Building an Effective Data Product Strategy — Taylor Wells, News Corp

With 1.5 years of runway before launch, built one first-party clickstream system (Glimpse) as the only tracking integration in the app, normalized across devices, so data went only to Disney (simpler GDPR/consent, no vendor seeing volume). Because the team had never done it at this scale, also signed an Adobe contract and integrated it lightly so it could be pulled out. Compared both in the first weeks after launch, then removed Adobe. Event volume later grew 3-4x.

Instead of events like download_clicked/started/paused/completed, track a generic click action with hierarchical keys (button within container within set within page). Map keys to meaning behind the scenes and enrich the stream downstream (e.g. classify as 'project created') for each team, while keeping the raw data. Capture state changes instead of repeating ~100 identical metadata fields on every event. Give the system a memorable, unambiguous name (Glimpse, replacing 'UAT', which clashed with User Acceptance Testing) so everyone knows what 'the events' means.

Every day by noon, a data person sent the Disney+ president an email with exactly three bullets (up to two sub-bullets each): what changed yesterday and what to focus on, with context a dashboard hides (e.g. an EU sign-up rise came from a Netherlands promo that missed its target). The writer must be freed from low-value report requests and trusted to pick the right things.

Bluey sat in a Disney Junior row at the bottom of adult profiles (25 rows down, 15-20 tiles across), dropped out of Continue Watching once finished, ranked below Blue's Clues in search, and kid profiles had no search. Engagement showed 98-99% episode completion, full-season completion and rewatching. Short term: move the title or row higher. Long term: nudge parents to set up child profiles by showing the benefit (their kid's shows always at hand, cleaner adult recommendations). Fast real-time signals can also feed merch, live shows and parks planning.

Executives assumed buckets like Star Wars/Marvel super fans or Disney animation fans. Data science and analytics teams were given room to explore and found groups such as parents with young kids, parents with older kids, adult couples, single adults and teens shared more behavior patterns, and the team reoriented around them.

Started with off-the-shelf Amplitude plus a few semantic events (app created, project created) and a generic page-loaded event with the path broken down, which allowed retrofitting later questions. When a restructuring of the product changed what core terms meant, added an ETL step that encodes each definition change by date into a normalized 'super definition'. Now close to running its own event tracking on top of its data warehouse.

Started with bare-bones logs written to a SQL database, before BI tools like Looker or a mature Tableau were affordable for startups.

Use off-the-shelf or open-source tools early unless there is a real chance of sudden massive scale. Agree up front on review points: if moderately successful, review within 3 months whether to cut events or offset cost; if wildly successful, review within 1 month how fast to deploy an in-house system, then run both for about a month to compare metrics before switching off the old one. Get leadership to accept that reports won't match year over year after a switch (the GA360-to-GA4 move stalled for years over this).