Engagement metrics beyond the click

Someone opens the Trainline app on a Tuesday evening. London to Edinburgh, Friday morning. She scrolls the fare list, jumps to Saturday, comes back to Friday, prices a return three days out, closes the app. No ad clicked, no email opened, no booking. On Thursday she books, on desktop, logged in.

Every click-based report missed the strongest buying signal of her week: a person pricing one route across five dates in a single session. The behaviour between browsing and buying (session depth, saves, date flexibility, pre-arrival and in-stay app use) predicts bookings that no click registers. It is also the easiest class of metric to fool yourself with, because it is cheap to generate and flattering to report.

The signals that sit between browsing and booking

Between the inspiration stage and the checkout stage the funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.View full definition → lesson maps, most of what a traveller does leaves no click record worth the name. The candidates worth instrumenting:

• Session depth: screens per visit in the app or booking engine.
• Date-flex breadth: how many distinct departure dates a user prices in one session.
• Save and wishlist actions against a specific product, not a destination page.
• Itinerary-builder completion, where the user commits dates, party size and a pickup point.
• Repeat sessions against the same search within a rolling window.
• Pre-arrival engagement: app opens, check-in started, add-on browsing in the two weeks before travel.
• In-stay behaviour: mobile key use, in-app messaging, activities booked after arrival.

Against those sit the metrics that proxy attention and nothing else: email opens, scroll depth, generic homepage visits, raw time on page. Time on page is the sneakiest of them, because a user stuck on a confusing fare table reads exactly like a user fascinated by it.

Which signals actually predict conversion

The distinction that matters is behavioural specificity: does the action require the user to commit information or effort that only a real trip would justify?

Low predictive value (dashboard-friendly, conversion-weak):

• Email opens. Since iOS 15 shipped Mail Privacy Protection in September 2021, Apple Mail pre-fetches images for anyone who turned it on, so a meaningful share of your opens are machines.
• Destination-guide and homepage visits with no product attached
• Passive scroll depth on marketing content
• Time on page without a funnel stage attached to it

High predictive value (harder to get, worth the instrumentation):

• Date-flex breadth in one session: five dates priced for the same route is a person with a trip to take
• A Klook-style wishlist save followed by a return visit inside seven days
• Itinerary-builder completions with specific choices locked in
• Push opt-in combined with in-app browsing of one named departure
• Check-in started more than 24 hours ahead, which puts the user in front of the add-on shelf while there is still time to buy

Actions that cost time, personal data or a small commitment correlate with intent. Actions an inbox can perform on the user's behalf do not.

Two edge cases stop this from being a rule you can apply blind. First, a wishlist save decays: on activities marketplaces the save is a shopping-list gesture, and one nobody returns to within a month is closer to noise than to intent. Score saves with a decay weight and a return-visit multiplier, not as a running total. Second, repeat visits can mean indecision rather than convergence. Six sessions on the same fare over three weeks often means someone waiting for a price drop, and the right response is a fare alert, not a premium upsell. Date flexibility carries the same double meaning: it marks real intent and real price sensitivity in the same breath.

A worked comparison

Illustrative numbers, an activities marketplace mailing 100,000 past customers.

• 38,000 open the email (38% open rate).
• 6,000 click into the itinerary builder (6% of delivered, roughly 16% of openers).
• 900 complete the builder, selecting dates, party size and pickup.
• 310 book within 14 days.

Completion to booking: 310 / 900 = 34.4%. Open to booking: 310 / 38,000 = 0.8%.

Builder completers convert around 40 times more reliably than openers. A team spending its quarter on subject lines is polishing the weakest link in that chain; a team cutting two screens out of the builder is working on the strongest.

Session depth: useful, but read it in context

Depth per visit is reported as a health metric for almost every travel app. It only means something once you know where the depth happens.

• Depth across customisation screens (excursions, upgrades, seat or dining choices) mirrors real planning.
• Depth across FAQ, cancellation policy and contact screens usually signals friction, and often precedes abandonment.

In-stay depth inverts again. A guest browsing "today" experiences on day two of a city break is minutes from a purchase. The same guest deep in voucher and refund screens is a service failure in progress, and treating both as engagement hides the second one.

A practical rule: tag session depth by page category. A twelve-screen session inside the booking flow and a twelve-screen session bouncing between refund policy and contact us are opposite events with identical counts.

Simple session-quality scoring

Weight screens by stage rather than counting them equally.

session_quality_score = 
    (pages_in_planning_stage * 3) +
    (pages_in_booking_stage * 5) +
    (pages_in_support_or_policy_stage * -2)

Simplified and illustrative, not an industry standard. The principle holds: unweighted depth lets anxious browsing show up as high engagement.

One second-order warning. The moment saves become a target, product puts a heart icon on every tile, save volume doubles and save-to-book halves. The metric moved, the predictive model behind it broke, and nobody notices for a quarter because the dashboard went up. Re-fit the weights whenever the interface changes.

Benchmarks to anchor your read

Directional figures drawn from marketing benchmark aggregators such as Litmus and Mailchimp's industry benchmark reports, 2024-2025 data. Both sell email tooling, and which external sources are genuinely comparable to your mix is the benchmarking lesson's territory.

Building an engagement-to-conversion dashboard that isn't vanity

1. Rank metrics by cost-to-fake. Could a pre-loader, a bot or idle curiosity produce this signal without intent? Opens and impressionsimpressionsThe total number of times an ad or piece of content is displayed, regardless of clicks. Each display counts as one impression, even to the same person.View full definition → score low. Completed itineraries, priced date ranges and returned-to saves score high.

1. Never report an engagement metric without its downstream rate. Open to click, click to builder-start, start to completion, completion to booking, as one connected chain. Some teams call this a micro-funnel: a funnel hiding inside what looks like a single number.

1. Segment session depth by intent stage, as above, and re-score it after every redesign.

1. Fix identity before you trust any of it. This is where most engagement models quietly die. A wishlist save on logged-out mobile web and a booking in the app land in two separate profiles unless the anonymous ID is merged at login. Customer data platformsCustomer data platformsSoftware that unifies customer data from every source into one persistent profile that marketing, sales and service teams can act on.View full definition → such as Segment (which sells exactly this event pipelinepipelineAll active sales opportunities across the stages of the sales process, together with their combined potential value and probability of closing.View full definition →) handle the stitching, and they bill on tracked users and event volume, so instrumenting every scroll has a real line-item cost. Pick fifteen events that change a decision and ignore the rest.

🎬 [VIDEO: "Vanity Metrics vs Actionable Metrics" - youtube.com/results?search_query=vanity+metrics+vs+actionable+metrics+marketing - a concise walkthrough of why surface-level engagement numbers often mislead marketing teams, applicable directly to travel booking funnels]

Key Takeaways

• The signals that predict a booking (date-flex breadth, saves with a return visit, builder completions, pre-arrival check-in) leave no click record, so click reporting misses them entirely.
• Commitment is the test. In the worked example, builder completers convert roughly 40 times better than email openers as a group.
• Saves decay and repeat visits can mean stalling, not converging. Weight for recency and read a sixth visit to the same fare as price-watching.
• Session depth is meaningless unweighted: depth in booking screens is positive, depth in policy and refund screens is friction, and in-stay depth flips the reading again.
• Broken identity stitching destroys the whole model before the analysis starts, and every extra tracked event costs money, so instrument few things properly.