How banks and fintech companies can connect business meaning, financial data, enterprise knowledge and tools to build agents that investigate, recommend and coordinate action.

A banking agent needs to know what “this customer” means before it can explain what changed—and what should happen next.

Financial intelligence brings customer economics into view. Agentic AI builds on that foundation by connecting business relationships, evidence and tools into a workflow that can investigate a problem and coordinate a response.

Give the agent a business model, a clear goal and a controlled path to action.

This reference architecture follows a Customer Relationship Intelligence Agent from a profitability question to an approved follow-up task. The concepts, figures and interactions are illustrative, with no client data or live system connections.

4. Separate ontology, analytical semantics and evidence

These layers answer different questions. Designing them together gives the agent a consistent route from a business phrase to a measurable result and its supporting context.

A Semantic View can capture substantial business meaning without being a complete ontology. A knowledge graph is one way to represent connected context; a carefully maintained mapping table may be sufficient for a simpler domain. Semantic Views documentation ↗

Start with canonical identifiers, aliases, relationship types, effective dates and the financial measures required by the first workflow. Map them to the underlying data. Where graph traversal or formal reasoning is needed, expose a tested query or custom tool that returns a bounded, traceable result.

Use Cortex Search for document evidence such as contracts and service records. Attach entity identifiers, document versions and effective dates to retrieved material so the agent can distinguish an active agreement from an expired one. Cortex Search with Cortex Agents documentation ↗

Keep meaning and enforcement separate. An ontology explains relationships and constraints. Authorization must be enforced by the data platform, retrieval configuration and action services.

5. Follow the investigation from goal to verified action

The following walkthrough shows an example execution path. A real agent may revisit a step, request clarification or stop when evidence is incomplete. The important property is that every step produces an inspectable result.

Customer Relationship Intelligence Agent
Inside the agentic workflow

Explore six steps using fictional records. This illustration has no live data connection and performs no external actions.

The financial bridge: 300 → 180 contribution. Revenue fell by 80 units and expenses rose by 40 units, explaining the 120-unit change. An agent must distinguish this measured decomposition from a claim about why the changes occurred.

6. Establish who is acting and what they may do

The ontology resolves the identity of the customer being investigated. Execution identity establishes which human or agent is requesting access. A customer relationship in the ontology does not grant permission to read that customer’s records.

Snowflake Horizon Catalog and native access controls provide the governance foundation: classification, lineage, role-based privileges, row access policies and masking. Enterprise SSO establishes human identity; provisioning and role assignments connect that identity to permitted resources. Horizon Catalog ↗ Access control ↗

For readers familiar with Databricks: Horizon Catalog plus Snowflake’s native security controls is the closest comparison to Unity Catalog. Both govern access to data and AI assets; enterprise identity providers establish who the users are. The specific execution identity and tool permissions still need to be designed. Unity Catalog overview

7. Secure every retrieval and tool boundary

Authorization must follow the execution path. Validate the role used for analytical queries, the access scope of each search service, the execution rights of stored procedures and the credentials used by external tools. A successful agent sign-in does not establish all of these permissions.

Cortex Search needs an explicit access design. Search services run with the owner’s rights. A role allowed to query a service can retrieve indexed content even when it cannot read the underlying source tables. Source-table permissions therefore do not automatically provide per-user document filtering. Use appropriately scoped services and trusted authorization enforcement; agent-selected filters are not a security boundary. Cortex Search security requirements ↗

In this banking blueprint, contract and service-record retrieval is restricted to an approved corpus for the authorized relationship scope. Retrieved text is treated as evidence: instructions embedded in a document cannot expand the agent’s permissions or authorize an action.

For external operations, use scoped credentials, validate inputs on the server and recheck destination permissions at execution. Allow only the intended operation and fields. Keep secrets out of prompts and redact sensitive content in application logs. These are implementation controls that complement Snowflake’s native policies.

8. Give actions an explicit execution boundary

MCP connectors let Cortex Agents discover and invoke tools in external applications. Custom tools can expose institution-specific operations. For this workflow, the useful action is a relationship-management task with an owner, evidence and a next step. MCP connectors documentation ↗

Design the action interface around a narrow operation such as “create an approved follow-up task.” Validate its inputs on the server, check authorization at execution time and return an operation identifier that can be verified.

The approval step is part of this reference architecture’s application and tool design; connecting an MCP server alone does not create that control. Credit decisions, pricing changes and financial transactions are outside this example’s permitted actions.

9. Compose expertise with focused tools and skills

Organize capabilities around responsibilities: customer identity, financial analysis, document evidence and action execution. Give each one a defined input, output and access scope. That makes it easier to test a capability independently and reuse it across workflows.

Snowflake agent toolsets can inherit tools from other configured agents. The tools are combined for the calling orchestrator; this is tool composition, rather than an automatic delegation to independently reasoning agents. Agent toolsets documentation ↗

If a workflow needs separate agents to hand off work, explicitly design the coordination, shared state, failure handling and completion criteria. Start with one orchestrator and focused tools when that is sufficient.

Agent skills add reusable instructions and scripts for recurring tasks. A financial-investigation skill might define the required comparison checks and report structure, while calculation and action tools enforce the underlying operations.

10. Use agentic engineering to build and improve the workflow

Snowflake CoCo supports development work across data engineering, analytics and agent building. Use it to accelerate the implementation work around the agent: draft semantic models, author queries, refine tool specifications and help investigate failed tests. Snowflake CoCo documentation ↗

The engineering loop should stay concrete: define a representative banking question, build the minimum context and tools needed, run it against known results, inspect the trace and correct the failing layer.

Define the task → Build context & tools → Evaluate → Review & version

Keep ontology mappings, semantic definitions, tool contracts and evaluation cases versioned together. When a customer hierarchy or profitability rule changes, the team needs to know which answers and actions may be affected.

11. Observe the run, then evaluate its quality

Cortex Agent evaluations support testing against expected results and other quality criteria, including tool-related metrics for supported tools. Live monitoring provides traces for investigating actual runs. Use both to establish where an error arose and whether a change improved the workflow. Cortex Agent evaluations documentation ↗ Agent monitoring documentation ↗

Make each run inspectable

Cortex Agent monitoring records execution traces with tool calls, SQL activity, responses and feedback. Query and access histories provide a complementary view of who accessed which objects. Connect these records to the application’s approval and external-action receipts using correlation identifiers. External approval and business-system events require their own instrumentation.

Use targeted integration tests for approval enforcement, permissions and retry behavior. Include adversarial document content, stale ownership mappings and partially available data. A correct final paragraph is insufficient if the workflow crossed an access boundary or created the wrong task.

12. Expand autonomy in measured steps

Meet Our Customer

VIRA Insight designs, engineers, manufactures and installs retail environments, fixtures and displays. Its work connects the commercial needs of retailers and brands with the operational demands of manufacturing, distribution and installation.

Behind those operations, information was spread across Salesforce CRM, Epicor ERP, SharePoint, SQL Server and custom APIs. Sales opportunities, accounts receivable, shipping and backlog each contributed part of the business picture. VIRA wanted a trusted, connected view of that information—and conversational access to trusted enterprise metrics.

NorthBay was engaged to deliver that foundation on Snowflake, bringing together enterprise data, consistent business definitions and production reporting.

The Challenge

VIRA’s business systems held the information its teams needed, but that information was difficult to bring together consistently.

Salesforce was isolated from backend operational databases. Epicor, SQL Server, SharePoint and custom APIs operated as separate sources, leaving departments with conflicting versions of core metrics. Building a consolidated view required repeated spreadsheet exports, manual preparation and reconciliation.

Those reporting cycles were slow and prone to error. Time spent assembling data reduced the time available to understand sales opportunities, receivables, shipping performance and backlog. The same fragmentation also made it difficult to establish a reliable starting point for advanced analytics and AI.

The brief was to replace siloed reporting with a governed enterprise data foundation: automate ingestion, refine data into trusted business metrics, deliver priority reports, and create a practical path toward natural-language access.

The Vision

NorthBay designed a Snowflake-powered lakehouse that brings enterprise sources into a common, governed environment. A layered data architecture separates raw ingestion from data integration and business-approved reporting, while a semantic layer gives those metrics a consistent meaning.

A single foundation across enterprise systems

Snowflake OpenFlow supports automated ingestion from Salesforce, Epicor, SharePoint and custom APIs. DBT transformations organize data into Bronze, Silver and Gold layers: raw source data lands in Bronze, is cleaned and conformed in Silver, and becomes curated business metrics in Gold. This provides a repeatable route from source systems to reporting.

Business meaning built into the data

A Cortex Analyst semantic layer aligns core KPI definitions across teams. Validated Gold metrics provide a consistent basis for business reporting and natural-language queries. Power BI serves executive dashboards and self-service reporting. Building on the Opportunities data, NorthBay also built a Cortex Agent within Snowflake Intelligence to provide semantic reporting through natural-language questions.

Governed by design

The architecture combines automated transformations, shared enterprise definitions and role-based access across the data layers. NorthBay’s end-to-end solution architecture was reviewed and approved by VIRA before production rollout.

The Q3 continuation focuses on expanding the lakehouse foundation, governance, additional Epicor tables, and customer and vendor master data. This builds on the delivered Q2 scope rather than treating modernization as a one-time reporting exercise.

Why NorthBay?

NorthBay brought together data architecture, ingestion, transformation, semantic modeling and reporting delivery to address VIRA’s immediate operational needs and longer-term AI ambitions. The engagement connected multiple enterprise systems, aligned business definitions and delivered four production reports within the agreed schedule and budget.

As an AWS Premier Consulting Partner, NorthBay combined cloud engineering with Snowflake delivery expertise to establish a governed platform that VIRA can extend. The result gives teams a more consistent view of the business, natural-language access to Opportunities data, and a foundation for further analytics and AI capabilities.

Which customers contribute most to the business? Where is revenue growing while profitability is weakening? Which products and channels deserve a closer look?

Banks and fintech companies hold much of the information needed to explore these questions. The challenge is connecting revenue, expenses and customer context into a consistent financial story that leadership and customer teams can use.

The first step toward customer intelligence is understanding the economics of the relationship.

This blueprint draws on NorthBay’s banking delivery experience to show how a Snowflake foundation can support financial dashboards, customer-level analysis and a next layer of conversational intelligence.

These patterns can coexist within the same platform. Choose the pattern for each dataset, then standardize quality and business definitions in the downstream layers. Snowflake data loading overview ↗

Useful quality checks include duplicate financial entries, missing customer keys, unmapped product codes, inconsistent signs and unexpected currency values. Exceptions should be visible to the people responsible for resolving them.

Snowflake Tasks can schedule processing or run in response to supported events. Refresh frequency should be designed around source availability and the business reporting window. Ingestion and downstream processing should be designed together to meet the required data freshness. Snowflake Tasks documentation ↗

4. Give leaders a connected view of financial performance

The executive dashboard organizes the financial story around revenue, expenses and profitability over time. Business breakdowns—such as product, branch, income type and currency—help leaders move from the headline trend to the area that deserves attention.

Amazon QuickSight is one of many BI tools that can connect to Snowflake. Microsoft Power BI, Tableau and other compatible tools can present the same governed financial models, allowing teams to use the reporting experience that fits their needs. Keep common metric definitions in reusable data models so the financial story remains consistent across tools. Snowflake partner and technology ecosystem ↗

A decline in contribution might reflect a revenue change, a cost change or a shift in product mix. Showing those measures together provides a more useful starting point for investigation than reviewing separate reports.

The model should preserve currency context. Comparing native-currency amounts is different from reporting consolidated performance: consolidated measures need an agreed exchange-rate source, conversion date and reporting currency.

5. Make customer profitability visible over time

The Customer 360 view brings the financial relationship into focus: revenue, expenses, recorded activity and the way contribution changes over time. This helps customer teams ask better questions about a relationship before deciding on an action.

A useful starting scope is the customer’s footprint in revenue and expense records. Transaction counts in this context describe those records; they should not be confused with a complete history of deposits, withdrawals or payment activity.

Work with the institution’s security, governance and risk teams to define the controls required for its environment and validate their operation.

8. Add conversational analytics to the trusted foundation

Once customer and financial measures have clear definitions, natural-language analytics becomes a useful additional way to explore them. The banking solution included Snowflake Intelligence alongside the dashboard experience; the architectural principle is to ground conversational questions in the same business meaning used for reporting.

Snowflake semantic views define business concepts, relationships, dimensions and metrics. That semantic layer helps supported AI tools interpret a business question in the context of the data. Semantic views documentation ↗

The analytics experience should make the reporting period, filters and metric definitions clear. Test representative questions against known answers, and retain human review when interpreting findings or deciding what action to take.

The analytics experience should make the reporting period, filters and metric definitions clear. Test representative questions against known answers, and retain human review when interpreting findings or deciding what action to take.

9. Expand from financial visibility to richer intelligence

The architecture provides a practical sequence for extending value. Start with the financial questions supported by the available records, then add the data and operating capabilities needed for the next decision.

Executive Summary

Tamimi Markets, one of Saudi Arabia's largest and longest-established grocery retailers, faced a familiar enterprise problem at an unfamiliar scale: hundreds of millions of operational data points locked behind BI queues and specialist teams, while high-volume retail work, ordering, forecasting, customer engagement — stayed manual and spreadsheet-bound. Rather than pursue a narrow point solution, Tamimi partnered with NorthBay Solutions, an AWS Premier and Anthropic partner, to build AskBaba: a bilingual (Arabic/English) agentic AI platform powered by Anthropic's Claude on Amazon Bedrock, running inside Tamimi's own AWS environment and delivered where employees already work, in Microsoft Teams.

The result is a business milestone as much as a technical one: Tamimi Markets is the first retailer in the Kingdom of Saudi Arabia to run agentic AI at enterprise scale, not a pilot, not a single chatbot, but a governed, production platform of 19+ specialized Claude-powered agents that answer questions in seconds, automate daily purchase ordering, forecast demand, and re-engage customers. AskBaba delivers measurable value across every persona in the organization — store managers, category buyers, HR, IT, finance, and executive leadership and across every operational stream, from analytics and procurement to forecasting and customer engagement, all while keeping enterprise data under Tamimi's own governance. Claude powers both what AskBaba does for the business and how NorthBay designed, built, and delivered it.

Results at a Glance

The Challenge

Tamimi Markets runs a nationwide network of stores generating enormous operational data, hundreds of millions of transaction line items across sales, inventory, stock movement, promotions, and procurement, flowing from SAP S/4HANA into an Amazon Redshift data warehouse of more than 130 tables. The data existed; access to it did not.

Store managers, category buyers, and executives depended on BI dashboards and analyst queues for even routine questions, how a store performed yesterday, which products were running low, whether a promotion moved volume. Answers took hours or days, arrived in English only, and required desk access. Meanwhile, high-volume operational processes such as demand forecasting, competitor price tracking, and customer win-back campaigns were manual, spreadsheet-driven, and dependent on a small group of specialists. Daily purchase-order creation, by contrast, already ran on TIPS, Tamimi’s own proven ordering platform, and one of the retailer’s strongest existing systems with room to add a real-time validation layer as volume and complexity grew.

Tamimi Markets wanted a single, trustworthy assistant its employees could talk to: one that understood retail language in both Arabic and English, answered from governed enterprise data rather than guesswork, respected each employee's role and permissions, and could safely take action, not just report. It had to serve every persona in the business, from the store floor to the boardroom, run inside the tools employees already use (Microsoft Teams), and scale from a pilot to the entire organization.

The Solution: An Agentic Platform Built on Claude & AWS

NorthBay Solutions designed AskBaba as an agentic AI platform with Claude at its core, running on Amazon Bedrock inside Tamimi Markets' own AWS environment, keeping the client's enterprise data under the customer's governance, so every persona gets a different, role-appropriate answer from the same trusted foundation.

A supervisor–specialist agent architecture

Rather than a single monolithic bot, NorthBay built a LangGraph (Day 1 Architecture and progressing towards an AgentCore architecture) multi-agent system in which a Claude-powered supervisor understands each request, then routes it to one of 19+ specialist agents, sales analytics, inventory, stock movement, sales budgets, product catalog, competitor pricing, predictive forecasting, purchase-order creation and inquiry, HR knowledge, IT incidents, business events, weather, scheduled reporting, and more. Each specialist combines Claude's reasoning with governed tools: text-to-SQL over the Redshift warehouse, retrieval-augmented generation (RAG) over HR and policy documents, ServiceNow ticketing, SAP purchase-order APIs, and Airflow job scheduling. Agents complete their task or escalate back to the supervisor, so conversations flow naturally across domains.

Why Claude

Reasoning over enterprise schemas. Claude reliably translates ambiguous, bilingual retail questions into precise SQL over a 130+ table warehouse, respecting business rules (posting vs. transaction dates, fresh vs. non-fresh classification, store hierarchies).

• Multimodal understanding. Claude's vision capability reads product photos, shelf images, and competitor brochures; voice queries flow through transcription into the same agents.
• Arabic + English fluency. One assistant serves the whole workforce, detecting and responding in each user's language.
• Safety and control. Amazon Bedrock Guardrails, role-based access control enforced at the agent layer, and prompt-injection hardening keep responses grounded and appropriate for the workplace.
• The Solution: An Agentic Platform Built on Claude & AWS

Built and Delivered with Claude

The platform is built on AWS with infrastructure-as-code (Terraform), CI/CD with environment promotion and production deploy authorization, SageMaker ML pipelines for demand forecasting with SHAP explainability, and event-driven integration (SQS, Lambda, Airflow) to SAP and messaging channels. NorthBay's delivery team also adopted Claude in its own engineering workflow, using Claude-assisted development with mandatory human review, plan-first tickets, and self-review gates. In effect, Claude powers both what AskBaba does for Tamimi Markets and how NorthBay built and delivered it, accelerating architecture, coding, and testing from the first release through every extension phase, shortening time-to-production without compromising the engineering rigor an enterprise platform demands.

Delivery & Adoption

AskBaba is in production and has been serving Tamimi Markets employees since 2025, delivered through four major releases and two funded extension phases, evidence of sustained customer commitment rather than a one-off pilot.

• Production platform. Live across development, QA, and production environments on AWS, with formal release management, a 12-week software-quality-assurance cycle, and a hotfix-governed production freeze discipline.
• Where employees work. Delivered inside Microsoft Teams (with an embedded web chat option), so adoption required no new application, employees message AskBaba like a colleague.
• Broad functional adoption. Natural-language sales/inventory/procurement analytics; HR-policy Q&A via RAG; ServiceNow incident creation from chat; a real-time validation layer on daily TIPS purchase-order creation, with SAP integration; ML demand forecasts with explainable drivers; image, barcode, and voice inputs; and administrator-composed broadcast messaging.
• Operational governance. Enterprise security program (IAM least-privilege, KMS encryption, secrets management, OIDC-only pipelines), PDPL-aligned data handling, and full auditability of agent actions.

Business Impact

AskBaba has changed the day-to-day economics of data access and retail operations at Tamimi Markets:

• From days to seconds. Routine analytics questions that previously waited on BI/reporting queues are now answered in under 30 seconds, self-service, 24/7, in Arabic or English.
• Growing daily usage. Adoption grew more than 40% within months of launch (from ~100 to 140+ daily active employees) across stores and head office.
• Half a billion+ records at conversational reach. The assistant reasons over 500M+ transaction line items and 130+ warehouse tables previously accessible only to data specialists.
• TIPS, validated. AskBaba adds a real-time validation layer on top of TIPS, Tamimi’s proven purchase-order platform, cross-checking daily purchase-order creation across the store network and giving every PO an explainable rationale freeing buyer hours for higher-value work.
• Forecasting with accountability. SageMaker demand forecasts with SHAP explainability inform ordering decisions, with drivers (events such as Ramadan and Eid, weather, promotions) visible to the business.
• Faster employee communications. Administrators broadcast rich announcements to 1,000+ recipients across Teams and web chat in minutes, with per-recipient delivery tracking.

Value Across Business

AskBaba's supervisor-agent architecture means the same platform serves every persona in the business differently — each getting the analytics, automation, or knowledge relevant to their role, without separate tools or integrations to manage.

What's Next?

Tamimi Markets and NorthBay Solutions continue to extend the platform on the same Claude-on-Bedrock foundation: competitor price intelligence across major KSA retailers, and personalized customer re-engagement identifying dormant customers and lapsed products, matching them to live promotions, and reaching them over WhatsApp and email with consent-managed, bilingual campaigns. Each increment reuses the same governed agent architecture, shortening time-to-value with every release and extending Tamimi's first-mover position as Saudi retail's digital transformation, aligned with Vision 2030, accelerates.

About NorthBay Solutions

NorthBay Solutions is an AWS Premier consulting partner and Anthropic partner specializing in data, analytics, and generative AI. NorthBay designs and delivers production AI systems, from agentic assistants to ML platforms, for enterprises across the Middle East, North America, and beyond.

Tamimi Markets is one of Saudi Arabia’s largest grocery retail chains, operating a nationwide network of stores and serving millions of customers with a commitment to quality and value. It is the first retailer in the Kingdom to run agentic AI at enterprise scale, built end to end on Anthropic’s Claude, extending value across every persona in the business, from store operations to executive leadership.

Availability

AskBaba was designed and delivered by NorthBay Solutions on Anthropic's Claude via Amazon Bedrock. To learn how NorthBay can bring production-grade Agentic AI to your enterprise, connect with an expert or find offerings on the AWS Marketplace.