Chief Data Officer / Europe
Four times the honest number was the smaller one
- Paying digital subscriptions, end of 2025
- 418,000
- Across Kronvig Tidende, Kapitalbladet and eleven regional titles; 251,000 in March 2021
- Page views read by logged-in readers, 2025
- 61%
- Up from 22 per cent in 2021, after refusal was given equal weight in consent
- Colleagues in the data office
- 74
- Data engineering, analytics, privacy engineering and audience research
Decisions & consequences
01Writing to 41,000 subscribers who had stopped reading
In spring 2023 our digital subscriptions stood at about 340,000 and were growing, but a quarterly analysis by my team showed what the headline concealed. About 41,000 paying subscribers, 12 per cent, had not opened a single article, newsletter or e-paper edition in ninety days. Most were on monthly plans renewing by card. Our churn model rated them low risk, because inactive people rarely cancel; they simply forget. Complaints about difficult cancellation were rising, card chargebacks had doubled in a year, and two regional editors were receiving letters from readers who felt cheated.
I proposed writing to every one of them, saying plainly that they had not used their subscription and offering three choices: a pause of up to three months, a cheaper plan with fewer titles, or cancellation with one click from the email. Our chief financial officer put the cost at up to DKK 30 million of annual revenue and argued that we were volunteering losses no competitor would accept. Our commercial director wanted a quieter re-engagement campaign without the cancellation link. The editor-in-chief of Kronvig Tidende supported me, and the chief executive took the proposal to the board, which approved it and asked for a report after six months.
We sent the letters in three waves between May and August 2023, each with a randomised holdout group so that we could measure the effect. Of the 41,000, 9,800 cancelled and 6,100 paused or moved to a cheaper plan; about 14,000 resumed reading within two months. The net loss was about DKK 23 million of annual revenue. Chargebacks fell by 60 per cent within a year, and complaints about cancellation almost disappeared. First-year churn among subscribers who joined after the programme fell from 31 to 24 per cent, which we attribute partly to a simpler exit making new readers more willing to subscribe.
The silent share is now a standing measure in the board pack, at 6 per cent in mid-2026, and dormancy letters go out automatically after ninety days.
What I would do differently: I sent the first wave before customer service had been trained on the pause option, and for three weeks readers who rang were told it did not exist. A kinder policy has to reach the people answering the telephone first.
Giving refusal the same weight as consent on the first screen
When I arrived in 2021, Vesterlund's thirteen titles used thirteen consent banners, configured separately as each site had been migrated. Most placed a large accept button on the first screen and hid refusal behind a settings link and two further screens. Acceptance averaged 84 per cent. Regulators across Europe were questioning designs of that kind, and our readers' panel told us they found the banners evasive. For me the deeper problem was that an audience built on reluctant consent was not one I could defend to advertisers or to our board.
I proposed one banner for the group, with accept and refuse buttons of equal size on the first layer, purposes written in plain Danish and a reader's choice remembered across all our titles. Our advertising director opposed it. Open programmatic auctions depended on consented identifiers, and he expected that line of revenue to fall by a fifth with no certain recovery. He also pointed out, correctly, that several competitors were keeping the older designs. The chief executive asked us to model the downside together, and we agreed to build logged-in reading before the change rather than after it.
We launched the new banner in March 2022, with a free login that unlocked three articles a week and personalised newsletters. Acceptance fell to 63 per cent. Programmatic revenue fell by DKK 41 million a year, about 16 per cent, and did not recover. Over the following three years logged-in reading rose from 22 to 61 per cent of page views, and we built advertising segments from declared interests and on-site reading that we could explain to readers. Direct-sold advertising using those segments grew by DKK 58 million a year by 2025, and total advertising revenue was back above its 2021 level that year, with a different mix. When a reader later complained to Datatilsynet about our banner, the complaint was closed without further action.
The larger change was internal. Consent became a product decision owned by a named executive, not a setting left to each site's developer.
What I would do differently is the sequence with advertisers. We told agencies about the change a fortnight before launch, and several held back spending until they understood the new segments. Three months' notice and a test campaign would have cost us less.
Declining fifteen-minute footfall counts for a shopping-centre owner
By 2013 Brosund Mobil's location insights business sold aggregated counts of mobile devices by grid square and hour to municipalities, regional transport planners and a few retailers. Revenue was DKK 19 million a year, and the plan approved by our chief executive called for DKK 40 million by 2015. The largest prospect was a property company that owned four shopping centres. It offered a three-year contract worth DKK 9 million if we would supply counts at each centre entrance every fifteen minutes, the share of visitors who returned within a month and the home districts of repeat visitors.
The request was technically feasible. My concern was what it meant in practice. At fifteen-minute intervals a side entrance on a weekday morning might see a handful of devices, and repeat-visit analysis requires following the same device across weeks, which is the beginning of a movement profile. Our lawyers advised that the anonymisation could be argued. I thought that was the wrong test: published research had shown how few time-stamped points identify a person, and our customers' privacy should not rest on whether an argument could be made. Our commercial director and the chief executive's office wanted the contract, and they had a fair point that a competitor might simply take it.
I recommended declining the specification and offering hourly counts per centre, with any cell under fifty devices suppressed and no repeat-visit or home-district measures. The property company refused and later bought a narrower product elsewhere. We missed the 2014 revenue plan by DKK 12 million, and I explained the shortfall at two quarterly reviews.
Over the next two years municipalities and the regional transport authority expanded their contracts, partly because our method note, which stated the floor and reported the share of suppressed cells in every delivery, made procurement easier for public buyers. Revenue reached DKK 31 million in 2015. When journalists later investigated another operator's sale of location data, our customers asked to see our method note rather than cancelling.
What I would do differently: I would have written the floor into the product's public description at launch in 2011, not in answer to a customer. Principles first announced under commercial pressure look like negotiating positions, however sincerely they are held.
Reserving a third of a streaming front page for editors
In 2017 Fjellvang Kringkasting's streaming player had 1.6 million weekly users, but viewers under thirty were spending less time with it each quarter. The head of streaming product proposed a fully personalised front page ranked by predicted viewing time, as global services did. Early tests were persuasive: viewing time per visit rose about 9 per cent. They also showed what the model did not value. News, regional programmes and Sámi-language content appeared far less often, because they were watched in shorter sessions or by smaller groups, and children's programmes, which we did not track by individual profile, were scarcely recommended to family accounts at all.
The editors' answer was to oppose personalisation altogether. The streaming team's was to treat editorial concerns as a small ranking penalty. I thought both were wrong. A public-service broadcaster has a duty to reach the whole audience with programmes a viewing-time model undervalues, and it also has a duty to be used. I proposed a reserved share: a third of front-page positions chosen each morning by an editorial desk, a fixed rule that at least one news item appears on the first screen, and no individual profiling of children's viewing. The rest of the page would be personalised. The head of streaming product argued that we were knowingly giving up viewing time, which was true. The director of programmes argued that two-thirds was still too much to hand to a machine.
The director-general approved the design in March 2018, and the board adopted it as policy, to be reported each year in the broadcaster's public-service accounts. In the first quarter viewing time per visit was 4.2 per cent below the fully personalised test; after a year, with better editorial selection, the gap was 1.5 per cent. The share of weekly users watching at least one news or documentary programme rose from 34 to 47 per cent, and weekly reach of Sámi-language programmes doubled from a small base.
The reserved share became published policy, accompanied by a plain account of how the rest of the page was ranked.
What I would do differently: I would have brought the editorial desk into the model's evaluation from the first test. For months the editors saw only results, and their scepticism was partly a reasonable response to being shown conclusions instead of evidence.