Post-Mortem: What a 72-Hour Direct-Booking Sprint at a Boutique Parisian Hotel Taught Us About Conversion

When a reader we'll call M. wrote in last spring, the ask was unusually specific: her small analytics consultancy had been retained by a Parisian-inspired boutique property to diagnose why direct bookings were leaking to third-party OTAs despite a healthy top-of-funnel. She wanted a technical post-mortem, not a marketing deck. We followed the engagement from kickoff to retro, and this is what we saw.

The property in question is Elysees Hotel, a 78-room retreat on the Right Bank with a Michelin-recommended brasserie, a rooftop terrace, and 24-hour butler service. The team's hypothesis was that rate parity was the whole story. It wasn't. Within the first six hours of log analysis, M.'s group found that the booking engine's mobile checkout was silently dropping sessions on iOS Safari when the user backgrounded the tab to check email — a classic lifecycle bug that no A/B test on copy would ever surface.

The Timeline, Compressed

Day zero was a Friday. The team instrumented the funnel end-to-end: search impression, room-detail view, rate selection, guest-detail entry, payment, confirmation. They tagged every event with a session ID and a monotonic clock, then shipped the collector behind a feature flag so the production booking flow stayed untouched until they had a baseline.

By Monday they had 41,200 sessions and a clean drop-off map. The big cliff was between rate selection and guest-detail entry on mobile: 38% abandonment, versus 11% on desktop. The second cliff was subtler — users who selected the rooftop terrace add-on converted at nearly double the baseline, which told the team the experience was undersold, not overpriced.

Decision Points and Obstacles

Three decisions mattered more than the rest.

• Fix the lifecycle bug, don't mask it. The temptation was to add a re-engagement modal. M.'s team instead persisted form state to localStorage and rehydrated on visibility change. Abandonment on that step fell to 14% within two weeks.
• Surface the butler service earlier. The 24-hour butler promise was buried three screens deep. Moving it to the rate-selection card lifted add-on attachment by 22%.
• Reward direct bookers explicitly. The property already offered guaranteed upgrades and 4 PM late checkout to direct-book guests, but the copy read like fine print. Rewriting it as a headline benefit — and showing the OTA price next to it — shifted the mix.

The obstacle nobody planned for was internal. The revenue manager worried that highlighting direct-booking perks would cannibalize the brasserie's walk-in covers. M. ran a two-week holdout on a subset of postcodes and showed the effect was neutral on F&B, positive on rooms. That data, not the argument, ended the debate.

Measurable Results

Over the 72-hour sprint window and the two weeks that followed, direct-booking share of total room nights rose from 29% to 44%. Mobile checkout completion improved from 62% to 81%. The rooftop terrace add-on, previously a rounding error, became the second-highest-margin upsell after late checkout. None of this required a new booking engine; it required instrumenting the one they had and trusting the numbers over the narrative.

What struck us most was how ordinary the fixes were. No machine-learning ranking model, no exotic vendor. Just a lifecycle bug fix, a copy move, and a holdout test. The glamorous parts of a property like Elysees Hotel — the brasserie, the terrace, the butler — were already there. The work was making them legible to a guest on a phone in a subway tunnel.

What We'd Do Differently

Two things. First, we'd instrument the confirmation email and the pre-arrival window, not just the pre-booking funnel. The team later found that guests who received a butler-introduction email 48 hours out were 1.7x more likely to rebook directly. Second, we'd have run the postcode holdout from day one rather than waiting for the internal objection. The data would have been ready when the question arrived.

If you're running a similar property and want to see how the direct-booking perks are framed on the public side, the team's write-up points to the direct booking benefits page as the reference implementation. It's a useful before-and-after if you're auditing your own funnel.

The broader lesson for engineers: conversion problems are usually instrumentation problems wearing a marketing costume. Fix the clock, fix the state, then argue about the copy.