From weighting panels to running a publisher's data office

Mandate

Give the group one honest set of audience and subscriber figures, a consent practice defensible to readers and regulators, and data products that serve journalism.

Outcomes

• Digital subscriptions grew from 251,000 in March 2021 to 418,000 at the end of 2025, while the silent share fell from 12 to 6 per cent.
• Replaced thirteen consent banners with one; logged-in reading rose from 22 to 61 per cent of page views between 2021 and 2025.
• Wrote to 41,000 silent subscribers in 2023; card chargebacks fell 60 per cent and first-year churn for new subscribers from 31 to 24 per cent.
• Direct-sold advertising on first-party segments grew by DKK 58 million a year by 2025, offsetting a DKK 41 million fall in programmatic revenue.
• Merged thirteen subscriber databases into one platform by 2024, with a single published definition of an active reader.

Decisions in context

I joined Vesterlund in March 2021 as its first chief data officer. The group had grown digital subscriptions fast but had built its data title by title: thirteen consent banners, thirteen subscriber databases and at least four definitions of an active reader. The chief executive asked me for three things: figures the board and advertisers could trust, consent we could defend, and data work that helped the newsrooms instead of competing with them. The first two years went mostly on consolidation and the consent redesign, which cost programmatic revenue before logged-in reading replaced it. The silent-subscriber programme in 2023 was the hardest commercial argument of my tenure. I lead 74 people across Copenhagen and Aarhus, present to the board every quarter and report to the Audit and Risk Committee on data protection and data risk; our data protection officer reports independently of me.

Reason for the move

Current role. Not seeking an executive move; considering one non-executive directorship with the board's consent.

Mandate

Build personalisation for the streaming player without compromising the broadcaster's public-service obligations.

Outcomes

• Designed a reserved editorial share of a third of the player's front page, adopted as board policy in 2018.
• Weekly users watching at least one news or documentary programme rose from 34 to 47 per cent.
• Weekly reach of Sámi-language programmes doubled from a small base.
• Kept children's viewing out of individual profiling while weekly use of the player by under-thirties grew 22 per cent between 2017 and 2020.

Decisions in context

I moved to Oslo in February 2016 to lead audience data and personalisation at Fjellvang Kringkasting, a public-service broadcaster whose streaming player was losing younger viewers. My team of 31 built the player's recommendation and measurement systems on a budget of NOK 64 million a year. The central question was how far a public broadcaster should personalise. A fully personalised test raised viewing time but nearly removed news, regional and Sámi-language programmes from most screens. I worked with the editors on a reserved share of the front page and a short set of rules no model could override, and the board adopted them as policy in 2018. I learnt to work in Norwegian, to sit in editorial meetings as a guest and to present algorithmic choices to a board as the editorial choices they are. I left to take charge of a whole data function in my own country.

Reason for the move

Invited to become the first chief data officer of a Danish news group, responsible for consent, subscriptions and measurement as well as recommendation.

Mandate

Build a data-products business from network insights and run analytics for the consumer business.

Outcomes

• Grew location-insights revenue from DKK 4 million in 2011 to DKK 31 million in 2015, mainly from municipalities and transport planners.
• Set a fifty-device reporting floor and stated the share of suppressed cells in every delivery.
• Declined a DKK 9 million shopping-centre contract that required fifteen-minute and repeat-visit data.
• Extended randomised uplift testing to every consumer retention and upsell campaign.

Decisions in context

In May 2010 I was promoted to direct Brosund's data products and analytics, with a team that grew to 42 people. The main new business was location insights: aggregated counts of devices by grid square and hour, sold to municipalities, regional transport planners and a few retailers. I wrote the product's rules before most customers had asked for anything: a floor of fifty devices for any reported count, suppression below it, no movement paths and no repeat-visit measures. Revenue grew from DKK 4 million in 2011 to DKK 31 million in 2015, short of the original plan, partly because I declined a large shopping-centre contract in 2013. I kept responsibility for analytics in the consumer business, where uplift testing became the rule for every retention and upsell campaign. By 2015 I wanted to work on questions that were editorial rather than commercial, and an approach from Oslo offered exactly that.

Reason for the move

Wanted to apply data work to editorial and public-service questions rather than network commerce.

Mandate

Make the postpaid churn model and retention programme measurably effective.

Outcomes

• Introduced randomised holdout groups for every retention campaign.
• Rebuilt retention targeting around measured response; retention spending fell by DKK 46 million a year.
• Annual postpaid churn fell from 18.5 to 16.9 per cent between 2007 and 2009.

Decisions in context

Brosund hired me in September 2006 to manage customer value analytics for its postpaid business. The existing churn model ranked customers by their probability of leaving and sent the highest-risk a discount or a new handset. I introduced randomised holdout groups and found that most discounts went to customers who would have stayed anyway, while calls to some undecided customers prompted them to compare competitors' offers. We rebuilt the targeting around the measured effect of contact rather than predicted risk. Retention spending fell by DKK 46 million a year, about a quarter of the budget, and annual postpaid churn fell from 18.5 to 16.9 per cent between 2007 and 2009. The work taught me to ask of every model what our action would change, and it gave me the standing to propose a data-products business to the commercial director.

Reason for the move

Promoted to lead data products and analytics.

Mandate

Run weighting and estimation for newspaper readership surveys and a television viewing panel.

Outcomes

• Rebuilt the weighting of a television viewing panel that under-represented young adults by almost half.
• Introduced published confidence intervals in readership reports for regional newspapers.
• Led readership estimation for fourteen newspaper titles.

Decisions in context

In January 2004 I joined Havnø Analyse, a Copenhagen firm of about sixty people that ran readership surveys for newspapers and weighted a television viewing panel for broadcasters and advertisers. I led the estimation work for fourteen newspaper titles and rebuilt the panel's weighting after we found that young adults were under-represented by almost half. It was my first experience of an industry that trades on its own audience numbers, where every methodological choice had a buyer and a seller. I introduced confidence intervals into our regional readership reports, which some clients disliked because they showed that a reported rise lay within the margin of error. I left in 2006 because I wanted to work inside a company that would act on what its data showed, rather than advise one that might choose not to.

Reason for the move

Wanted to work inside a company that acted on its data, rather than advise clients who might not.

Mandate

Design weights and nonresponse adjustments for large household surveys.

Outcomes

• Built nonresponse adjustments for four large household surveys.
• Co-wrote a methods note on weighting for hard-to-reach groups, used across the institute for several years.
• Trained six student assistants in survey estimation.

Decisions in context

After completing my master's degree in statistics at Nørrestrand University in June 2001, I joined the Lerbjerg Institute for Survey Research, a non-profit institute in Copenhagen that ran household surveys for ministries, municipalities and research councils. I designed weights and nonresponse adjustments for four large surveys. Most of the craft concerned people who did not answer: night-shift workers, students who moved every year, recent arrivals who did not read Danish. The published estimate depended on what we assumed about them, and I learnt to put those assumptions in the methods note rather than leave them in my code. I co-wrote a note on weighting for hard-to-reach groups that colleagues used for years afterwards. I left at the end of 2003 to apply the same methods to audience measurement, where every assumption had a commercial consequence as well as a statistical one.

Reason for the move

Wanted to apply survey methods to commercial audience measurement.