ABM Hotels · Data Engineering & ML

We taught a closed hotel system to speak cloud.

A bespoke operational platform connecting two hotels, an on-premise property-management system, a cloud backend and applied machine learning.

2Hotel properties

5 minData synchronisation

130,897Historical stays analysed

25Report fields reconstructed

3.45×Return concentration

66.6%Future revenue identified

The challenge

Critical operations, locked inside a system that was never designed for the cloud.

Two properties operated independently. Revenue, rooms, guests and check-ins lived inside proprietary software with no external API and no unified operational view.

We mapped the relevant data model and built a secure integration layer around it—without interrupting day-to-day hotel operations or exposing the source system publicly.

The transformation

From isolated property systems to one operational platform.

Before → Integration → After

BeforeDisconnected

▣

ExpressProperty system

RoomsDB-1

BookingsDB-2

GuestsDB-3

RevenueDB-4

▣

InternationalProperty system

RoomsDB-1

BookingsDB-2

GuestsDB-3

RevenueDB-4

×No shared view

×Desktop-bound

×Manual comparison

×No external API

Data routingIntegrationEvery 5 minutes

Secure·Normalised·Reconciled

AfterUnified

◇

Cloud operational modelOne trusted application layer

Dashboard

Room status

Reports

Audit trail

Revenue

Guest intelligence

✓One operational view

✓Cloud accessible

✓Auto-reconciled

✓Analytics ready

The architecture

A private edge-to-cloud data platform.

Hotel systems remain the source of truth. A private integration worker reconciles operational changes into a cloud model designed for fast, reliable application access.

At the hotelsProperty systemsTwo live operations

Private edgeIntegration workerSecure ingestion

Cloud dataOperational modelReconciled state

ApplicationBackend APIsRoles & workflows

Decision layerOperations + MLActionable intelligence

Eventually consistentIncremental syncBackdated reconciliationHealth monitored

Designed for operational reality

The past can change.

Hotels can post charges against an earlier business date. The platform revisits recent history so reporting converges instead of preserving an incomplete snapshot.

IncrementalIdempotentAuditable

Revenue completenessConverged

Live dayCurrent postings

76%

Night auditLate entries arrive

92%

ReconciledHistorical refresh

100%

The platform

Not an off-the-shelf dashboard. A system designed around how the hotels operate.

Private PMS integration

A secure integration layer that reads two on-premise hotel systems without exposing them to the public internet.

Cloud operational model

Legacy records transformed into a consistent, read-optimised data model built for modern applications.

Backend platform

Authenticated APIs, role-based access, property isolation and operational workflows delivered as a production service.

Live hotel operations

Room state, booking changes, revenue, audit events and notifications available across both properties.

Reporting engine

Daily, weekly, monthly, PDF and Flash-style reporting reconciled against real hotel operations.

Reliability layer

Health checks, freshness monitoring, failure isolation, regression testing and query optimisation.

Operational room board

Every room, one current state.

A unified live view derived from the hotel's operational source of truth.

Occupied44

Vacant20

Dirty6

Maintenance4

Blocked2

76 rooms mapped · representative state5-minute refresh

The intelligence layer

Once the data was trustworthy, we asked what it could predict.

41,766 completed stays. 30,468 anonymised guest profiles. Models validated against later, unseen behaviour across both properties.

Guest retention

3.45×normal return rate

The model's highest-ranked guests returned at roughly 69%, compared with a 20% portfolio baseline.

Baseline · 20/100

Model selected · 69/100

Future guest value

66.6%of future revenue

The highest-ranked 10% of guests represented approximately two-thirds of future room revenue.

TOP 10%

Guest portfolioRevenue concentration

Explainable RFM score80.1% ranking score

Validation, not presentation theatre

Tested on behaviour the model had never seen.

Guests were ranked using information available at the prediction date, then evaluated against later stays. The highest-ranked 10% captured 66.6% of future room revenue.

81.1%Validation AUC

10%Priority cohort

Future room revenue captured by the top 10% of guests

100%75%50%25%0%Future revenue captured

10%

Random selection

66.6%

Model-ranked

Random selection benchmark: approximately 10% of future revenue.

The impact

From operational records to operational intelligence.

↗One operational layer across two hotels

↗Five-minute edge-to-cloud visibility

↗Historical corrections reconciled automatically

↗A validated foundation for guest intelligence

Lamax led systems architecture, PMS discovery, data engineering, backend development, reporting, reliability engineering and independent analytics research. The application was delivered with an implementation partner responsible for frontend and partner modules.