Maya Travel AI

Maya is a hosted, white-label AI agent that travel companies embed on their own site and messaging channels. It is not a new company's first product: PhocusWire's August 2023 AI Insights column has Joris Vanherp, CEO of Belgian travel content company Live the World, describing Maya as the company's own consumer travel assistant, with current work focused on "the development of a white-label software-as-a-service version of our AI travel assistant Maya" (phocuswire.com, 7 Aug 2023). Live the World runs the consumer sites livetheworld.com and itinari.com.

That lineage is the most useful thing to understand about the product. Maya is a travel-operator tool built by people who ran a consumer travel content platform first, which is a different starting point from a horizontal support bot pointed at a travel customer. It also explains the shared corporate paperwork covered further down.

The same PhocusWire interview is candid about the hard part in a way the marketing site is not. Asked about the biggest challenge, Vanherp answered: "the randomness of the results... there is no single correct answer anymore that you can easily compare the result to while testing." That is an accurate description of the problem and worth holding alongside the site's claim that answers are "never hallucinated" with "99%+ answer relevance" — a figure published with no stated methodology. A sensible buyer asks how relevance is measured and over what sample.

Vendor logo walls usually can't be verified. This one largely can, and it is the strongest thing in Maya's favour.

Maya's widget loads from a fixed mount point — a div with id maya-b2b-full-page and a data-organization-slug attribute — which makes deployments identifiable from a customer's own served HTML. We found it live on two named customers today: zoover.nl carries a MayaChatLoader component with data-organization-slug="zoover", and a microcopy key setting the chat title to "Vera" — matching Maya's claim that Zoover's agent is branded Vera. bestarctic.com, a WordPress site, carries the same mount with data-organization-slug="bestarctic", under an HTML comment reading "Maya Chatbot - Popover".

Neckermann, the largest brand on the list, corroborates independently in plain language on its own Dutch homepage: "Via onze chat kom je 24/7 in contact met onze AI-chatbot Maya en kan je tijdens onze openingsuren ook rechtstreeks in contact komen met een Travel Specialist" — you reach the AI agent Maya around the clock, and a human Travel Specialist during opening hours. That sentence also confirms the intended operating model: AI always on, humans in business hours.

We did not find the widget on travelbase.nl. That is not a contradiction — Travelbase's claimed use is email drafting rather than a site widget — but it is unconfirmed rather than confirmed.

One inconsistency: the homepage FAQ says "35+ travel companies across 20 countries" while the site's own llms.txt, last updated 7 July 2026, says "30+". PhocusWire reported "more than 20 companies" in February 2025. The direction of travel is consistent even if the current number isn't stated consistently.

Maya leads with large conversion figures, and the site contains enough detail to work out what they mean — which is more than most vendors provide.

The headline is "up to 5x". The Zoover case study states it precisely: "Travelers who interact with Vera convert at 3-5x higher rates than those who don't." That is a comparison between travellers who chose to open a chat assistant and travellers who didn't. People who engage are, by construction, further along in buying intent, so this figure mixes the agent's effect with the effect of already wanting to book. It is a real and reportable operating statistic; it is not a measure of causal lift.

The number to plan against is elsewhere on the same homepage: Maya states that "Neckermann measured a +6% conversion lift in an A/B test". An A/B test is the design that isolates the tool's contribution, and +6% on a large OTA's booking volume is a commercially meaningful result. Maya deserves credit for publishing it; a buyer should build the business case on it rather than on the 3-5x.

The Travelbase figures — "up to 30% increase in bookings", "up to 82% of inbound email replies drafted by AI" with review taking roughly 30 seconds against 4+ minutes manually, and "estimated cost savings of €18,000/month" — appear in Maya's blog post on AI agent adoption, not on the Travelbase case study page, which is entirely qualitative. Note the vendor's own hedges: "up to", "attributed to", "estimated". The email-drafting ratio is the most concretely checkable of these in a pilot.

This is a hosted, vendor-operated SaaS, not something you run. The public loader at maya-b2b.s3.eu-central-1.amazonaws.com/loaders/prod.js is a small script that waits for the maya-b2b-full-page element and then injects a React bundle from CloudFront; the customer's tenancy is identified by a data-organization-key attribute. Integration on the customer side is a script tag and a div. AWS eu-central-1 is Frankfurt, consistent with Germany appearing in the vendor's processing locations.

Maya's FAQ puts implementation at 4 to 6 weeks, covering training on the client's website content, product catalogue and FAQ; channel configuration; branding the agent with its own name and persona; then feedback rounds and operational setup. The case studies bear out that this is a configured-for-you engagement rather than self-serve — Zoover's agent is "Vera", Best Arctic's is "Maria", Travelbase's is "Atlas", each with its own persona.

Channels claimed: website chat, WhatsApp, Instagram, Facebook Messenger, Gmail and Outlook, with handover into Slack and Microsoft Teams. Named integrations: HubSpot, Salesforce Marketing Cloud, Zendesk, Zoho CRM, Travelspirit, Mailchimp and Front. There is no public API documentation — /docs returns 404 — so the integration surface beyond that list is not established. If you need a specific reservation system connected, treat it as scoped project work and get it in writing.

Budget for the calendar as much as the licence: 4-6 weeks of vendor-led onboarding plus your own content preparation is the real cost of entry, and it means this is not a tool you trial in an afternoon.

There is no pricing page. /pricing returns 404, no pricing URL appears in the sitemap, and the only occurrences of the word "pricing" on the homepage refer to the product answering travellers' questions about trip prices. Every route to a number is a booked call — a demo via TidyCal or a "free AI consulting session".

That is ordinary for a vendor selling 4-6 week managed deployments to mid-market travel companies, and it is not a criticism. It does mean you cannot size this against alternatives without entering a sales process, and you should expect pricing to be quoted per deployment rather than from a rate card. Given the delivery model, ask specifically whether the fee scales with conversation volume, channels, languages or brands — a multi-brand operator like Travelbase and a single-country tour operator are very different shapes of contract, and nothing public indicates which axis the price moves on.

Maya's FAQ states it is "built in Belgium in accordance with GDPR, CCPA/CPRA, and the Data Protection Act", supports compliant WhatsApp opt-ins, and keeps conversation data isolated per customer. For an EU-headquartered vendor selling to EU travel companies that is a reasonable baseline claim.

The supporting documentation is thinner than the claim. The privacy policy is the only legal or compliance document published: /security, /dpa, /terms and /trust all return 404, and there is no public subprocessor list, no trust centre and no named certification such as SOC 2 or ISO 27001. Whether Maya holds any is not established either way.

The policy itself is a single document covering livetheworld.com, app.livetheworld.com and mayatravel.ai together — consistent with the shared corporate origin — and it routes data access, erasure and portability requests to [email protected]. It is stated as effective 1 January 2021, which predates the Maya product. Three consequences a buyer's data protection officer will care about: it names no LLM or AI subprocessor and takes no stated position on whether customer or traveller data is used to train models; it states that personal information "is stored and processed in Belgium, Germany, United States and United Kingdom", so US processing is in scope despite the Belgian framing; and being a pre-AI general policy, it was not written for a product that pipes traveller conversations through a model.

None of this is unusual for a company of this size, and a private DPA and subprocessor list may well exist. But for any tool reading customer conversations, get the model provider named, the training position in writing, and a DPA covering the US transfer before a security review — none of it can be settled from public sources today.

PhocusWire reported on 27 February 2025 that Maya raised €1 million, with investment from "key executives from Lighthouse", to be used for global growth and to grow its sales and engineering teams. The article names Joris Vanherp as CEO and co-founder and Benjamin Manzi as chief commercial officer and co-founder. That article stated "more than 20 companies" were then using Maya. We found no public record of a subsequent round, so as of today the most recent disclosed funding is roughly eighteen months old.

Note that a widely-repeated "founded 2023 in Antwerp" detail does not appear in the PhocusWire article text and we could not source it to a primary record; treat the founding year and city as not established, though the product clearly existed in some form by mid-2023.

The team page lists 14 people, including one advisor, one summer intern and one freelance marketer — so roughly eleven core staff, with a named CTO (Ali Sherazi), a product and AI lead, an AI engineer and a data engineer. That is a small company, but a genuinely staffed one with in-house engineering rather than a two-person reseller.

What this means practically: Maya is past the prototype stage and has production deployments at recognisable brands, but it is a company of about a dozen people on a €1M raise serving customers in 20 countries. If you are a large operator, weigh the support model and ask about escalation cover and roadmap commitments. There are no independent user reviews on G2 or Capterra to triangulate service quality against, which is normal at this size but does mean references from existing customers are the only real check available — and given the deployments verified above, those references exist and are worth asking for by name.

It fits best if you are a tour operator, OTA, DMC or travel agency with a real catalogue and a real content library, enough enquiry volume that repetitive questions are consuming expert time, and a multilingual audience. The verified deployments cluster exactly there: a Dutch review platform with millions of reviews, a Norwegian Arctic operator with 100+ tours, a Belgian mass-market OTA. Maya's pitch depends on having proprietary content to ground answers in; if your site is thin, the grounding advantage largely disappears.

It fits poorly if you want self-serve deployment this week, need published pricing to build a business case before talking to sales, require a signed DPA and named subprocessors up front from public documentation, or want an API-first component to build your own experience on — there is no public API documentation and the delivery model is a managed widget.

If you are evaluating it, the highest-value things to request are: the Neckermann A/B methodology, a named customer reference on your channel mix, the model provider and training position, and a clear statement of what the price scales with.