Skip to content
Voraal

Guide · Engineering

How to build a travel booking portal: the architecture, the hard parts and a phased delivery plan

A travel booking portal looks like a search box and a checkout. Underneath, it is a distributed system that queries slow external suppliers, prices against volatile fares, takes money before inventory is confirmed and then has to service the booking for months. This guide is written for product and engineering leads deciding how to build one.

Built for

  • CTOs and product leads
  • OTA founders
  • Banks and wallets adding travel
  • Agencies replacing legacy portals

Short answer

How do you build a travel booking portal?

Define the products and channels first, then assemble five layers: supplier connectivity (GDS, NDC, LCC, hotel APIs), a normalisation and pricing engine, booking and payment orchestration, post-booking servicing, and the front end. Most teams buy the connectivity layer through a travel API or licensed platform and build only the customer experience. Voraal supports either route with Avunzo or Rezencia.

By the Voraal team · Updated · 7 min read

What should you decide before writing any code?

Who sells what to whom, and who issues the ticket. A consumer flight portal, a B2B agent portal and a corporate booking tool share components but differ in pricing, payment and permissions. Equally important is the merchant-of-record question: if you are not IATA-accredited, someone else issues tickets and settles with airlines, and your portal must integrate with their ticketing and settlement, not just their search.

Write a one-page scope: products (flights, hotels, transfers, packages), markets and currencies, channels (web, app, B2B), payment methods, and the after-sales actions customers or agents can trigger themselves. Every later architecture decision traces back to this page.

What are the layers of a travel booking portal?

A reference architecture with the typical build-or-buy call for each layer.

LayerResponsibilitiesTypical decisionWhy
Supplier connectivityGDS, NDC, LCC and hotel API sessions, certification, version upgradesBuy via a travel API or platformEach connection needs contracts, certification and constant maintenance
NormalisationOne data model for offers, fares, rooms, baggage, rulesBuy, extend where neededSupplier formats differ widely; mapping is where bugs hide
Pricing engineMarkups, discounts, service fees, currency conversion, roundingConfigure on platform or buildYour commercial logic; must be explainable per booking
Booking orchestrationRe-price, hold, pay, confirm, ticket; failure handlingBuy core, own the flowPartial failures (paid but not ticketed) need careful state handling
ServicingChanges, voids, refunds, reissues, schedule changes, EMDsBuyFare-rule and supplier-specific logic, high error cost
FinanceInvoices, ledgers, reconciliation, FXBuy or integrateMust match supplier statements
Front end and accountsSearch UX, checkout, account area, content, SEOBuildYour differentiation

Why is supplier connectivity the hardest part?

Because every supplier behaves differently and none of them is fast. GDS interfaces carry decades of conventions; NDC airlines each implement their own subset of the IATA schemas and versions; low-cost carriers expose proprietary APIs that change with their product; hotel suppliers differ on cancellation policies, room mapping and rate types. Each connection also needs a commercial agreement and, for GDS and many NDC airlines, a certification process before you can go live.

This is why most new portals start with a multi-supplier travel API that has already done the connecting, mapping and certification. You can still add a direct connection later for a supplier that matters to your margin. If you do build connections yourself, budget for the ongoing work, not only the first integration: schema upgrades, airline-specific quirks and certification renewals never stop.

How do you make search fast enough?

By combining parallel supplier calls, short-lived caching and progressive results. A single flight search may fan out to several sources that each take seconds to respond. Call them in parallel with per-supplier timeouts, stream the first results to the screen as they arrive, and de-duplicate identical itineraries from different sources by keeping the best price and noting the source.

Caching needs discipline. Cache shopping results for minutes, not hours, and always re-price before taking payment. Watch your look-to-book ratio: some suppliers and aggregators charge for searches beyond an agreed ratio — Duffel, for example, publishes an excess-search fee above a 1,500:1 search-to-book ratio — so caching and sensible search throttling are also a cost control.

How should the booking transaction be orchestrated?

The booking path must survive partial failures. A safe sequence for a paid flight booking:

  1. 01

    Re-price the selected offer

    Confirm the fare and availability with the supplier immediately before checkout. If the price changed, show the new price and ask the customer to accept.

  2. 02

    Create the booking or hold

    Create the PNR or order with passenger data. Record the supplier reference and the ticketing time limit in your own database.

  3. 03

    Take payment with an idempotency key

    Authorise payment once, using a key that prevents double charging if the customer retries or the network drops.

  4. 04

    Issue or confirm

    Ticket the PNR or confirm the order. If issuance fails, the booking enters a 'paid, not ticketed' state that alerts the operations desk immediately.

  5. 05

    Deliver documents and post finance

    Send e-tickets or vouchers, raise the invoice and post ledger entries in the same transaction as the status change.

  6. 06

    Reconcile asynchronously

    A background job compares your bookings with supplier and payment records and flags mismatches for a person to resolve.

Should you build from scratch, license source code or use white-label?

Choose by how much differentiation you need and how much engineering you can sustain. A white-label travel booking engine is fastest: your brand on an existing engine, with configuration rather than code. Licensing travel booking software source code suits organisations that need data residency or deep customisation and have an in-house team to maintain it. Building a custom front end on a travel API gives you full control of the experience while the provider runs connectivity and servicing. Building everything from scratch is justified only when travel technology is itself your product and you can fund a permanent supplier-integration team.

What does a phased delivery plan look like?

PhaseScopeExit criteria
0. DiscoveryScope page, supplier and ticketing agreements, payment provider, compliance reviewSigned content and payment contracts; sandbox access
1. MVPOne product (often flights), one currency, web checkout, e-ticket delivery, manual servicingEnd-to-end live bookings with real money; reconciliation matches
2. ServicingSelf-service cancellations, automated voids and refunds, schedule-change noticesSupport contacts per booking falling month on month
3. BreadthHotels, transfers, packages, more payment methods, mobile appAttach rate of second product to flights
4. ChannelsB2B agent portal or white-label sites on the same engineAgent onboarding without code changes

Durations depend heavily on supplier contracting and certification, which often take longer than the software.

How should you test a travel portal before launch?

Test the unhappy paths more than the happy one. Build a test matrix covering price changes at re-price, supplier timeouts during booking, payment success with ticketing failure, duplicate submissions, name-length and special-character limits, infant and child pricing, multi-city itineraries, and cancellations on non-refundable fares. Supplier sandboxes rarely behave exactly like production, so plan a controlled live pilot with real low-value bookings and a staff member ready to fix anything by hand.

Load-test the search path with realistic fan-out to suppliers, and agree rate limits with each supplier before you run the test, or you may get blocked for a denial-of-service pattern. The Voraal travel API integration checklist lists the controls we check before go-live.

How Voraal fits

Voraal covers the layers most teams prefer not to build. Avunzo provides a REST/JSON API for flights from Amadeus, Sabre, Travelport, NDC airlines and low-cost carriers, plus hotels, transfers, car rental, activities and dynamic packages, with a sandbox, OpenAPI specification, webhooks, versioning and scoped credentials. Rezencia provides the full platform — booking engine, white-label front end, B2B portals, ticketing automation and finance — as cloud SaaS or licensed source code. For a custom build, the travel portal development page explains how projects are scoped.

Key takeaways

  • Write the scope page first: products, channels, currencies, payments and who issues tickets.
  • Buy supplier connectivity and servicing unless travel tech is your core product.
  • Search in parallel with timeouts, stream results and cache for minutes, not hours.
  • Always re-price before payment and use idempotency keys on payment.
  • Design an explicit 'paid, not ticketed' state with alerts.
  • Keep card data out of your systems with hosted payment fields.
  • Test failures and edge cases more than the happy path.
  • Phase delivery: MVP, servicing, breadth, then channels.

FAQ

Frequently asked questions

Questions buyers ask about how to build a travel booking portal.

Ask us anything
  • How long does it take to build a travel booking portal?

    A front end on an existing travel API or white-label engine can reach a live MVP far faster than a from-scratch build, because supplier contracting and certification are already done. From-scratch builds are usually dominated by supplier onboarding, which can take months per source.

  • Which programming stack is best for a travel portal?

    The stack matters less than the architecture. Choose what your team can maintain, with good support for concurrent outbound calls, background jobs and strong typing of supplier data. The supplier-integration layer is the part where language choice has the most impact.

  • Do I need my own GDS contract to build a travel portal?

    Not if you connect through a travel API provider or consolidator that holds the GDS agreements. You need your own contract when you want direct pricing, your own office IDs and full control, which usually follows IATA accreditation.

  • What is a look-to-book ratio and why does it matter for a portal?

    It is the number of searches per completed booking. Suppliers and aggregators may limit or charge for excessive searching, and every search costs compute and latency. Caching and sensible search design keep the ratio healthy.

  • How do portals handle prices that change between search and booking?

    They re-price the selected offer immediately before payment. If the price has moved, the customer sees the new price and must accept it. Taking payment on a stale cached price is a common cause of losses and complaints.

  • Can the same portal serve consumers and agents?

    Yes, if pricing, payment and permissions are channel-aware from the start. Retrofitting B2B accounts, credit and agent markups onto a consumer-only design is expensive, so decide early.

Sources and further reading

Regulations, airline programmes and supplier terms change. Confirm requirements with the relevant authority or supplier before acting. Contact [email protected] if you spot something out of date.

Let’s build what comes next.

Tell us how your travel business works. We’ll recommend the right products and walk you through them on your own use case.

WhatsApp