Six months of a part-time technical lead is $15,000 at our published rate. A one-off architecture review is $3,000. A full-time hire covers 37% of the same six months, because the vacancy stays open for most of it. And below about $60,000 of decisions in those six months, none of the three pays.
Six months is the horizon because it is the shortest one on which an advisory fee and a build price are the same size: our lead runs from $2,500 a month, our data platform band starts at $12,000, and 12,000 ÷ 2,500 = 4.8 months of retainer buys the cheapest platform we would build. Every figure below is a published price of ours, a line quoted from a document with the date I read it, or a division you can redo, and the chosen numbers say so.
What a part-time technical lead costs against a hire, a review or doing nothing
Over six months at our published prices it’s $15,000 for the fractional technical lead, $3,000 for a one-off architecture review, $0 for doing nothing, and $44,863 for a full-time hire who starts when the search ends. The first three are arithmetic on a price list: our advisory card reads “architecture review from $3,000. Fractional technical lead from $2,500/month”, and $2,500 × 6 = $15,000. The fourth takes the next two sections, because the salary is the small half of it.
Fees over the same six months in dollars, at our published prices and the chosen salary derived below; coverage is the share of those months with somebody in the room.
| Option | Fee over six months | Clock starts | Share of the horizon covered |
|---|---|---|---|
| Nothing | $0 | at once | 0% |
| Architecture review | $3,000 | at once | one look, no duration published |
| Fractional technical lead | $15,000 | at once | six months, part-time |
| Full-time hire | $44,863 | after 2.8 months | 37% |
None of the four is silly. A full-time lead is the only one who still knows your system in year three; a review costs a fifth of the lead’s fee; and a decision you can reverse for less than $3,000 is cheaper to get wrong than to get advice about. And the fractional lead is ours, which is why most of what follows is about when not to buy it.
A fifth line sits on our data engineering page and I left it out of the table deliberately: the Data Platform Review, two weeks at a fixed price, whose fee comes off the build if you build with us. It’s the only price we publish that can actually finish at zero.
What does a full-time engineering hire cost beyond the salary?
About 40% of the salary again, on four components I chose rather than looked up. I’ll put the base at $120,000 so the divisions stay legible. It is not a market rate or a claim about pay anywhere, and every figure below scales linearly in it.
Employer contributions and insurance are 12 points of base; equipment and software seats are 8; the recruiter’s fee is 15, paid once; and management and onboarding time is 5, which has a ladder under it. A 40-hour week is 40 × 52 = 2,080 paid hours, less 20 days of leave at 8 hours (160), 10 public holidays (80) and 5 sick days (40): 2,080 − 160 − 80 − 40 = 1,800 hours actually worked. Two hours a week of somebody senior is 2 × 52 = 104 hours, and 104 ÷ 1,800 = 5.78%, which I rounded down to 5 because it prices the manager’s hour at the hire’s, and nobody’s manager is that cheap.
So 12 + 8 + 15 + 5 = 40, the multiplier is 1.40, and $120,000 × 1.40 = $168,000 a year, or $168,000 ÷ 12 = $14,000 a month.
The obvious version of this number is wrong. Ask around and you get two of the four: contributions and kit, 12 + 8 = 20 points, a multiplier of 1.20. The recruiter’s fee and the manager’s hours are the other 20 points — half the loading, and neither one a tax or a benefit. They are the price of the search and the price of the ramp, which is what this article turns out to be about. I used to carry 1.3.
How long an engineering lead vacancy stays open, derived from openings and hires
Between 1.3 and 2.8 months, off two published tables. The Job Openings and Labor Turnover Survey technical note, read on 29 October 2026, counts the two sides on different clocks: “Job openings include all positions that are open on the last business day of the reference month”, countable only where “The job could start within 30 days”, while hires are “all additions to the payroll during the entire reference month”.
So one is a stock and the other a flow, and mean time in a queue is the queue length over the rate it drains. Table 1 of the August 2026 release, seasonally adjusted and preliminary, gives 123 thousand openings in Information and 1,186 thousand in Professional and business services; Table 2 gives 44 thousand hires and 935 thousand, both tables read on 29 October 2026. Information is 123 ÷ 44 = 2.7955 months and Professional and business services is 1,186 ÷ 935 = 1.2684. Both figures stay in play, because Information is NAICS 51 and engineering services sit in Professional and business services, so a lead could be hired into either.
Dividing openings by hires is an old proxy and not my invention, and the objection to it is as well worn as the method: a stock over a flow is only a duration if the queue is in a steady state, and I have no reason to think this one is. Two more caveats, and neither is fatal. Both levels are industry means over every occupation and grade in the industry, and neither table splits by seniority, so a technical lead’s own vacancy might well sit above them. And the published rates give 4.3 ÷ 1.6 = 2.6875 months, which is a rounding check on the same two series rather than a second opinion.
Treat 2.8 months as an order of magnitude. Two errors run opposite ways and I can’t net them: hires never posted as openings mean fewer than 44 thousand drained the stock, which pushes the duration up, while openings withdrawn rather than filled push it down. I don’t know which dominates.
Check how much of a six-month window a full-time hire covers
A full-time technical lead covers 37% of a six-month decision window. Take the Information duration of 2.7955 months and add a ramp of one month. The ramp is chosen, not measured — put your own in, anywhere from zero to two. Coverage is (6 − 2.7955 − 1) ÷ 6 = 2.2045 ÷ 6 = 36.74%, and I’ll round that down rather than up.
You pay from the start date rather than from the day you decide, so the bill is 6 − 2.7955 = 3.2045 months at $14,000 = $44,863 for 2.2045 months of leadership, or $44,863 ÷ 2.2045 = $20,351 a month of it, against the fractional lead’s $15,000 ÷ 6 = $2,500. The 20,351 ÷ 2,500 = 8.1 between them is not a price comparison, and I’d argue with anyone who read it as one: a full-time month buys full-time hours and a retainer month buys part of one. It’s the calendar.
Choosing a full-time hire for a six-month decision window is choosing to do nothing for the first 2.7955 ÷ 6 = 46.6% of it, whether or not that is what you thought you were choosing.
What the decision being wrong costs, and why it is not the severance
One week’s notice is 2.41% of what a wrong hire costs over twelve months, and the rest is calendar. Section 86(1) of the Employment Rights Act 1996, read on 29 October 2026, says the notice owed to anyone with a month or more of continuous employment “is not less than one week’s notice if his period of continuous employment is less than two years”. Paragraph (c) is where almost everybody misreads it: the twelve weeks owed after twelve years is a minimum and not a cap, because all three paragraphs say “not less than”. That statute is probably not yours, and what carries over is the size of the term.
So price the wrong person in calendar. Twelve-month horizon, three months before anybody admits the hire was a mistake — chosen, and the input I would argue with hardest. The sequence runs 2.7955 of search, 1 of ramp, 3 you later undo and 2.7955 of a second search: 2.7955 + 1 + 3 + 2.7955 = 9.591 months, so leadership kept is 12 − 9.591 = 2.409 months, or 20.1% of the year. You paid for the ramp and the three months, 1 + 3 = 4 at $14,000 = $56,000, which is $56,000 ÷ 2.409 = $23,246 for every month of leadership you kept. The hire who works out pays 12 − 2.7955 = 9.2045 months at $14,000 = $128,863 and keeps 8.2045 of them: $128,863 ÷ 8.2045 = $15,706. Being wrong costs 23,246 ÷ 15,706 = 1.48 times as much per month kept. And the statutory week, at 52 ÷ 12 = 4.3333 weeks a month, is 1 ÷ 4.3333 = 0.2308 months, or 0.2308 ÷ 9.591 = 2.41% of the time you lost.
When is a one-off architecture review better value than a fractional lead?
Between $60,000 and $240,000 of six-month decision exposure, and above $240,000 the fractional lead takes over. Decision exposure is money you will commit over the horizon on the strength of technical judgement: build fees, embedded engineering, anything you cannot unwind without paying twice. A small exposure is a top-of-band data platform at $20,000 plus a top-of-band RAG build at $20,000, so $40,000. A large one adds a web platform at $50,000 and two embedded engineers for six months at 2 × $4,000 × 6 = $48,000: 40 + 50 + 48 = $138,000.
Nobody can tell you how much rework a reviewer prevents, myself included, and no measured figure goes on this page because we have not measured one. Invert the question instead. An option earns its fee when the reduction in rework it buys, times your exposure, beats the fee — so fee ÷ exposure is the reduction it has to deliver.
The table gives that reduction, as a percentage of decision exposure, at three exposures built from the bands above.
| Option | Fee | E = $40,000 | E = $138,000 | E = $276,000 |
|---|---|---|---|---|
| Architecture review | $3,000 | 7.5% | 2.17% | 1.09% |
| Fractional lead, six months | $15,000 | 37.5% | 10.87% | 5.43% |
One relationship in it never moves: 15,000 ÷ 3,000 = 5.0 at every exposure, because the exposure cancels. Six months of a fractional lead has to be five times better than one review, whatever you have at risk.
So name the tolerances before the numbers, because a threshold without one is an opinion with a decimal point. I will credit six months of senior attention on a system somebody is actively building with ten points of rework probability, and no more. The review gets half of that, five points, because it reads the system once and sees none of the decisions that come after. The lead therefore breaks even at $15,000 ÷ 0.10 = $150,000 of exposure and the review at $3,000 ÷ 0.05 = $60,000, and the two cross where 0.10E − 15,000 = 0.05E − 3,000, which is 0.05E = $12,000 and E = $240,000, or 12 top-of-band platform builds inside one six-month window. At $150,000 the lead nets $0 and the review nets 0.05 × 150,000 − 3,000 = $4,500, so our dearer product stops losing money there rather than becoming the right buy.
The floor deserves a second derivation, from a different published fact. Every build card on our pages bills “30% up front, 40% at an agreed midpoint, 30% on delivery”, so 30 + 40 = 70% is committed by the midpoint, and counting only that money puts the floor at $3,000 ÷ (0.05 × 0.70) = $85,714 rather than $60,000. I’ll quote the lower one, because exposure includes the rework you pay for after delivery as well as the money you can’t claw back. Every band here is on our data engineering page.
Below $60,000 of decisions in six months, hire nobody
Nothing clears its own gross fee down there, and that includes both of our advisory lines. The exception nets to zero and it is the last paragraph here. A buyer with one platform decision and one build carries the $40,000 above, where the review needs 7.5 points against the five I claim for it and the lead needs 37.5. I wouldn’t sign for that. So we don’t take the money.
Our own page makes the cheaper instrument hard to price, which is the next thing I would fix. The price table carries a row reading “Data Platform Review, two weeks, fixed scope” at $4,000, and the platform block higher up the same page says “the $6,000 doesn’t change” about the same two weeks. Read that block against the price table and the gap is in front of you inside a minute. At $40,000 of exposure the two fees need 4,000 ÷ 40,000 = 10.0% and 6,000 ÷ 40,000 = 15.0%, so our own disagreement moves the break-even by half again.
The advisory line is worse, because it carries no denominator at all. The first engagement card says “from $4,500/month for a senior data engineer, 160 hours — about $28 an hour. Minimum 20 hours.” The third says “architecture review from $3,000. Fractional technical lead from $2,500/month.” So bracket the lead with our two other published embedded rates, $5,000 ÷ 160 = $31.25 and $4,000 ÷ 160 = $25.00: 2,500 ÷ 31.25 = 80.0 hours a month at the dearer and 2,500 ÷ 25.00 = 100.0 at the cheaper, a 1.25 spread on a quantity we never print. The page doesn’t say.
But the real alternative below the floor isn’t a cheaper adviser. Cut the build instead. Dropping from our published web platform band to our published MVP band takes $50,000 of scope to $20,000, which removes $30,000 of exposure for a fee of $0 and drops the large case to $108,000, a 30,000 ÷ 138,000 = 21.7% cut. It is not free, and the cost is in the same unit: $30,000 of scope you decided not to buy. Descoping beats the fractional lead when the dropped features are worth less than the fee you save plus the rework you avoid, which at a chosen baseline rework probability p is $15,000 + p × $30,000: $24,000 at p = 30% and $30,000 at p = 50%.
One instrument survives below the floor, and it’s the one I left out of the first table. The Data Platform Review’s fee comes off the build, so if you were going to build with us anyway its net fee is $0, and it beats doing nothing at any exposure, $40,000 included. That is a quirk of our pricing rather than a property of reviews — the same inversion token prices fell 73% and the same answer costs 2.7 times more runs on inference bills.
What flips the answer: a longer horizon, or double the exposure
The horizon flips it. The exposure doesn’t. Tolerance first: a search eating more than a tenth of the horizon it covers is buying the wrong thing. The distribution is the 1.2684 to 2.7955 months derived above, I design to the worse end, and 2.7955 ÷ 0.10 = 27.955 months. Call it 28, the horizon line, above which a full-time hire is right and below which it isn’t. At twelve months that hire covers (12 − 2.7955 − 1) ÷ 12 = 68.37% against 37% at six, so the calendar argues against hiring on a short horizon and clears it on a long one.
A year of 4,800 Microsoft salaries put a neighbouring line at eighteen months off the same August 2026 release. That article spends a quarter of an engagement on ramp and chose 4.5 months for it; this one spends a tenth of a horizon on search and derives 2.7955. Put my search time into its tolerance and the line is 2.7955 ÷ 0.25 = 11.2 months; put its 4.5 months into mine and the line is 45. The tolerance is doing more work than the data in both. Eighteen and 28 bracket an answer rather than agreeing on one.
BLS Employee Tenure for January 2026, read on 29 October 2026, puts median tenure for “management, professional, and related occupations” at 4.9 years, so 4.9 × 12 = 58.8 months, or 58.8 ÷ 27.955 = 2.10 times the line. That is median time already served by people still in post rather than the expected length of a hire you are about to make, so it is a sanity check with the wrong denominator in it.
Derate the exposure 4x to $138,000 ÷ 4 = $34,500 and the lead needs 15,000 ÷ 34,500 = 43.5% while the review needs 3,000 ÷ 34,500 = 8.7% against the five points I claim, so the decision hardens toward nothing rather than flipping. Double it to $276,000 and the lead clears at 5.43% and sits above the $240,000 crossover, which is the one case on this page where I would sell it. The input the answer really turns on is my own tolerance: at five points instead of ten the horizon line is 2.7955 ÷ 0.05 = 55.9 months, that 58.8-month median only just clears it at 58.8 ÷ 55.9 = 1.05 against 2.10, and the margin on a full-time hire all but disappears.
We don’t sell a fractional technical lead below $60,000 of six-month exposure, and between $60,000 and $240,000 I’d point you at the $3,000 review instead of at six months of us. Below the floor, I’d rather tell you to cut the build.