Auditable decisioning, human control, measured outcome.

68-element rule and control catalogue

The RateMind™ decision system consists of three separated groups of elements. They are not all price-changing rules: controls hold back or block a recommendation, and measurement rules only measure.

• 24 price signals: OTB load, pickup, historical pace, days to arrival, booking window, room-type price position, guest-rate exposure and hotel-specific demand patterns.
• 40 data-quality, decision and safety controls: source freshness, room-type mapping, contradictory signals, minimum and maximum rate, daily movement limit.
• 4 measurement rules: evaluating the post-decision outcome without modifying the original reasoning.

Hotel DNA

The system learns from the hotel's own booking history how individual weekdays, seasons, arrival horizons and room types perform. It does not force a generic hotel pattern on the property.

• Weekday × lead time behaviour
• Seasonal position and season transitions
• Booking window length per segment
• Room-type hierarchy and rate steps
• With a small sample: minimum data, confidence threshold and safe rate band

Individual apartment strategies, one portfolio

The apartment engine maintains a pricing policy for each unit. HUF, EUR and USD properties can be calculated in their native currency. Season, weekday and date-specific rates shape the starting rate in an explicit order.

• Individual base rates, floors, ceilings, rounding and daily movement limits.
• Seasonal and weekday factors, with date-specific rates taking precedence.
• Versioned strategic rate plans, traceable reviews and apartment-level calendar results.

Economics of the complete stay

For stays of 1–28 nights, the engine evaluates the applicable, configured discount combinations. It calculates a cost-based rate floor from the contribution remaining after commissions, payment fees, cleaning fee income and operating costs.

• Length-of-stay discounts apply only to eligible stay lengths.
• Protection includes the configured combination that produces the lowest contribution.
• Contribution is a calculated operating amount based on the recorded revenue and costs.

Bookability and calendar protection

The engine checks a rate recommendation against the stays that can actually be sold. The review covers the entire stay and the surrounding calendar dates.

• Minimum and maximum stays, arrival and departure restrictions.
• Detection of leftover gaps that would be too short to sell.
• Specific review flags when calendar or cost data is missing or contradictory.

Events, holidays and weather context

Alongside event locations, timing and relevance, the system maintains date-specific environmental context. Hungarian public holidays, rescheduled working days and archived weather forecasts help explain the situation.

• Forecast source, issue time and freshness remain traceable.
• Information available at the decision time stays separate from later observations.
• Holiday and weather information provides context; it does not independently change a rate.

Pace Navigator and demand velocity

Pace is not a single number. Velynto measures a date's booking velocity against the hotel's own historical curve and flags when the current pace is persistently accelerating or slowing.

• Daily pickup and cumulative OTB by days to arrival
• Historical pace curve for the same weekday and season
• Confidence level tied to the deviation

Promotion-aware guest price

The shelf rate and the rate a guest pays differ. Velynto also examines the effective guest rate expected after active Booking.com promotions, Genius levels, rate-plan scopes, suspensions and stacking rules. This is an analytical capability: the sources involved always depend on the setup.

• Minimum and maximum guest-rate exposure
• Verified discount stacking
• Rate-floor and value-add warning

Evidence gate and rate candidates

Twelve evidence checks control whether a recommendation may proceed and how high its confidence can be. Four rate candidates are then evaluated within the hotel's own safety limits.

• Hold, conservative, balanced and assertive options
• Applicable guardrails and reasoning attached to every candidate
• Held-back or blocked recommendation with incomplete or contradictory data

Immutable decision passport

At decision time we record what the system knew: OTB, days to arrival, pickup, current and recommended rate, promotion exposure, inventory, event and market signals, active and contradictory rules, confidence level. It cannot be rewritten later.

• A later outcome cannot leak back into the original reasoning
• Important decisions remain auditable afterwards
• Approval, modification and rejection are logged

Application verification

An approved recommendation does not end with the decision. A later source confirms that the applied rate actually appeared and that the guest rate sits in the expected band.

• Applied rate compared with promotion exposure
• Warning on deviation
• Velynto does not publish hotel rates on its own

Outcome measurement

After the decision we measure impact at D1, D3, D7, D14 and at the final result. Measurement happens without modifying the original evidence, and we do not state causal revenue impact.

• Pickup, bookings, revenue, ADR, occupancy, RevPAR and cancellations
• Final result after the stay closes
• Measurement is attached to the decision log, not a rewrite of it

Decision Replay

Decision Replay reproduces the decision from the original evidence and flags any difference in rate, rule, modifier or confidence, so changes in decision logic stay traceable.

Forecast backtesting and shadow models

New forecasting and strategy models are tested in chronological validation, isolated from live pricing. There is no automatic model deployment and no automatic rate change.

• Decision-time features and verified final outcomes only
• Forecast challenger: comparison on MAE, WAPE and bias
• Strategy ranker: only actually selected and measured strategies are evaluated
• The production recommendation is still produced by the auditable rule engine