← Back to work

From dead ends to dynamic choice

How I designed the multi-provider architecture for a travel platform with 90,000+ activities, and discovered through a hands-on data audit that the real problem wasn't the UI, it was the data.

Role Principal Product Designer
Company Civitatis (travel tech)
Scope Data audit, information architecture, interaction design, prototyping
Year 2026
Marketplace platform, two people reviewing Civitatis product comparison interface
# marketplace-hero.jpg · image · fill · 21:9

Executive summary

Civitatis operates a catalog of 90,000+ travel activities globally. The booking flow was tightly coupled to a single provider per activity. When availability failed, the user hit a dead end.

No alternatives. No fallback. No recovery.

The business opportunity was clear: recover lost bookings by enabling multi-provider comparison at the moment of unavailability.

What started as a UI redesign became a re-architecture: I mapped the data model that had to exist before any comparison UI could work, and designed the flows on top of it.

Concretely: I ran the catalog data audit, proposed the normalized schema, designed the end-to-end booking flow, and built the prototypes used to validate it.

Strategic context

In a catalog of this scale:

  • Availability varies constantly
  • Providers differ in pricing models
  • Modalities are inconsistently structured
  • Core booking inputs are overloaded

The result: Silent revenue leakage.

The hypothesis: decouple availability from a single provider, compare alternatives in real time, and booking probability goes up.

But early exploration revealed a deeper constraint.

You cannot compare what the system cannot define consistently.

This case study is password-protected.

Request access via email

Before
user selects activity PDP Product detail page one provider Single source not available No options found
After
user selects activity PDP Product detail page several providers Multiple sources option 1 Available option 2 Not available option 3 Available

The data audit

Before designing the comparison UI, I audited the catalog structure to understand what we were actually comparing.

01

Modality field overload

A single field was being repurposed to represent 9 different meanings: accommodation type, menu tier, duration, equipment, meeting point, pickup, route, schedule, and service plan.

02

People input semantics

The universal "number of travelers" field was overloaded: sometimes vehicle type, equipment, or even duration (1 "person" = 1 hour).

03

Hidden critical information

Key booking details, duration, meeting point, what's included, were buried inside unstructured description fields instead of structured data.

The UI wasn't the issue. The system lacked a consistent schema.

A unified booking flow

I redesigned two things: how availability is queried, and how options are presented.

Simplified availability input

The booking form was reduced to its essentials: Date + People. No upfront provider selection, no modality filtering. This single query fans out to every connected provider simultaneously.

Intelligent option surfacing

All returned alternatives are displayed. When identical options exist from different providers, an algorithm selects the "optimal option". I defined the scoring criteria, price, ratings, cancellation flexibility, and provider reliability, working with product and engineering on the weighting.

1 Select date & people
→
2 Query all providers
→
3 Compare alternatives
→
4 Show optimal option
Before

Book your activity

Select your preferences

Date
15 March 2025
Time
10:00 AM
Activity type
Guided tour
Number of people
2 adults
Check availability
Activity not available
After

Book your activity

Select date and group size

Date
15 March 2025
Number of people
2 adults
Show options
Walking tour 2h · Small group
€25
Guided visit 3h · Skip the line
€42
Private tour 4h · Exclusive
€89

Designing for low / uncertain availability

Not all providers expose real-time inventory. Some calendars sync with latency, creating states where availability is uncertain rather than definitively unavailable.

1. All dates remain selectable

Even dates marked low or unavailable stay selectable: one provider may be sold out while another still has inventory. Blocking selection would prematurely close the booking path.

2. Availability as signal, not barrier

Instead of disabling dates, I surfaced availability states while preserving interaction.

Never fabricate scarcity, only surface real uncertainty. In a distributed marketplace, availability is probabilistic.

Interactive prototypes

I built high-fidelity interactive prototypes for both desktop and mobile to validate the flow in stakeholder reviews.

Desktop

Product detail page, hero and gallery
Product detail page, details and booking widget
Calendar date picker
Booking options, accommodation tiers

Mobile

Beyond screens: designing the data model

The UI was only the visible layer. These are the deliverables I owned end to end.

01

Data audit & schema

Audited the full catalog, mapped how 9 different concepts were collapsed into a single field, and proposed the normalized schema that makes cross-provider comparison possible.

02

End-to-end booking flow

Designed the full booking flow for desktop and mobile: simplified Date + People search, real-time provider comparison, and optimal-option highlighting.

03

Prototypes

Built high-fidelity interactive Figma prototypes for desktop and mobile, the artifacts used to validate the flow in stakeholder reviews.

04

Scoring criteria

Defined when one provider's option outranks another's, the logic behind the "optimal option", and shaped the phased rollout strategy with PM and engineering.

Measured impact

Validated through A/B testing on the live funnel, the interaction model and availability pattern produced significant lifts across the key conversion metrics:

+10.31% Conversion rate 99% statistical significance
+33.45% Add to cart 99% statistical significance
-14.19% Bounce rate 99% statistical significance

What this work enabled

While the full marketplace rollout is part of the long-term roadmap, this phase produced three concrete deliverables:

01

Data standardization roadmap

The audit and taxonomy engineering is now using to normalize the catalog's data model, findings that elevated data quality from a "nice to have" to a roadmap prerequisite.

02

Validated interaction model

The simplified Date + People input and multi-provider comparison pattern, validated through prototyping and stakeholder reviews.

03

Uncertain-availability pattern

A calendar pattern where every date remains selectable and availability is surfaced as a signal, not a barrier, reflecting real uncertainty instead of fabricating scarcity.

The best match starts
with a conversation.