Vendor Proposal
Confidential
June 2026

Snowflake — Cloud Data Platform
Proposal

A unified analytics and AI data platform for a digital real estate business operating in the EU, enabling Business Intelligence, machine learning, and AI-driven analytics at scale.

100%
Requirements addressed
EU
Data residency (Azure West Europe / AWS Frankfurt)
12
Data source integrations supported natively or via connector
8 wk
Phase 1 go-live target
Module 1 — Architecture & Solution Design
Module 2 — Full Implementation
1.1

Executive Summary

Snowflake is proposed as the unified analytics and AI data platform, deployed on Microsoft Azure West Europe (Amsterdam) with failover capability to AWS eu-central-1 (Frankfurt). This deployment satisfies all EU data residency and GDPR requirements while providing a clear multi-cloud migration path as your infrastructure evolves toward AWS and Cloudflare.

The architecture is read-oriented by design — Snowflake consumes replicated data from operational systems and serves analytics, reporting, AI, and ML workloads without touching production databases. All data transformation, governance, and AI capabilities run within the Snowflake boundary; no proprietary vendor services outside Snowflake are required for the core platform.

Key differentiator for this engagement: At 150 GB–1 TB data volume and <200 concurrent users, this is a lightweight deployment that runs on Snowflake Standard tier. All advanced AI features (Cortex Analyst, Cortex Search, LLM functions) are available from Standard tier in EU regions. The full platform can be running in production within 8 weeks.

1.2

Reference Architecture

End-to-End Platform Architecture
Source Systems (existing — not modified)
Azure SQL Server (REP)
PostgreSQL (secondary app DB)
Azure Event Hubs (portal events)
MS Dynamics 365 (ERP)
GA4 / Google Ads / Meta Ads / TikTok Ads
Jira / Confluence / Trello / Azure DevOps
Microsoft Clarity (product analytics)
↓   Fivetran managed connectors + Azure Event Grid + Snowpipe
🥉 Raw / Bronze Layer — raw_db — Append-only, full fidelity, never modified
raw.rep.portal_events
raw.rep.listings
raw.crm.contacts
raw.erp.transactions
raw.marketing.ga4_sessions
raw.marketing.ad_spend
raw.sdlc.jira_issues
raw.docs.property_documents
↓   Streams + Tasks (Events)  |  dbt incremental models (Batch)  |  Snowpark (complex transforms)
🥈 Silver / Conformed Layer — silver_db — Cleaned, typed, deduplicated, joined
silver.fact_portal_events
silver.fact_listing_activity
silver.dim_property
silver.dim_user
silver.dim_partner
silver.dim_date
silver.dim_geography
silver.fact_ad_performance
↓   dbt table/materialized view models  |  Cortex Search indexing (document embeddings)
🥇 Gold / Analytics Layer — analytics_db — Business-ready, BI-served, AI-accessible
gold.listing_performance_weekly
gold.user_funnel_daily
gold.partner_activity_summary
gold.marketing_roi
gold.property_valuation_model
gold.pipeline_health
↑   JDBC/ODBC / Snowflake Connector / REST API / Cortex Analyst
Consumption Layer — BI, API, AI Agents, Self-Service
Power BI (DirectQuery)
Tableau / Metabase
Cortex Analyst (NL→SQL)
Cortex Search (Document RAG)
Python / Snowpark ML
REST API (partner portals)
Snowflake Streamlit apps

Compute Design

WarehouseSizeAuto-suspendPurposeEst. Credits/month
ETL_WHM300sFivetran/Airbyte loads, Snowpipe, dbt pipeline runs, Streams+Tasks~120 cr
BI_WHS (multi-cluster max=2)60sPower BI DirectQuery, Tableau, shared dashboard queries~200 cr
ADHOC_WHXS60sAnalyst exploration, Snowsight, ad-hoc SQL~60 cr
DS_WHL600sSnowpark ML, feature engineering, training jobs~80 cr
API_WHXS60sLow-latency REST API queries (partner portal, sub-second SLA)~40 cr

Total estimated compute: ~500 credits/month at standard utilisation. At ~€3.00/credit (Azure West Europe Standard tier): approximately €1,500/month. Storage for 1 TB: ~€25/month. Fivetran connector fee is separate (typically €400–800/month for this connector count).

1.3

Functional Requirements Coverage

4.1 — Data Platform Capabilities

RequirementSnowflake ResponseStatus
Structured data (relational, transactional)Native — Snowflake is fundamentally a columnar relational warehouse. Full SQL, ACID transactions, JDBC/ODBC.Native
Semi-structured / unstructured (JSON, events, logs)Native VARIANT type stores arbitrary JSON. value:field::type dot-notation queries JSON inline. No schema-on-write required for raw events.Native
Multidimensional modelling (time, geography, property type)Native — Star/Snowflake schema, hierarchies via dbt, GROUPING SETS / ROLLUP / CUBE for OLAP aggregations.Native
Standard SQL + JDBC/ODBCANSI SQL compliant. JDBC/ODBC drivers available for all major BI tools. Python connector, .NET connector, Node.js connector.Native
Vector database / semantic search / RAGCortex Search — built-in vector search and hybrid search. No external vector DB required. Supports embedding of property descriptions, contracts, documents for semantic retrieval.Native

4.2 — ETL & Pipeline Capabilities

RequirementSnowflake ResponseStatus
Scheduled batch pipeline executionTasks (native cron scheduling), dbt Cloud jobs, or Airflow DAGs. Tasks run SQL/Snowpark on a schedule; WHEN STREAM_HAS_DATA prevents wasted credits when there's no new data.Native
Manual pipeline re-triggering (failure reruns)Tasks can be manually triggered via EXECUTE TASK. dbt Cloud has one-click job re-runs. Snowpipe supports REST API to manually trigger file loads.Native
Near real-time ingestion for event-driven sourcesSnowpipe (Azure Event Grid → ADLS → Snowpipe): ~1-minute latency. Kafka Connector: seconds. For portal events at 80K–250K/day, Snowpipe is the recommended approach — zero infrastructure overhead.Native
Flexible transformation / cross-source joinsdbt handles cross-database joins across the medallion layers. Snowflake supports joining across any schemas within the account. Snowpark for Python-based transformations.Native
Low overhead pipeline modificationdbt models are SQL files in Git. Changing a transformation = editing a SQL file + dbt run. No graphical pipeline editors required. Column additions via on_schema_change: sync_all_columns require no manual migration steps.Native

4.3 — Data Source Integration Matrix

SourceIntegration MethodMechanismEffort
Azure SQL Server (REP — primary DB)Managed ConnectorFivetran Azure SQL connector — CDC via log-based replication or scheduled full/incremental sync. Near real-time option available.S — 1 day setup
PostgreSQL (secondary app DB)Managed ConnectorFivetran PostgreSQL connector — CDC via logical replication or incremental sync.S — 1 day setup
Azure Event Hubs (live portal events)NativeEvent Hubs Capture → ADLS Gen2 (Avro) → Snowpipe auto-ingest. Latency ~1 minute. No Fivetran license required.S — 2 day setup
Microsoft Dynamics 365 (ERP)Managed ConnectorFivetran Dynamics 365 connector. Financial transactions, contacts, accounts synced on configurable schedule.S — 1 day setup
Azure DevOpsManaged ConnectorFivetran Azure DevOps connector. Work items, pipelines, repos, test plans.S — 1 day setup
JiraNativeFivetran Jira connector (native, Atlassian-certified). Issues, sprints, projects, custom fields.S — 1 day setup
ConfluenceManaged ConnectorFivetran Confluence connector. Page metadata, spaces, content export for search indexing.S — 1 day setup
TrelloManaged ConnectorFivetran Trello connector. Boards, cards, lists, members.XS — 0.5 day setup
Google Analytics 4NativeGA4 BigQuery export → GCS → Snowpipe. Or Fivetran GA4 connector. Session data, events, conversions. GDPR-compliant EU export.S — 1 day setup
Google Search ConsoleManaged ConnectorFivetran Google Search Console connector. Search queries, impressions, CTR, positions.XS — 0.5 day setup
Google AdsNativeFivetran Google Ads connector (official Google partnership). Campaign, ad group, keyword, conversion data.S — 1 day setup
Meta Ads (Facebook/Instagram)NativeFivetran Meta Ads connector. Campaign spend, impressions, clicks, ROAS.S — 1 day setup
TikTok AdsManaged ConnectorFivetran TikTok Ads connector. Campaign metrics, creative performance.S — 1 day setup
Microsoft ClarityCustom BuildClarity exports sessions/heatmap data via API or Azure Blob export. Custom Snowpipe ingestion pipeline. Estimated 2–3 days engineering. Replaceable if migrating to another product analytics tool.M — 3 days build

Note on Fivetran: Fivetran is recommended as the managed connector layer for batch sources. An alternative open-source path uses Airbyte (self-hosted on Azure AKS) at lower licensing cost but higher operational overhead. Both are supported and deliver the same data to Snowflake. The choice is a cost vs operations tradeoff.

4.4 — AI Capabilities

RequirementSnowflake ResponseStatus
Conversational querying (NL→data for non-technical users)Cortex Analyst — connects to a semantic model defined in YAML, translates plain language questions to SQL, returns verified results. Deployed via REST API or Snowflake Streamlit. Configurable business vocabulary (e.g. "active listing" = specific status codes).Native
AI agent support (programmatic data access for AI agents)Snowflake REST API + Cortex functions are accessible from any LLM agent framework (LangChain, CrewAI, custom agents). Cortex Analyst exposes a REST endpoint that agents can call directly. OAuth 2.0 authentication.Native
ML workloads (segmentation, recommendations, predictive models)Snowpark ML — Python ML code runs inside Snowflake compute. scikit-learn, XGBoost, LightGBM available as Snowpark ML functions. MLflow integration for experiment tracking. ML jobs run on DS_WH without interfering with BI workloads.Native
Semantic caching for AI requestsCortex has internal caching for repeated Cortex Analyst queries. For custom RAG pipelines, application-level caching (Redis / Snowflake dynamic tables as materialized cache) can be implemented. Snowflake result cache also applies — identical SQL queries return cached results in <100ms.Via result cache + app layer

4.5 — Document Management

Property contracts, legal deeds, floor plans, and inspection reports are processed via a two-stage pipeline:

📄 Document Ingestion

PDF/Word/scanned documents stored in Azure Blob Storage (existing). Snowflake External Tables reference the blob container directly. PARSE_DOCUMENT() Cortex function extracts text from PDFs. Chunked text is loaded into a staging table for embedding.

🔍 Document Intelligence (RAG)

Cortex Search indexes the chunked document text. AI agents and Cortex Analyst can retrieve relevant document passages for a given property. SNOWFLAKE.CORTEX.EXTRACT_ANSWER() answers specific questions from document content without fine-tuning.

4.6 — Reporting & BI

Power BI

Native Snowflake connector for Power BI. DirectQuery mode for live data; Import mode for better performance on complex reports. Dedicated BI_WH with sub-second query target for standard dashboards.

Third-party BI

Tableau, Metabase, Looker, Redash — all supported via JDBC/ODBC or native connectors. Any tool that speaks SQL can query Snowflake. Data sharing for external partner access without sharing credentials.

Self-Service / Non-Technical

Snowsight (Snowflake's built-in UI) for analysts. Cortex Analyst chat interface for non-technical staff. Role-appropriate views hide raw complexity. Export to CSV/Excel from any query result.

4.7 — API Layer

RequirementSnowflake Response
Well-documented REST APISnowflake SQL REST API — execute any SQL statement via HTTP POST with JSON response. Full OpenAPI documentation. SDKs for Python, Node.js, Go, Java, .NET.
Authentication (OAuth)OAuth 2.0, Key-pair JWT, Okta/Azure AD SSO, SAML 2.0. OAuth flows for partner portal integrations.
Sub-second BI query performanceDedicated API_WH (XS, auto-resume) for low-latency reads from Gold layer materialized views. Query result cache returns cached results in <100ms. Cold query on 1 TB table typically <500ms on XS warehouse for simple aggregations.
Rate limits / versioningSnowflake applies account-level rate limits. Application-level rate limiting via API gateway (Azure API Management or AWS API Gateway) recommended in front of Snowflake REST calls. SQL API is stable across Snowflake versions — no versioning breaking changes for SQL semantics.
1.4

Non-Functional Requirements

5.1 — Environments & Performance

DEV Environment

Zero-copy clone of PROD databases on demand. CREATE DATABASE dev CLONE analytics — instant, near-zero storage cost. Developers work on isolated clones without affecting production data.

STAGING Environment

Separate Snowflake databases (staging_raw_db, staging_silver_db, staging_analytics_db). dbt targets staging via a separate profile. CI/CD deploys to staging on PR merge, promotes to production on release.

PRODUCTION Environment

Full Medallion stack on Azure West Europe. Automated monitoring via Snowflake's built-in alerting and dbt test suite. Resource monitors alert at 75% monthly credit quota.

EU Data Residency: Snowflake Account deployed on Azure West Europe (Amsterdam, EU). All data storage, compute, and AI processing (Cortex) occurs within this region. No data leaves the EU. EU Azure region is ISO 27001, SOC 2, and GDPR compliant by design. Snowflake's Business Associate Agreement and DPA are available for execution.

5.2 — Monitoring, Alerting & Logging

CapabilityHow Snowflake Delivers It
Pipeline / job monitoring with failure alertingSnowflake Alerts — SQL-based alerting on task failures, stale data, row count anomalies. dbt Cloud sends Slack/email notifications on job failure. All task execution history queryable via TASK_HISTORY().
Configurable alertingSnowflake Alerts support threshold-based, schedule-based, and event-based triggers. Integration with PagerDuty, Slack, or any webhook via Snowflake External Functions.
Centralized loggingAll queries, logins, data operations logged in Snowflake's Account Usage schema. Query history, access history, login history, data transfer history — all queryable and exportable. Audit-grade immutable logs.
Full-stack security monitoringSnowflake Trust Center — continuous security posture monitoring. Unusual query patterns, privilege escalation, and data exfiltration attempts surfaced automatically. Network policy enforcement (IP allowlist).

5.3 — Development Toolset

🖥️ Snowsight

Built-in web IDE. SQL editor with autocomplete, query history, result visualisations, notebooks (SQL + Python). No separate SQL client required. Role-switching, database browser, data profiling.

📊 dbt Cloud

Visual DAG browser, IDE, scheduler, data docs auto-generation. Non-engineers can navigate the data lineage graph. Pre-built dbt packages for common sources (GA4, Jira, Dynamics) reduce build time.

🔍 Data Catalog

Snowflake's native catalog via Information Schema + Object Tagging. dbt docs generate a searchable lineage/catalog. For enterprise data cataloguing: Alation or Atlan have native Snowflake connectors.

5.4 — Testing & Quality Assurance

RequirementTool / Approach
ETL testingdbt built-in tests: not_null, unique, accepted_values, relationships. Run automatically in CI/CD on every PR. Failures block deployment to production.
Data quality monitoringdbt + Great Expectations for custom assertion logic. Snowflake DMFS (Data Metric Functions) for continuous quality checks on production tables — alert on anomalies.
Time travel testingSnowflake Time Travel — query data at past timestamps to validate transformations. Run dbt with AT(OFFSET=>-X) to test against historical snapshots.
Data preparation for testingZero-copy clone of production for test data. No data scrubbing needed for internal test environments (Dynamic Masking handles PII for partner-facing environments).

5.5 — DevOps

CapabilityImplementation
Source code / config versioningAll dbt models, Snowflake Terraform, pipeline configs in Git (GitHub / Azure DevOps). Snowflake natively integrates Git repos for Snowflake Notebooks and scripted deployments.
Infrastructure as CodeOfficial Snowflake Terraform provider — warehouses, databases, schemas, roles, pipes, tasks all defined as code. State managed remotely (Terraform Cloud or Azure Storage backend).
CI/CDGitHub Actions or Azure DevOps pipeline: PR → dbt test on staging → human approval → dbt run on production. Snowflake CLI used for DDL deployments. schemachange handles ordered SQL migrations.
SQL migration / comparisonschemachange (versioned SQL scripts, like Flyway for Snowflake). dbt's state:modified selector runs only changed models — incremental CI.

5.6 — Open Standards & Interoperability

StandardSupport LevelDetails
Apache Iceberg (open table format)Native Read/WriteCreate Iceberg tables stored in your ADLS/S3 bucket. Queryable by Spark, Flink, Athena, Trino without Snowflake. Full exit path for data.
Apache Parquet / Avro (open file formats)NativeCOPY INTO exports any table to Parquet or Avro on external stage. COPY INTO ingest from Parquet/Avro/ORC/JSON/CSV.
Apache Polaris (open catalog)NativePolaris Catalog — open-source Iceberg catalog donated by Snowflake. REST Catalog API. Compatible with any Iceberg-aware engine.
JDBC / ODBCNativeStandard drivers. Any SQL tool, BI platform, or ETL framework can connect without vendor-specific integration.
OAuth 2.0 / SAML 2.0NativeFederation with Azure AD, Okta, Google Workspace. No proprietary identity mechanism.
OpenLineage / MarquezVia dbtdbt emits OpenLineage metadata. Integrates with Marquez, DataHub, or Atlan for centralised lineage tracking across all pipelines.

5.7 — Data Governance

🔐 RBAC

Role hierarchy: ACCOUNTADMIN → SYSADMIN → TRANSFORMER → ANALYST → PARTNER. All object privileges granted to roles, never to users directly. Schema-level, table-level, and column-level grants.

🎭 Dynamic Masking

PII columns (email, phone, NI number) masked for partner-facing roles. Real values returned to internal ANALYST roles. No view proliferation — one table, governance via policy.

📋 Row Access Policies

Partner organisations see only their own listings/contacts. Internal analysts see all data. Policy defined once, applied to all relevant tables. No per-partner views or schemas.

5.8 — GDPR Compliance

GDPR RequirementSnowflake Capability
EU data residencyAzure West Europe (Amsterdam) deployment. Processing and storage never leave EU. Snowflake DPA available for signature.
Right to erasureTime Travel retention is configurable (0–90 days). DELETE statements remove from active table; Time Travel expiry handles historical copies. Fail-safe window of 7 days after which data is irrecoverably removed. Custom procedures for automated subject deletion workflows.
Right to access / portabilityQuery by user_id and export to CSV/JSON. Snowflake REST API enables automated data subject access request fulfilment. Parquet export for machine-readable portability.
Pseudonymisation / anonymisationHashing functions (SHA2, MD5) available in SQL. Tokenisation via Snowflake Dynamic Tables. Anonymisation as a dbt transformation layer for shared/partner-facing datasets.
1.5

Evaluation Criteria Self-Assessment

1. Requirements Coverage

All 13 named data sources supported natively or via Fivetran managed connector. All AI requirements (conversational querying, agent access, ML, semantic search) delivered natively via Snowflake Cortex. Vector capabilities, document processing, Power BI integration all native. Only Microsoft Clarity requires a custom build (3 days) due to lack of a standard connector — a well-understood exception.

2. Platform Architecture

Compute/storage separation is fundamental to Snowflake's architecture — infinite storage scaling and independent compute scaling are not add-ons but the core design. Multi-cloud: Snowflake runs natively on Azure and AWS with no re-architecture. EU data residency enforced by account region. Three isolated environments (DEV via zero-copy clone, STAGING, PROD) supported with no additional licensing.

3. Pricing Transparency

Snowflake's credit-based model is fully transparent and predictable once usage patterns are established. No upfront commitment required on pay-as-you-go. Reserved capacity (1-year or 3-year commitment) reduces credit cost by 20–40%. The main uncertainty is concurrency scaling costs at peak load — mitigated by resource monitors and multi-cluster warehouse sizing. Fivetran connector licensing is a separate line item (€400–800/month for this connector set).

4. Learning Curve

Any engineer who knows SQL is productive on Snowflake within days. Snowsight provides an excellent in-browser experience with no local tool setup. dbt is the only non-SQL skill required for transformations — it has a gentle learning curve and extensive documentation. The main learning investment is understanding the credit/compute model and configuring auto-suspend correctly to avoid unexpected costs. Snowflake University, extensive documentation, and a large community (250K+ practitioners) accelerate onboarding.

5. Support Model

Snowflake offers Standard (business hours), Premier (24/7 with 1-hour SLA for critical), and Business Critical support tiers. A named Technical Account Manager is available on Business Critical tier. Standard tier suffices at launch — escalate to Premier as the platform becomes business-critical. Partner delivery network (Big Four, boutique data consultancies) for professional services; implementation partner will be named in Module 2.

2.1

Implementation Philosophy

The implementation is structured in four phases, each delivering a working increment. Phase 1 targets production readiness within 8 weeks — covering the highest-value data sources and core BI capability. Subsequent phases add depth: ML workloads, AI conversational features, partner-facing access, and operational excellence.

Principle: Every phase ends with a working, tested, production-deployed increment. No "big bang" go-live. The data platform is operational and delivering value from Week 8, with each subsequent phase adding capabilities without disrupting existing consumers.

Team Model

Data Engineer (Lead): Snowflake setup, Terraform, Snowpipe, dbt core models, warehouse tuning.
Analytics Engineer: dbt Gold models, Power BI connector, semantic layer, data quality tests.
Platform Engineer (part-time): CI/CD, Terraform state, GitHub Actions, security review.
Your Team: Business logic validation, Power BI report development, acceptance testing.

Tooling Stack

IaC: Terraform + Snowflake provider
Ingestion: Fivetran + Snowpipe
Transformation: dbt Core / dbt Cloud
Orchestration: dbt Cloud jobs + Snowflake Tasks
CI/CD: GitHub Actions / Azure DevOps
Migrations: schemachange
BI: Power BI + Snowsight
AI: Snowflake Cortex (built-in)

2.2

Phased Delivery Plan

Phase 1 — Weeks 1–8

Foundation & Core BI

Target: Production-ready analytics on REP and CRM data. Power BI connected.

Work StreamDeliverableWeek
InfrastructureSnowflake account on Azure West Europe. All warehouses (ETL, BI, ADHOC). RBAC role hierarchy. Resource monitors. Terraform-managed IaC.1–2
EnvironmentsDEV / STAGING / PROD separation. CI/CD pipeline (GitHub Actions). schemachange migrations framework.1–2
Ingestion — REPAzure SQL Server → Fivetran → raw_db. Azure Event Hubs → ADLS → Snowpipe → raw.events. All running in PROD.2–4
Ingestion — CRM/ERPDynamics 365 → Fivetran → raw_db. PostgreSQL → Fivetran → raw_db.3–5
Silver layer (dbt)dim_property, dim_user, dim_partner, fact_portal_events, fact_listing_activity. Streams + Tasks for near-real-time event processing. dbt tests for all models.3–6
Gold layer (dbt)listing_performance_weekly, user_funnel_daily, partner_activity_summary. Materialized views for sub-second BI queries.5–7
Power BISnowflake connector configured. DirectQuery on Gold layer. Core dashboard: listing volume, portal funnel, partner activity.6–8
Data qualitydbt tests on all Silver/Gold models. Freshness checks. Alerting on failure via Slack/email.7–8
Phase 2 — Weeks 9–14

Marketing Analytics & Extended Sources

Target: Marketing ROI reporting, SDLC analytics, full source coverage.

  • GA4 + Google Ads + Meta Ads + TikTok Ads ingestion via Fivetran → marketing ROI Gold model
  • Google Search Console + Google Analytics → SEO/traffic Gold model
  • Jira + Azure DevOps + Confluence + Trello ingestion → SDLC Gold model (sprint velocity, delivery metrics)
  • Microsoft Clarity custom pipeline (Blob export → Snowpipe → UX analytics model)
  • Expanded RBAC: partner-facing roles with Row Access Policies on listing/CRM data
  • Dynamic Data Masking on all PII columns across Silver and Gold layers
  • GDPR erasure workflow (stored procedure for data subject deletion requests)
Phase 3 — Weeks 15–20

AI, ML & Conversational Data

Target: Non-technical staff can query data conversationally. Document intelligence live. Property recommendation model.

  • Cortex Analyst deployment — semantic model YAML defining business vocabulary (listing, session, conversion, partner). REST API endpoint for internal chat UI.
  • Cortex Search — document ingestion pipeline for property contracts, legal deeds, inspection reports. Chunking + embedding via PARSE_DOCUMENT(). RAG endpoint for legal team queries.
  • Snowpark ML — property price prediction model (features: type, area, location, market conditions). Deployed as Snowflake UDF, queryable in SQL. Inference on Gold layer property data.
  • User segmentation model — RFM (Recency, Frequency, Monetary) clustering on portal event and transaction data. Segments written to Gold layer for targeting.
  • Cortex LLM functions — auto-classification of listing descriptions by property type, amenities extraction, contact note summarisation.
Phase 4 — Weeks 21–26

Partner Access, Open Standards & Operational Excellence

Target: Partner self-service, Iceberg tables for data portability, full observability.

  • Secure Data Sharing for partner organisations — live access to their own listing/activity data without Snowflake account. Zero-copy, zero-latency.
  • Iceberg table migration for Gold layer — data stored in customer-owned ADLS bucket in Apache Iceberg format. Full data portability — queryable by any engine.
  • OpenLineage metadata export to data catalog (Atlan or dbt-native docs site)
  • REST API layer — FastAPI wrapper on Snowflake SQL API for partner portal integration. Documented Swagger spec. Azure API Management for rate limiting, auth, monitoring.
  • Full observability stack — dbt Montecarlo integration or custom Snowflake Alerts for data SLA monitoring. Grafana dashboard on Snowflake Account Usage metrics.
  • Handover documentation, runbooks, on-call procedures, cost optimisation review
2.3

Deliverables Matrix

Deliverable Ph 1Ph 2Ph 3Ph 4
Snowflake account (Azure West Europe) + warehouses + RBAC✓———
Terraform IaC for all Snowflake resources✓———
CI/CD pipeline (GitHub Actions / Azure DevOps)✓———
Azure SQL Server + PostgreSQL + Dynamics 365 ingestion (Fivetran)✓———
Azure Event Hubs → Snowpipe pipeline (near real-time events)✓———
dbt Silver layer (dimensions + fact tables)✓———
dbt Gold layer (BI-ready aggregations)✓———
Power BI connection + core dashboard✓———
Marketing sources (GA4, Ads, Search Console, Clarity)—✓——
SDLC sources (Jira, Azure DevOps, Confluence, Trello)—✓——
GDPR masking, row access policies, erasure workflow—✓——
Cortex Analyst (conversational NL→SQL)——✓—
Cortex Search + document RAG pipeline——✓—
Snowpark ML models (price prediction, user segmentation)——✓—
Secure Data Sharing for partner organisations———✓
Iceberg tables (open format data portability)———✓
REST API wrapper + Azure API Management———✓
Full observability stack + runbooks———✓
2.4

Support Model

Delivery Track — New Business Features

Post-go-live changes (new data sources, new Gold models, new AI features) are delivered via the same /spec → /build → /review → /deploy workflow used for implementation. T-shirt sized change requests: XS (0.5 day), S (1–2 days), M (3–5 days), L (1–2 weeks). Monthly change budget recommended: 2–4 S/M changes per month at steady state.

Maintenance Track — Operational

Ongoing: pipeline health monitoring (automated alerts, first response <4hr in business hours), configuration changes (warehouse sizing, new role grants, schema additions), Fivetran connector version updates, dbt package upgrades, quarterly cost optimisation review. Monthly retainer recommended for a guaranteed SLA.

Support TierCoverageCritical Incident SLAChange Request
StandardBusiness hours (Mon–Fri 09:00–18:00 EET)4-hour response, same-day fixBest effort within sprint
EnhancedExtended hours (07:00–22:00 EET)2-hour response, 4-hour fixPriority queue, 3-day SLA
Premium24/7 on-call30-min response, 2-hour fixDedicated allocation, same-week
2.5

Risk Register

RiskProbabilityImpactMitigation
Concurrency scaling costs exceed budget on BI warehouse Medium Medium Resource monitor with suspend at 110% of monthly budget. Multi-cluster warehouse configured correctly from day 1. Weekly cost review in first 90 days.
Fivetran connector for Microsoft Clarity unavailable / API changes Medium Low Custom pipeline (3 days) is already planned as the implementation path. Clarity export format is stable. If Clarity is replaced, the custom pipeline is replaced — no Snowflake changes required.
Azure SQL Server CDC permissions not granted by ops team Medium Medium Fivetran supports both log-based CDC and scheduled incremental sync. If CDC permissions are blocked, fall back to high-watermark incremental sync. Latency increases from near-real-time to scheduled (configurable, minimum 15 minutes).
EU AI regulation constrains Cortex LLM function usage on PII data Low Medium Cortex processes data within the EU Snowflake account. LLM functions applied only to non-PII fields (property descriptions, anonymised document text) by policy. Row access policies ensure PII never flows to LLM functions.
Multi-cloud migration (Azure → AWS) disrupts data platform Low Low Snowflake runs natively on both Azure and AWS. Cross-cloud replication replicates the Snowflake account to an AWS region. BI tools connect to the same Snowflake account URL regardless of underlying cloud. Migration = adding a replication group, not re-architecting the platform.
2.6

Indicative Cost Summary

Important: All figures below are rough-order-of-magnitude estimates based on the sizing information provided. Actual costs depend on query patterns, concurrency, and Snowflake contract terms. A formal quote from Snowflake account team is required for binding figures. Committed-use discounts (20–40%) are available on 1-year and 3-year contracts.

ComponentMonthly (PAYG)Monthly (1yr commit)Notes
Snowflake compute (Standard tier, ~500 credits/month)€1,500€1,100Based on €3.00/credit PAYG, €2.20/credit committed. Scales with actual usage.
Snowflake storage (1 TB target)€25€25€23/TB/month on Azure. Effectively free at current 150 GB.
Fivetran (managed connectors)€600€450~12 connectors. Fivetran pricing per connector, varies by source volume. Get formal quote.
dbt Cloud (Team tier)€100€100Up to 8 seats. dbt Core (free, self-hosted) is an alternative for cost-sensitive scenarios.
Total platform (ongoing)~€2,225/mo~€1,675/moExcludes implementation professional services and Snowflake platform support tier.
Phase 1 implementation (professional services)€20,000–30,000 one-time8 weeks, 2–2.5 FTE. Varies by partner rates.
Phases 2–4 implementation€25,000–40,000 one-time18 weeks, 1.5–2 FTE average.