Measuring a Trip You Don’t Sell: An Analytics Practice for Visit California

The Challenge: Destination Marketing’s Measurement Problem

Someone opens four tabs at eleven at night and starts planning a week in California. They read about Joshua Tree, price a hotel in Palm Springs, check whether the rental car will find a charger and eventually book. On an airline’s site. On a hotel’s site. On a national park’s reservation system.

Never on the site that convinced them.

That is the position every destination marketing organization works from. Visit California inspires a trip it never sells. The transaction, the revenue and the proof of the whole thing happen on someone else’s platform, and the usual measurement stack — the one built for e-commerce, where the sale lands in your own database — has nothing to say about it.

The site had grown to match the ambition. Millions of page views a month, roughly three quarters of them on phones. Ten language editions. Hundreds of destination guides, seventy-plus road trip itineraries, a podcast library, hundreds of event pages and a set of campaign hubs running against national media buys. Behind it, a Google Tag Manager container holding years of accumulated instrumentation from a decade of agencies, campaigns and vendors, including a Universal Analytics property still firing from inside a JavaScript package that had not been updated in seven years.

Traffic was never the difficult part. Knowing what the traffic meant was.

The Strategy: When You Can’t Measure the Booking, Measure the Intent

Then build the measurement so the team trusts it on an ordinary Tuesday, not only in the quarterly deck. Four years of a DMO analytics practice, in six sustained moves.

Rebuild the instrument before reading it. We audited the tag container line by line and decided the fate of every tag: keep, repurpose, pause. Then we migrated to GA4 and designed a custom event taxonomy around the moments that indicate a real trip forming rather than a page merely loading: partner referral, itinerary download, map open, language selection, road trip game engagement and dwell thresholds at thirty, forty-five and sixty seconds. Roughly two dozen custom events, each one defined because someone could name the decision it would inform.

Treat consent as a data-quality problem. Most teams treat a consent platform as paperwork. We mapped 166 tags against five CPRA purposes, eighteen GDPR purposes and thirty-three custom vendors, with a minimum of four triggers on every tag, because the configuration decides how much of the audience the analytics can legally see at all.

Put site performance on a schedule. From January 2024 we pulled Core Web Vitals weekly across the site’s main templates, archiving every run. Later we automated it into a daily job across twelve template URLs covering the homepage, road trips, places to visit, things to do, the family hub and campaign pages. One rule made it operational: any score that fell more than five percent opened a ticket in the backlog with a named reviewer. Performance stopped being a project with an end date and became a system with an owner.

Make the content readable by machines. We added structured schema to the content types that had the most to gain, including events, podcasts, video, cities, regions and the Experience type that makes up the largest share of the site, so that search engines and AI crawlers could parse a road trip, a festival date or a destination as an entity with properties.

Instrument the search box properly. When we implemented Algolia for global site search, results rendered without a page load, which meant the default search tracking recorded nothing. We deployed code to capture both the search event and the query itself, validated it in production and documented the tracking gap in the reporting so no one would later read the silence as a drop in demand.

Move the data somewhere it can be asked questions. We automated the data pull into BigQuery and designed a Looker dashboard set covering website performance, international audiences, site search, Search Console and real-user monitoring, so the questions the team asked monthly stopped requiring a person to assemble the answer by hand.

The Outcomes: The Numbers Became Defensible, and the Audience Came Back Into View

Trust. The numbers became defensible. Rebuilding redirects as a server-side solution took production crawl errors down to 2,312, a 99.8% reduction, with zero 500-level errors remaining. The staging environment fell from 223,207 errors to 6,972. Every one of those errors had been a page a crawler could not read and a session that could not be attributed to anything.

Engagement. Performance stopped being a project and became a watched number. Three quarters of the audience arrives on a phone, which is where the work concentrated. Specific releases moved specific numbers: replacing an icon library with a sprite took mobile Largest Contentful Paint from an average of 21.28 seconds to 12.92 seconds, a 39.38% improvement; a layout-stability release improved Cumulative Layout Shift by ten points and reduced errors by 74%; and a redesigned 404 page, given real navigation and clear pathways onward, produced a 33% reduction in exits and 24% fewer 404 page views within a month of launch. What made those durable rather than momentary was the monitoring underneath them — a daily job across twelve template URLs, and a rule that any score falling more than five percent opened a ticket with a named reviewer.

Growth. Structured content earned its way up the results page. After event schema shipped, average search position improved by 7.46 positions, clicks rose 40.55% and impressions rose 158.92%. Podcast schema improved average position by 13.91 positions. Five months later the follow-up report showed events ranking 37% better on average, with more than thirty-five events entering the top ten and more than fifty improving their ranking by 30% or more. Six new podcasts entered search results with strong positioning on day one.

Intelligence. The audience came back into view. The consent configuration returned 99.3% of consumers trackable under CPRA and a 49.6% opt-in rate under GDPR, which is the difference between reporting on a fraction of the audience and reporting on nearly all of it. With the instrument rebuilt, questions became answerable that had not been before: which language editions convert attention into partner handoffs, which campaign hubs keep a visitor past sixty seconds, what people type into the search box when the navigation doesn’t have the word they want. In the final months of the engagement, AI assistants appeared in the reporting as their own acquisition channel for the first time. That category did not exist when the instrumentation was designed, and the instrumentation was ready to see it.

A destination marketing organization will never own the booking. It can own the evidence of intent, and that turned out to be a measurable thing.