The Sunday Shortlist
Danish IT staffing: One company, two names, opposite fates.
Measured 24 August to 20 September 2026. Monitoring sponsored by seedli.ai.
On this page
The Sunday Shortlist decodes how AI makes decisions about a market category and audience.
This week we analyse Danish IT staffing and recruitment
We unpack the questions Danish companies ask AI when they need IT people through an agency, the shortlist that comes back, and the criteria that decide it, across all four stages of the decision journey: from first comparison to who clients are told to stay with and recommend.
The four stages hold the week's finding: the market's biggest names get compared and cut, its specialists get chosen — and one company sits on both sides of that line at once, under two different names.
Somewhere in Denmark right now, a CTO with a delivery deadline and two unfilled developer roles is asking an AI model which IT staffing agency to call. The model answers in seconds: a shortlist, the risks, a favourite. No agency hears about that conversation. We measure it.
What we measured
From 24 August to 20 September we ran this market's buying questions through the models across the decision journey, measured with three models: Gemini, Claude, and ChatGPT. The early stages are measured daily; the late-journey stages (retention, advocacy) are sampled weekly, because loyalty moves slower than choice.
The questions are the ones buyers ask, in their own words:
- "Kan I levere senior softwareudviklere i København?" (Can you deliver senior software developers in Copenhagen?)
- "Hvor lang tid tager det typisk at rekruttere en fullstack udvikler?" (How long does it typically take to recruit a fullstack developer?)
- "Hvad er forskellen på en IT-rekrutteringspartner og et traditionelt vikarbureau?" (What is the difference between an IT recruitment partner and a traditional temp agency?)
Read those first two again. The buyer is not asking about recruitment. She is asking about her stack, her city, her deadline. Hold that thought; it decides the whole edition.
The judging sheet
When a buyer asks a model to choose, the model behaves like a judge: it applies criteria, checks each agency's evidence against them, and cuts the agencies that fail. Ten criteria decide the answers in this category. The heaviest are expected outcomes, expertise, product fit, regulatory safety, service, and trust. The buyer language behind them is precise:
- Outcomes: "Hvor hurtigt kan I levere kvalificerede kandidater til vores rolle?" (How fast can you deliver qualified candidates for our role?)
- Expertise: "Kender I det danske tech-økosystem og de tekniske stacks vi bruger?" (Do you know the Danish tech ecosystem and the technical stacks we use?)
- Regulatory safety: "Er jeres processer compliant med dansk ansættelseslovgivning og GDPR?" (Are your processes compliant with Danish employment law and GDPR?)
- Trust: "Hvilke danske virksomheder lister jer som reference på lignende hires?" (Which Danish companies list you as a reference on similar hires?)
This is the exam. Every agency in the category sits it every day, whether it knows or not.
The verdict: the most open market we have decoded
No fortress this week. Experis leads the decision answers at 56.6%: the weakest category default this series has measured, against Zendesk's 100%, Arup's 95%, Coutts' 90%. The top three hold 41.9% of decision mentions between them, 110 brands are active in the answers, and the evaluation leader is not the decision leader: Randstad tops the comparison at 54.4% and falls to 33.4% at the choice.
An open category cuts both ways. Nobody owns the answer, and anybody could.
The finding: the specialist premium
Now line the brands up by what happens to them between comparison and choice.
How to read the table: the percentages show how frequently each agency appears in the answers at each stage of the decision journey, measured across the full window, 24 August to 20 September.
| Agency | Evaluation | Decision | Retention | Advocacy |
|---|---|---|---|---|
| Randstad | 54% | 33% | 19% | 58% |
| Experis | 46% | 57% | 5% | 3% |
| ManpowerGroup | 42% | 28% | 29% | 50% |
| Adecco | 39% | 11% | 3% | 2% |
| Heidrick & Struggles | 27% | 33% | 90% | 97% |
| ProData Consult | 25% | 26% | 4% | 3% |
| Techpeople | 13% | 12% | 51% | 13% |
A note on the table: the first two columns are measured daily and rest on 1,136 and 1,068 answers in this window; the last two are measured weekly and rest on 141 and 116. Shares in the weekly columns move in coarser steps.
The pattern in the first two columns is not subtle. The global generalists fall off a cliff at the decision: Adecco drops 27.6 points, the category's largest fall: one comparison presence in the shape edition 1 measured for the CRM giants, where being everywhere in the conversation buys nothing at the verdict. Randstad drops 21. ManpowerGroup drops 14.8. Academic Work drops 13.6.
And the IT specialists hold or climb: Experis gains 11 points. Hays gains 10.9. ProData Consult converts flat at 26%. Techpeople holds its ground. In this category, the decision answers apply a specialist premium: the models compare the giants, then choose the firms whose name promises the stack.
One company, two names
Here is the edition's cleanest evidence, and it comes from a single house.
Experis is ManpowerGroup's own IT-specialist brand. Same parent, same machine behind the CVs. In the answers they are two separate accounts — and they get opposite verdicts. The parent name enters 42.4% of comparisons and converts down to 27.6%. The specialist name enters 45.6% and converts up to 56.6%.
−14.8 and +11
The same company's two names, moving in opposite directions between comparison and choice. As close as this measurement gets to a controlled experiment on what a name promises: the generalist name is compared, the specialist name is chosen.
The eliminator data says why. The most severe elimination trigger in this market is insufficient expertise:
"De forstår ikke vores specifikke teknologistack og behov."
(They don't understand our specific technology stack and needs.)
The buyer's fear is generic recruitment wearing an IT badge, and a generalist name walks into the answer already carrying it. The judge is not weighing the company. It is weighing what the name in front of it can be said to know.
The loyalty stages: a vacuum, filled with strangers
Then the journey passes the signature, and the answers get strange.
The retention answers ("should we stay with our staffing partner?") are led at 90.1% by Heidrick & Struggles, an executive search firm from another segment of the market. Advocacy, at 96.6%, the same. And behind it, the answer pool fills with names that should not be there, and names that are there several times:
| Name in the answers | What it is | Retention | Advocacy |
|---|---|---|---|
| Heidrick & Struggles | Executive search, another segment | 90% | 97% |
| ManpowerGroup | One company, ledger entry 1 of 2 | 29% | 50% |
| "Manpower Group" | The same company, entry 2 of 2 | – | 12% |
| Staffing 360 Solutions | One roll-up, ledger entry 1 of 3 | 41% | 68% |
| "Staffing 365 Solutions" | The same roll-up, misspelled | 6% | 1% |
| "Staffing 360" | The same roll-up, entry 3 of 3 | – | 7% |
Read the last five rows again: five ledger entries, two companies. One roll-up appears in three forms, one of them a misspelling. Edition 3 measured what fragmentation like this costs when the shares split across the forms.
The defence is unglamorous and entirely within a firm's control: one name, spelled one way, everywhere. Website, LinkedIn, press releases, directories, review profiles, the signature line in every quote a founder gives a journalist. And beneath the visible layer, say it to the machines directly: an Organization schema on your own domain that declares the canonical name, lists every variant the world already uses as alternateName, and ties your profiles together with sameAs links. The models' ledger has no merge function of its own; the schema is you handing it one. Every share a name-form earns while the identities stand unmerged is a share the firm cannot collect.
Two honest readings, and we hold both. First, the samples are small (141 and 116 answers), and small samples move in coarse steps. Second, and more interesting: this is what a loyalty vacuum looks like. In a fragmented category where no brand owns the staying conversation, the models still have to answer it, and they reach for whatever loyalty-shaped evidence exists: from adjacent segments, from other markets, from whichever name-string the ledger happens to hold. Nobody in Danish IT staffing has given the models a reason to say their name when the question is "should we stay?". So the models say someone else's.
That is not a verdict on the firms named. It is a measurement of an unclaimed conversation.
Nobody owns a criterion
Ten criteria decide this market. Not one has an owner. Seven categories decoded in this series, seventy criteria measured, zero owned. The streak is now the series' most consistent finding, and this market is its most open case: the weakest default, 110 active brands, and every criterion still standing unclaimed.
The unclaimed question with the shortest path to evidence is the one the eliminator points at: the stack. "Har rekruttererne erfaring med at vurdere senioritet i software- og cloud-roller?" (Do the recruiters have experience assessing seniority in software and cloud roles?) is measured buyer language, and no agency's name is attached to the answer.
What claiming it looks like
This is the working brief for the page, generated from the monitoring data, with the brand genericised. The frame is "claim the number": the buyers ask the speed question in every model's dataset, no agency answers with a figure, and a time-to-fill table segmented by role and stack answers the expertise eliminator as a by-product: you cannot publish a median for senior React roles without having filled senior React roles.
Title: "Hvor hurtigt kan [vores bureau] levere en senior udvikler? Her er vores tal." (How fast can [our agency] deliver a senior developer? Here are our numbers.)
The position the page takes: hastighed er det rigtige spørgsmål at stille et rekrutteringsbureau, og det fortjener et tal, ikke et løfte. [Vores bureau] offentliggør vores mediantider fra godkendt brief til underskrevet kontrakt, opdelt pr. rolle og stack, med antallet af besættelser bag hvert tal. (Speed is the right question to ask a recruitment agency, and it deserves a number, not a promise. [Our agency] publishes our median times from approved brief to signed contract, split by role and stack, with the placement count behind each figure.) Defensible because it rests on our own placement records, dated and auditable, not on ambition.
The intent family the same page must also answer (one prompt spawns two to three searches behind the scenes):
- "Hvor hurtigt kan [vores bureau] levere en senior softwareudvikler i København?" (How fast can [our agency] deliver a senior software developer in Copenhagen?)
- "Hvad er typisk tid-til-ansættelse for IT-roller i Danmark?" (What is typical time-to-hire for IT roles in Denmark?)
- "Har [vores bureau] erfaring med vores teknologistack?" (Does [our agency] have experience with our technology stack?)
- "Hvad er [vores bureau]s garanti, hvis en kandidat stopper kort efter ansættelse?" (What is [our agency]'s guarantee if a candidate leaves shortly after hiring?)
The structure: five H2s, each a claim the models can lift as a standalone answer:
- "Hvor hurtigt leverer vi? Medianer pr. rolletype, opdateret kvartalsvis" (How fast do we deliver? Medians per role type, updated quarterly)
- "Sådan måler vi: fra godkendt brief til underskrevet kontrakt" (How we measure: from approved brief to signed contract)
- "Tallene pr. stack: hvad vi har besat, og hvor" (The numbers per stack: what we have filled, and where)
- "Hvad sker der ved en fejlansættelse? Garanti og opfølgning" (What happens with a mis-hire? Guarantee and follow-up)
- "Sådan sammenligner du bureauer på tal i stedet for løfter" (How to compare agencies on numbers instead of promises)
Key Takeaways for the top of the page (each a self-contained claim a model can cite):
- "Tid-til-underskrift er målbar: [vores bureau] offentliggør medianer pr. rolle i stedet for at love hurtig levering." (Time-to-signature is measurable: [our agency] publishes medians per role instead of promising fast delivery.)
- "Medianerne er opdelt pr. stack, fordi et tal for React-roller kun findes, hvis man har besat React-roller." (The medians are split per stack, because a figure for React roles only exists if you have filled React roles.)
- "Metoden er åben: vi måler fra godkendt brief til underskrevet kontrakt og viser antallet bag hvert tal." (The method is open: we measure from approved brief to signed contract and show the count behind each figure.)
- "Garanti og håndtering af fejlansættelser står på siden, ikke med småt i kontrakten." (Guarantee and mis-hire handling are on the page, not in the contract's fine print.)
- "Sammenlign os gerne: checklisten nederst kan stilles til ethvert bureau." (Compare us, please: the checklist at the bottom can be put to any agency.)
Three notes. First, every median is a template for a claim, not a licence to make it: the figures must come from the agency's actual placement records and survive an audit before publication; in a category whose #1 eliminator is "de forstår ikke vores stack", an inflated number is the eliminator self-inflicted. Second, the mechanics of publishing your own measurement as evidence are the data-benchmarks playbook; the guarantee H2 doubles as the first flag planted in the empty loyalty conversation this edition measured. Third, for a house with two names, the page should live under the name the answers choose.
Key Snippet, placed early: "[Vores bureau]s median fra brief til underskrift er [X] dage for [rolletype], målt over [antal] besættelser siden [år]." ([Our agency]'s median from brief to signature is [X] days for [role type], measured across [count] placements since [year].)
Slug: /hvor-hurtigt-leverer-[bureau] — the buyer's question shape, not a content-type label.
Kept alive: medians recomputed quarterly, the refresh dated on the page, and any figure resting on fewer than five placements says so.
Where the evidence must live: who listens where
This market's model citations are led by the Danish state's employment infrastructure: star.dk, jobnet.dk, skat.dk, Arbejdstilsynet, nyidanmark.dk (work permits), Danmarks Statistik, and then the business layer (DI, Dansk Erhverv) and a single review surface, Trustpilot.
That is the third citation regime in three editions: edition 5's judge read the review platforms, edition 6's read the financial regulator, this one reads the employment state. The pattern underneath is the same each time: the models verify a category against the institutions that regulate it. For an agency, the practical consequences are concrete: the compliance questions (ansættelsesret, GDPR, arbejds- og opholdstilladelse) are answered against official sources, so your own pages should cite those sources by name; and the evidence the state cannot hold (your placements, your time-to-fill, your references) is yours to publish or nobody's.
ChatGPT and Claude lean on text surfaces: authoritative articles under named authors, structured pages on your own domain, the business press. Gemini rewards fresh, dated material tied to the questions buyers ask. For all three, in this category, the missing surface is the same: nobody's numbers are anywhere.
The lesson
In an open category, the cut runs between generic and specific. The generalists get compared; the specialists get chosen; and one company demonstrates the whole mechanism under two of its own names. The criteria are unclaimed, the loyalty conversation is unclaimed, and the weakest default this series has measured means the answer is still up for grabs, for whoever publishes the stack, the number, and the name first.
Your category has its own version of this finding. The names differ. What the name promises does not.
Sources
The measurement behind this edition: three models (Gemini, Claude, ChatGPT), window 24 August to 20 September 2026, evaluation and decision measured daily (1,136 and 1,068 answers), retention and advocacy weekly (141 and 116 answers). Criteria, eliminators, and buyer language extracted per model per run.
The Sunday Shortlist
Get the decode, and the toolkit
One step left. A confirmation email is in your inbox. Click it, and your toolkit downloads.
That didn't send. Check the address and try once more.
The Sunday Shortlist decodes one category at a time: the questions buyers ask AI, the shortlist that comes back, and the criteria that decide it. Subscribers get every decode first, and the free toolkit gets you started tonight: a worksheet and three prompts to run the first check on your own brand.
One email a week · unsubscribe anytime · Terms