01 / Problem
An independent hotel pays commission on every booking that comes through an OTA such as Booking or Expedia, and many of those bookings start with a question its website did not answer in time. Building an assistant that answers and sends guests to the booking engine is technically manageable. The expensive part is building it and then finding out hotels will not pay for it.
02 / Context
Context
The product hypothesis has four parts: the hotel loses direct bookings to unanswered questions, an assistant that answers and hands off to the booking engine wins some of them back, extras offered at the right moment add revenue, and the hotel pays if it can see the attribution. Phase 1 exists to test whether that holds up with real hotels, not to prove the software works.
03 / Solution
How we solved it
Phase 1 is deliberately small: an indexable landing page, a revenue calculator on its own page and a form that stores the lead together with the calculation behind it. There is no chatbot, no login, no integrations and no payments. What there is, is rigour in what gets measured: the calculation lives in versioned pure functions, the server always repeats it instead of trusting the browser, and analytics carry no personal data.
Flow
- 01
Calculation engine in pure functions: no state, no network, tested in isolation
- 02
Strict limits on every input: anything out of range is rejected, not clipped
- 03
The endpoint recalculates on the server and stores the formula version with the lead
- 04
Calculator set to noindex with its own canonical so it does not compete with the landing page
- 05
Funnel events in the dataLayer with a closed list of parameters
Workstreams
Validate before building
The Phase 1 question is commercial, not technical: whether a hotel will accept a pilot and what price it considers reasonable. The software is sized to answer that and nothing more.
- No published price until there have been conversations
- Commercial hypothesis kept in code, not shown
- Lead status tracked in the database through to pilot
A calculation that can be audited
If the formula changes, the version changes. That way a lead from today and one from three months on can be compared, knowing which numbers each was calculated with.
- Formulas, limits and scenarios in a single module
- Amounts in whole cents to avoid rounding errors
- The server never accepts the result sent by the browser
An estimate that says what it is not
The page states that the savings are gross, that there is no seasonality and that the scenario percentages are working hypotheses, not industry averages.
- Indicative-estimate notice always visible
- Model limitations explained next to the result
04 / Scope
What it does
- Explains the offer on a landing page built for searches about chatbots for hotels
- Estimates commission savings and revenue from extras using the hotel's own data
- Shows three scenarios (conservative, realistic and potential) and states their limitations
- Collects pilot requests with versioned consent and an internal email alert
- Measures the whole funnel without sending personal data to analytics
05 / Impact
What it solves
The product moves from idea to measurable experiment. Each lead arrives with the hotel's data, the chosen scenario and the formula version, so the diagnostic call is about concrete numbers and can ask directly whether the hotel would accept a pilot and at what price. Phase 2, the working assistant, only gets built if those answers justify it.
Security and confidentiality
- The leads table is only accessible with the server's service key; the browser never touches it.
- Logs are structured and contain no personal data.
- The form consent is versioned: it is known which text each hotel accepted.
06 / Questions
Frequently asked questions
Why is there no chatbot yet?
Because the first thing to prove is that there is a hotel willing to pay for it. Building the assistant, the booking-engine integrations and the attribution takes weeks. Measuring interest takes days. If Phase 1 does not land pilots, that time has been saved.
Where do the scenario percentages come from?
They are working hypotheses, and the page says so. They are not industry averages or client results. They let the hotel see a range based on its own data, not a promised figure.
Why is the calculator not indexed?
The same tool sits inside the landing page. Two indexable URLs with the same functional content would split the search traffic. The separate page serves as a destination for ads and direct links.