Twenty-seven years of measuring people carefully

Mandate

Make subscriber, viewing and network data useful to the business on terms households can understand, and give the board an honest account of data risk.

Outcomes

• Moved outside use of viewing logs to opt-in consent; 33,600 boxes enrolled within nine months
• Split churn-model output into repair and price routes; repair-group churn fell from 2.7% to 1.5% a month
• Introduced AI call summaries under a no-training contract and a monthly audit of 400 summaries
• Began a single register of partner data arrangements, each with a review date

Decisions in context

The appointment made data a standing item for the executive committee rather than a technical service to it. I report to the president and present to the board on data risk, partner arrangements and our use of AI. My office answers to our audit and supervisory board members for personal-data handling, which keeps me careful about the difference between what I recommend and what I can decide alone. The three cases on this site all belong to this period, and each involved a colleague with responsibility for revenue or service who disagreed with me for sound reasons. My team includes engineers who maintain the platform, analysts who build models and four people who handle subscriber data requests and partner reviews. I spend part of every month in the contact centres and with field crews, because that is where the effects of our data decisions first become visible.

Reason for the move

Present role, which I intend to keep; an outside directorship would sit alongside it with my president's approval.

Mandate

Bring the merged company's personal data under one inventory and build analytics the business would actually use.

Outcomes

• Inventoried personal data across 41 systems and retired 9 of them
• Unified two customer numbering schemes ahead of a billing migration
• Built the first broadband churn model, later revised after review
• Supported network planning through the surge in home-broadband demand during the pandemic

Decisions in context

The merger had joined companies without joining their records. Each original station kept its own customer numbers, its own terms and its own habits about what to keep. My first year went on an inventory: which systems held personal data, why, who could see it and how long it stayed. We found forty-one systems and retired nine. That work was unglamorous and made everything later possible, including a single consent record. I also built our first broadband churn model in 2020, as demand for home connections rose sharply while people worked from home. The model was accurate by the measures I chose at the time. I did not ask field technicians to examine its inputs, and four years later that omission became the subject of one of my cases. I mention it here because the lesson started in this job rather than the next one.

Reason for the move

Appointed chief data officer when the board created the role as an executive-officer position.

Mandate

Give marketing and customer operations a reliable view of subscribers, tenure and reasons for leaving.

Outcomes

• Replaced a free-text cancellation field with recorded reasons agreed with contact-centre supervisors
• Linked evening speed complaints to specific network segments
• Produced the first monthly subscriber report used by the board

Decisions in context

Moving from a broadcaster to a cable company changed the person at the end of my analysis. At Sennami a viewer was an estimate. At Tamakazura a subscriber had a contract, a router, a payment history and a technician who might have visited last week. My six analysts and I began with cancellations, because the company recorded them as free text that nobody could count. We introduced a short list of recorded reasons, agreed with contact-centre supervisors, and kept a free-text field for anything the list missed. Within a year the reports showed that evening speed complaints clustered on particular network segments in hill towns, and network operations used that to plan inspections. I also learnt how the merged stations' different terms limited what we could do with each subscriber's data, which prepared me for the governance role that followed.

Reason for the move

Promoted to lead data governance and analytics across the company.

Mandate

Measure catch-up viewing and advertising delivery in a form that agencies and advertisers would accept.

Outcomes

• Defined completed-advertisement measurement for the catch-up service
• Kept streaming counts and broadcast ratings in separate, defined columns in sales reporting
• Introduced a correction process that told affected advertisers about reporting errors

Decisions in context

The catch-up service gave us more data in a week than the research department had seen in a year. Every start, pause and completed advertisement arrived as a record. My nine colleagues and I defined what counted as a completed advertisement and how to treat viewers who skipped ahead. The difficult arguments were internal. Sales colleagues wanted a single audience figure combining broadcast ratings and streaming plays; I argued that they described different things and should sit side by side, each with its definition. Some presentations became less striking. Agencies, who compared our reports with their own tags, came to rely on them. When we found an error in our own counts in 2014, we sent a correction to the affected advertisers before any of them asked. By 2016 distribution was moving to a shared industry platform, and I wanted work closer to the households being measured.

Reason for the move

Catch-up distribution moved to a jointly operated industry platform, narrowing the role; joined a cable operator closer to the people being measured.

Mandate

Interpret ratings and survey results for programming and advertising sales.

Outcomes

• Built a weekly programme-retention report used by the programming department
• Wrote advertiser reports that stated sample sizes and margins of error
• Ran viewer surveys for regional news and drama programming

Decisions in context

I arrived in Osaka with a panel analyst's caution and found myself between two departments with different questions. Programming wanted to understand why a drama lost viewers in its third week or why a regional news segment held its audience. Sales wanted a figure an advertiser would accept. I built a weekly retention report for programming and wrote advertiser reports that stated sample sizes and margins of error, which some sales colleagues initially thought made us look uncertain. Over time the agencies treated those reports as more credible than the alternatives. I led four researchers and managed the purchase of ratings and our own surveys. The job taught me to translate between people who use the same numbers for different purposes, and to refuse a figure that answered neither question well. That discipline became more necessary once streaming multiplied the volume of records.

Reason for the move

Asked to lead analytics for the broadcaster's new catch-up service.

Mandate

Weight panel results so that they represented the populations clients paid to understand.

Outcomes

• Introduced a monthly report on households that stopped responding
• Revised weighting targets after a census update
• Trained two junior analysts in panel quality checks

Decisions in context

My first employer sold panel research to broadcasters, agencies and advertisers. I weighted diary and online panels so that their households resembled the populations clients wanted to understand. The technical work was careful arithmetic. The lasting lesson came from the households who stopped responding. They were not a random group: younger single people and shift workers dropped out faster than retired couples, and a report that ignored them looked tidier than it should have. I introduced a monthly report on non-response so that clients could see how the panel was changing. I also learnt how much panel households were trusting us with, and how carefully the company explained what it would and would not do with their answers. By 2005 I wanted to work for an organisation that acted on audience research rather than produced it, which is why I moved to a broadcaster.

Reason for the move

Moved to a broadcaster to work with the people who acted on audience research.