work you absorb internally. When a gap goes to your own best people, you pay a senior engineer's salary for coordination work, and you lose the engineering you hired them to do. That loss is invisible precisely because no one issues a bill for it, which is why it runs for years.
The 8 KPIs That Hold the Model Accountable
Any engagement model that cannot be measured eventually becomes a habit. These are the eight I would hold an on-demand relationship to. Track them in a shared document from day one, not retrospectively when someone questions the spend - and note that the first one exists specifically to tell you when to stop.
01 Repeat-Task Rate The one that matters most
engagements addressing the same problem ÷ total engagements, rolling 12 months
How often you are calling the same person about the same thing. This is the single honest test of whether the model is working, because a genuinely closed problem does not recur. It is also the metric a vendor has every incentive not to give you, which is why you should ask for it in writing at the start.
02 Hiring Cost Avoided
recruiter fee + cost per hire + (interview hours × loaded hourly rate)
The most straightforward saving and the easiest to defend to a CFO, because every component has a benchmark behind it. Count it only where you genuinely would have hired - counting hypothetical hires you were never going to make is how these models get discredited.
03 Cost of Vacancy Avoided
(days to fill − days to start on demand) × daily contribution of the role
The gap between 44 days and this week, valued at what the role actually contributes. Derive the daily contribution from your own revenue per employee rather than from a published benchmark - the vendor-published cost-of-vacancy figures in this space are the least reliable numbers in the whole category.
04 Ramp Cost Avoided
months to productivity × monthly salary × (1 − average productivity)
What you do not pay for the period a new hire is learning. The interesting version of this metric is the second engagement onward, where the operator's own ramp is zero because the context was retained. Track ramp days per engagement over time: if it is not trending to zero, the relationship is not compounding and you are paying relationship prices for transactional service.
05 Internal Senior Hours Released
hours returned to your own people × their loaded hourly rate
The cost that never appears on an invoice. When a senior engineer stops coordinating a migration and goes back to engineering, that is real recovered capacity, and it is usually the largest number in the whole exercise. Measure it by asking the person, before and after, what fraction of their week the work was consuming.
06 Tool and License Recovery
(seats paid − seats actually used) × unit cost, before and after
Gartner's research puts roughly 25 to 30 percent of SaaS spend on unused or underused licenses, against enterprise SaaS spend above $300 billion in 2025. A material share of "we need a new tool" turns out to be "we own a tool nobody was shown how to use." This KPI captures both the licenses eliminated and the purchases avoided.
07 Handover Half-Life
months the team runs the process unaided after the engagement ends
The honesty metric, and the hardest one to game. If the written handover works, the team keeps running the thing without help. If the process quietly reverts within a quarter, the engagement produced output but not capability - and you should say so at the next review rather than re-buying the same work.
08 Blended Effective Day Rate
total paid over 12 months ÷ days actually worked, vs. FTE loaded cost for the same days
The comparison that keeps everyone honest in both directions. A senior day rate looks expensive next to a daily salary and cheap next to a fully loaded employment cost at 1.25 to 1.4 times salary - and the on-demand side is only genuinely cheaper if you are buying a fraction of a year. Run this annually. If you are buying most of a year, the answer has changed.
When This Is the Wrong Model
The strongest argument for an engagement model is a clear account of where it fails, so here it is. Two of these are hard constraints rather than preferences.
On demand fits when
- The work is bounded and has an end state you can describe
- The capability is needed now and the headcount is not coming
- The problem crosses functions and nobody owns the seam
- You need a specific person's judgment, not a team's capacity
- The need is real but genuinely temporary - a leave, a launch, a migration
- You want the capability transferred rather than retained by the vendor
Hire permanently when
- The same problem has recurred more than twice in a year
- The work is continuous, not project-shaped
- The role requires formal authority over people
- Regulation requires a named, accountable employee in the seat
- The knowledge is a competitive asset that must live in-house
- On-demand days are approaching a full-time equivalent year
The recurring-problem case deserves emphasis because it is the one everybody rationalises. A problem that keeps coming back is a headcount problem wearing a project costume, and continuing to buy it in project form is more expensive than hiring - it just spreads the cost across enough invoices that nobody adds them up. That is exactly what KPI 01 is for.
Sources
Graded by strength, because in this category that matters more than usual.
- StrongSHRM, 2025 Recruiting Benchmarking Report. US average time to fill of 44 days; average cost per hire of $5,475 for non-executive roles and $35,879 for executive roles. Published methodology, membership-scale sample. shrm.org
- StrongMBO Partners by Beeline, 2025 State of Independence in America (15th annual). 72.9 million Americans working independently; 5.6 million independents earning above $100,000, up roughly 19 percent year over year; 74 percent using generative AI. The longest-running study of the independent workforce. mbopartners.com
- StrongGartner, SaaS spend and license utilisation research. Roughly 25 to 30 percent of SaaS spend estimated as wasted on unused or underused licenses, against enterprise SaaS spend above $300 billion in 2025. gartner.com
- EstimateRecruitment agency fee structures. 15-30 percent of first-year salary, with 20-25 percent typical in technology roles. Widely reported industry standard rather than a single published study; verify against your own quotes.
- EstimateLoaded employment cost multiplier. Commonly modelled at 1.25 to 1.4 times base salary once employer taxes, benefits, equipment and space are included. A planning convention, not a measured figure - your finance team's number is better.
- EstimateTime to full productivity. Commonly cited at 6 to 12 months for senior and executive roles across HR industry sources. Directional; there is no authoritative cross-industry measurement.
- EstimateFractional executive market sizing. Published 2025 estimates range from roughly $5.7 billion to $9.4 billion with growth rates of 11 to 14 percent, mostly from vendors and content aggregators. The spread is wide enough that these figures are indicative at best, and are not relied on for any argument in this piece.
This article is general commentary on engagement models and cost structures. It is not financial, legal, tax, or employment-law advice, and none of the figures above are a prediction of results for any particular company. Employment classification rules for independent contractors differ by jurisdiction and are worth checking with a licensed professional before structuring any engagement.
May Mor
Talent on demand for growing tech companies. 10 years inside fintech, digital banking, adtech and capital markets - full-stack developer, then product and program management, then the organizational work. M.Sc in AI. I take the work you have nobody for, finish it, and leave your team able to run it without me. Full bio →