How do I measure if my CV improved?

Fix the outcome first, or you are measuring your mood

Most people measure a CV against getting hired. That is the wrong outcome. Between the CV and the contract sit the interview, the second interview, the reference calls and a decision between two near-equal candidates — stages where your document plays no part at all. A CV decides exactly one transition: from the pile to the first human reaction.

So count first human replies. An invitation, a phone call, a question about your background, a rejection written by a person who clearly read your file. An automated acknowledgement does not count. Neither does a template rejection. Draw that line once and never move it, or you will quietly shift the yardstick along with the document.

You also need a deadline. Without one, every application stays open forever and your rate keeps creeping up long after you have moved on. Four weeks from sending is a workable cut: after that the application counts as no reply, even if something arrives later.

The arithmetic almost nobody runs

Assume your true invitation rate is ten per cent. That is an illustrative assumption, not a market figure — nobody publishes a reliable Swiss invitation rate, and yours is yours alone. You send ten applications, rewrite the CV, send ten more. If the rewrite changed nothing whatsoever, the new version comes out ahead 34.4 per cent of the time, behind 34.4 per cent of the time, and level 31.3 per cent of the time.

Put plainly: with a completely unchanged CV, the ten-against-ten comparison is a coin toss with a thick edge. Anyone who concludes "the new version works" from that comparison has consulted a coin with equal justification.

It gets worse when the rewrite did damage. If the true rate falls from twelve per cent to eight, the worse version still wins the ten-against-ten comparison 54.1 per cent of the time. Even at thirty applications per version, the worse CV comes out ahead 38.2 per cent of the time. Small samples do not merely fail to confirm optimism — they cheerfully confirm a regression.

And if you want an actual demonstration: doubling from ten to twenty per cent needs 199 applications per version at conventional statistical standards. Going from ten to fifteen per cent needs 686. Those numbers are not an argument against rewriting your CV. They are an argument against judging the rewrite on the replies to twenty applications.

Zero replies from ten applications is not a finding

The most common misreading is the zero. Ten applications, not one reply, therefore something is wrong with the CV. That conclusion arrives far too early. At a true invitation rate of ten per cent, 34.9 per cent of all ten-application runs end at zero. A third of people with a perfectly ordinary document experience exactly that.

The zero is not a signal, then. It is an instruction to keep counting. Only at around twenty applications with no reply at all does it become a hint worth acting on — and even then the cause may sit in which jobs you chose as easily as in the document.

The second round is never the same exam

Even if you push the number of applications high enough, one problem does not go away by counting: the adverts you answered before are not the adverts you answered after. And they differ far more than it feels like they do.

We computed this on our own running analysis of open Swiss adverts. Draw ten nursing adverts at random and, in nine draws out of ten, somewhere between three and eight of them require German at working level; the median is five. Draw two batches of ten from the same field and they differ by 1.7 adverts on average, and by three or more in a good quarter of comparisons. The picture repeats in IT, commercial and administrative roles, retail, hospitality, construction, finance and engineering — the spread is practically identical everywhere.

Language level is simply the dimension we can measure most cleanly. The language the advert is written in scatters just as much: among ten randomly drawn finance adverts, between three and eight are not written in German. So two rounds of applications can differ, on a requirement that decides invitation or rejection, by half the batch — with no involvement from you at all.

If you want to compare, hold constant what you can: the same occupational field, the same region, the same application channel. And accept that a residue of difference remains that you cannot control.

Ten applications are not ten independent verdicts

The arithmetic above assumes each application is an independent trial. In practice it often is not. A small number of employers and staffing agencies post a large share of the adverts, so several of your applications land on the same desk.

We counted how often that happens. Draw ten adverts from a single field and the same employer appears at least twice in 73.3 per cent of nursing draws, 71.7 per cent of hospitality draws, 90.1 per cent of construction draws, 95.8 per cent of IT draws and 98.4 per cent of retail draws. In retail and IT, roughly four draws in five contain the same sender three times over.

For your measurement that means one person who likes or dislikes your CV can generate three data points. It makes the numbers noisier than the clean arithmetic suggests, and it is one more reason not to trust a small difference.

When silence turns into a data point

To count anything you need to know when a missing answer counts as an answer. The lifetime of the adverts gives you a reference. In our analysis, a quarter of open positions have been online for ten days or less, half for around twenty days, and a quarter for 66 days or longer. Just over six in ten open adverts are a month old or younger.

That spread is why silence means nothing for so long. A position that has been open for two months often has a selection process running just as long. And the lifetime depends heavily on the field — construction adverts stay up markedly longer than technical and engineering ones.

arbeit.swiss recommends following up in exactly this situation, noting that after some time without a response you can ask, factually and politely, about the state of the process. For measurement purposes that is useful: it converts an open application into a countable outcome instead of leaving it suspended indefinitely.

Four things you can measure the same day

Because the outcome is so slow, you measure leading indicators instead — properties of the document that can be checked immediately and that do not depend on which adverts happened to be open. They do not prove you will get more invitations. They prove that a known source of failure has been removed.

First, the machine read-back. Have your CV parsed and compare what comes back with what is on the page: name, every employer, every date range, every qualification. Count the fields that come back correct, before and after. That is a hard number, it changes in minutes, and it routinely drops with two-column templates or details placed in page headers.

Second, the language levels. Of the open adverts in our analysis, 72.9 per cent carry a stated language requirement; where one exists, 71.9 per cent name exactly one language and 21.2 per cent name two. German at working level is required by 46.4 per cent of all open adverts, English by 25.8 per cent, French by twelve per cent. The demand is therefore almost always explicit — which makes it checkable whether your CV answers it explicitly, with a stated level rather than bars or dots.

Third, the official pre-send checklist. arbeit.swiss names three things to check before sending: "spelling and formatting errors", "completeness of documents" and "clear and correct file names", with the documents saved and sent as PDF files from a reliable email address. That is five yes-or-no questions. A version that passes all five is better in a demonstrable sense than one that does not.

Fourth, the blind pass. Give both versions to somebody who does not know your history, allow thirty seconds per document, and then ask them to write down what you did most recently, for how long, and whether you qualify for the job. Two versions, one reader, the same clock. It is the only test on this list that measures an effect on a human being, and it takes a minute.

The log: four fields per application

The slow number still decides the question. You simply cannot have it in a fortnight. What you can have is a record that will be analysable in six months. arbeit.swiss advises keeping one anyway: "Keep a record of applications sent, ongoing application procedures and outstanding responses."

Four fields are enough: the date, the field and region of the job, the channel (advertised role, speculative application, personal contact), and which version of your CV you sent. The outcome gets added later. With those four you can compare at the end without relying on memory — and you will see immediately when the two rounds were never comparable in the first place.

Two rules separate a log from a measurement. Change one thing at a time: rewrite the CV and start applying to different roles in the same week, and afterwards nobody can say which one did the work. And fix the deadline before you analyse rather than after, or you will unconsciously pick the deadline that produces the answer you wanted.

What a checker score measures, and what it does not

A checking tool, ours included, measures conformance to checkable rules: whether the advert's stated requirements appear in the document, whether the structure is machine-readable, whether details are missing that Swiss adverts routinely ask for. It does not measure an employer's interest, and it cannot — that interest only comes into existence once a person reads your file.

The value of such a score lies precisely in the fact that it moves immediately. It is the proxy you can work on while the real number accumulates slowly. Take it for what it is: evidence that a known source of failure was removed, not evidence that you will be invited.

Take that distinction seriously and you stop passing judgement after twenty applications, and start checking the document against the individual advert instead. That is the level at which an application is actually decided.