Danish debt collection: The second-most recommended name was retired in 2017.

Somewhere in Denmark right now, a bookkeeper with a stack of overdue invoices is asking an AI model which debt collector to trust. The model answers in seconds: a shortlist, the risks, a favourite. No bureau hears about that conversation. We measure it.

What we measured

From 22 September to 4 October we ran this market's buying questions through the models across the decision journey, measured with three models: Gemini, Claude, and ChatGPT. All four stages are measured daily; the late-journey stages (retention, advocacy) run on a smaller set of questions each day, so their shares rest on fewer answers.

The questions are the ones buyers ask, in their own words:

• "Hvad koster det at sende en sag til inkasso?" (What does it cost to send a case to debt collection?)
• "Hvad er jeres succesrate med at inddrive gæld?" (What is your success rate at recovering debt?)
• "Hvad betyder 'sober tone' i praksis for jer?" (What does 'measured tone' mean for you in practice?)
• "Hvad er jeres provision, hvis jeg ikke får pengene tilbage?" (What is your commission if I do not get the money back?)

Notice the third one. The buyer is not only hiring an enforcer; they are lending out their own reputation. The debtor being chased is often a customer they hope to keep, and the buyers ask about tone before they ask about speed.

The judging sheet

The verdict, filtered through the state's own list

The police's register of authorized inkasso companies is public. The table below shows the measured shares for the recommended brands that appear on it.

A note on the table: it is limited to companies on the authorization register as of 5 October 2026 (law firms, which are exempt from the authorization requirement, also collect debt and are not listed). The first two columns rest on 519 and 456 answers in this window; the last two on 74 and 59, because fewer loyalty questions run each day, so shares in the loyalty columns move in coarser steps. *Danske Inkasso's evaluation share fell below the export's reporting cut; its decision share is shown.

Intrum is the category default: 69.7% of decision answers, converting cleanly from 70.5% at evaluation. That part of the answer is real, authorized, and earned.

It is everything around the default that misbehaves. The models' full answers do not look like this table, because the full answers are not limited to companies that exist.

The runner-up retired in 2017

The second-most recommended name in this market is Lindorff: 51.1% of evaluation answers and 40.4% of decision answers.

Lindorff and Intrum Justitia merged on 27 June 2017, and that October the combined group dropped both legacy names to operate as Intrum. As a condition of the merger, the European Commission required Lindorff's businesses in Denmark, Estonia, Finland and Sweden to be divested; Lowell completed that purchase in March 2018, and the Danish business operates today as Lowell Danmark A/S. As of 5 October 2026, no Danish company operates debt collection under the Lindorff name, and no Lindorff entry appears on the authorization register. The only Lindorff in the Danish company registry is a music business.

Now hold the two name strings up against the measurement. The retired name, Lindorff: 40.4% of decision answers. The living name of the company that bought its Danish business, Lowell: 1.9% of evaluation answers. The models are not recommending a company. They are recommending a memory, and the memory outpolls the successor twenty to one.

This is the pattern this series has been tracking for seven editions, taken to its end state. Edition 7 measured one company converting in opposite directions under two of its own living names. Edition 8 found retired name forms drifting in the fog around a three-way tie. This market puts a retired name on the podium itself: the recommendation ledger is keyed to name strings, nobody maintained the string after the rename, and eight years later it is still collecting 40% of the choice.

Every point of that 40% is unclaimed market share, because the name that holds it cannot take the call.

The loyalty stages belong to the ghosts

Then the buyer becomes a client, and the answers leave the register entirely.

The retention answers are led by Kreston Danmark at 86.5%. Kreston Danmark is a network of independent audit and accounting firms, part of Kreston Global; it is not an inkasso bureau and holds no entry on the authorization register. The advocacy answers are led by Kreston (79.7%) and Bisnode Inkasso (78%). Bisnode was acquired by Dun & Bradstreet, completed on 8 January 2021; the Danish entity today carries the name Bisnode D&B Danmark A/S and is registered for data processing, not debt collection. Meanwhile the category default, Intrum, falls to 33.8% of retention answers.

The loyalty samples are small (74 and 59 answers), so these shares move in coarse steps. But the shape survives any rounding: at the two stages where the models describe who clients stay with and recommend, the leading names are an accounting network and a brand that was absorbed five years ago. The same vacuum this series measured in Danish IT staffing is here filled not by names from other segments, but by names from other decades.

The fog has more layers. Visma Severa appears in the decision answers; it is project-management software for consultancies, not a debt collection service, and the Danish company Visma Collectors A/S ceased in 2013. A further set of generic provider names in the answers exists nowhere at all: no registry entry, no company. And in the opposite direction, the register itself holds a trap the models are poorly built for: Dansk Inkasso and Danske Inkasso are two separate, real, authorized companies whose names differ by one letter. A ledger keyed to name strings has no safe move in this category: the strings point at the dead, the miscast, the invented, and the nearly identical.

The defence does not change, but the stakes here are the highest the series has measured: one name, spelled one way, everywhere; an Organization schema on your own domain that declares the canonical name, lists variants as alternateName, and ties your register entry, CVR record, and profiles together with sameAs. The mechanics are the schema-as-AI-context technique. In a market where the second-most recommended name has been retired for eight years, being unambiguous is not hygiene. It is the whole game.

The eliminator: the fifth market cut on the same question

The most severe elimination trigger in this market is conflict of interest:

"Potentiel interessekonflikt med skyldneren eller andre parter er uacceptabelt."

(A potential conflict of interest with the debtor or other parties is unacceptable.)

Every model applies it. The buyer's version is sharper still: "Arbejder de også for vores konkurrenter?" (Do they also work for our competitors?) In debt collection the conflict question has a particular edge, because the bureau stands between the buyer and someone the buyer often wants to keep as a customer. Who else the bureau works for is not a procurement detail; it decides whose relationship gets protected when interests collide.

This is the fifth market in nine where conflict of interest is the hardest cut, after Danish IT outsourcing, Danish spend management, UK wealth management, and Danish insurance brokers. Five regulated or trust-bearing categories, one recurring verdict: the models eliminate on the question no provider has answered in citable form. Silence is scored as an answer.

Nobody owns a criterion

Ten criteria decide this market. Not one has an owner. Nine categories decoded in this series, ninety criteria measured, zero owned.

This market's open questions are unusually concrete: no bureau owns cost transparency ("Er der skjulte gebyrer udover selve inkassosalæret?"), no bureau owns documented outcomes ("Hvad er jeres succesrate for virksomheder som min?"), no bureau owns tone ("Er tonen i jeres kommunikation professionel og respektfuld?"), and no bureau owns the conflict question that eliminates hardest. A category with a 70% default and zero owned criteria is not a settled market. It is a market where the default wins on familiarity while every question that could unseat it stands open.

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 "verify us": in a category where the models recommend the retired and the miscast, the strongest position a real bureau can take is to make its own existence, authorization, and loyalties effortless to check.

Where the evidence must live: who listens where

This market's model citations are the purest institutional layer the series has measured: the courts (domstol.dk), the data authority (Datatilsynet), the financial supervisor (Finanstilsynet), the business authority (Erhvervsstyrelsen), the consumer ombudsman, the industry body (inkassobranchen.dk), and the parliament's own site. Not one bureau site leads, and no review platform breaks the pattern the way it did in customer service software.

The consequence is the same one this series measured in Danish IT staffing, sharpened: in categories the state supervises, the models verify the market against the state. Your register entry is a load-bearing surface of your AI reputation. Your own pages win by citing the institutions by name and by publishing the evidence the institutions cannot hold for you: your prices, your conflict policy, your documented outcomes.

ChatGPT and Claude lean on text surfaces: official registers, authoritative articles, structured pages on your own domain. Gemini reads the same institutional layer and rewards fresh, dated material tied to the questions buyers ask. For all three, in this category, the verification questions are standing open on every surface at once.

The lesson

Sources

The measurement behind this edition: three models (Gemini, Claude, ChatGPT), window 22 September to 4 October 2026, evaluation and decision resting on 519 and 456 answers, retention and advocacy on 74 and 59. Criteria, eliminators, and buyer language extracted per model per run. Company facts verified against the police's register of authorized inkasso companies (5 October 2026), the Danish company registry (CVR), and the named merger announcements from Intrum, Lowell, and Dun & Bradstreet.

The Sunday Shortlist

All editions