200+ agents in production.
One platform underneath.
A Fortune 200 financial services firm runs Dataworkz across advisor enablement, regulatory lineage, operations, engineering, and mainframe modernization. In their VPC. With the audit trail their regulators expect.
across 5 business domains
What this firm built on Dataworkz.
Agents at scale on a single foundation, not point tools stitched together. Across multiple workflows and applications, every one of them inherits the same connectors, the same audit trail, and the same in-VPC deployment.
- 5Business domains: advisor enablement, regulatory and audit, operations, engineering productivity, mainframe modernization.
- 16Systems connected: SFDC, MongoDB Atlas, DB2, Pentaho, Collibra, ServiceNow, and more: across data, tools, and platforms.
- 1Platform underneath: Build · Connect · Trust · Measure. Composable across every agent.
- Customer
- Fortune 200 financial services firm
- Industry
- Wealth management & financial planning
- Scale
- 50,000+ employees · U.S. and Canadian branch network
- Deployment
- In customer VPC · SOC 2 Type II · audit trail by default
- Agents in production
- 200+ across 5 business domains
- Status
- Live
- Connected systems
- SFDCMongoDB AtlasOneDriveJIRAGitHubConfluenceSharePointServiceNowDB2KafkaSparkPentahoCollibraBroadcom ClarityApptioOracle
Six problems on the engineering and data team's desk that no point tool was going to solve.
Senior advisors retiring faster than juniors could replace them.
Field advisors escalated client questions to a central domain-expert team that hand-searched internal PDFs. Slow for clients. Single point of failure on the firm's most experienced people.
Lineage built by hand, every reporting cycle.
Analysts reverse-engineered Pentaho ETL into Collibra lineage records auditors would accept. Effort scaled linearly with every new pipeline.
3,000 developers bottlenecked on ticket overhead.
Engineers wrote ServiceNow submissions by hand, copying context out of JIRA, GitHub, Confluence, and SharePoint. Per-ticket overhead was the cap on shipping speed.
15,000 daily calls the ops team couldn't keep up with.
Mostly status questions the underlying DB2 systems already had answers to. The data existed; the path to self-serve didn't.
Contractor invoices reconciled line by line.
Hours and rates checked against Broadcom Clarity and Apptio by hand. Overbilling slipped through. Hours of weekly reconciliation per person.
COBOL nobody wanted to touch.
Legacy programs the firm depended on. No path to safe modernization. No API standardization across DB2, Oracle, and MongoDB.
The firm had tried building this in-house. Three to six months in, they had cloud AI infrastructure stood up and no production use case shipped. They needed a platform, not another framework.
One platform. Five domains. Composable agents.
A selection of the firm's production workflows, in detail. Each ships against the systems the firm already runs on, on the same Build / Connect / Trust / Measure foundation.
Advisor enablement
Surface institutional knowledge to the people meeting clients.
- What it does
- Field advisors ask questions in plain language ("IRA distribution rules for a client in this scenario"). The agent runs permission-aware retrieval across institutional PDFs and joins the answer with the advisor's relational client profile.
- Connects
- OneDriveSFDCMongoDB Atlas
- What it changed
- Junior advisors deliver senior-quality answers. The central domain-expert team is no longer the bottleneck.
Regulatory & audit
Lineage and evidence at the speed of the business, not the auditor.
- What it does
- Scans Pentaho .kjb and .ktr files, resolves placeholders, queries the Collibra API, generates structured lineage mappings, and uploads CSVs that Collibra renders as visual lineage at the app, table, and column level.
- Connects
- PentahoCollibra
- What it changed
- Lineage work that used to take weeks of analyst time now completes in minutes, with documentation auditors accept.
Operations
Deflection and reconciliation built on the firm's real systems.
- What it does
- Detects DB2 change-data-capture events on Kafka, generates Spark jobs that transform the data and load it into the client portal API. Clients self-serve in the portal instead of calling the contact centre.
- Connects
- DB2KafkaSparkClient Portal
- Scale
- 15,000 calls/day routed through self-serve deflection.
- What it does
- Ingests contractor invoices from email, extracts names / hours / rates / dates, fuzzy-matches against Broadcom Clarity, flags discrepancies, auto-approves or routes for review, and logs an audit trail.
- Connects
- EmailBroadcom Clarity
- What it changed
- Hours of weekly reconciliation per person returned to the team. Overbilling caught before payment.
Engineering productivity
Make the dev org's day-to-day work consistent and standards-compliant.
- What it does
- Reads JIRA tickets, GitHub commits, Confluence pages, and SharePoint docs. Synthesises a change-control submission with all required fields. Validates against compliance templates. Submits to ServiceNow. Tracks approval.
- Connects
- JIRAGitHubConfluenceSharePointServiceNow
- Scale
- 3,000 developers across the firm.
- What it does
- Generates API code for any database, DB2, Oracle, MongoDB, against the firm's internal API standards.
- Connects
- DB2OracleMongoDB
- What it changed
- Closes the gap between "we have a database" and "we have a standards-compliant API on top of it."
Mainframe modernization
Make legacy code legible before anyone changes it.
- What it does
- Maps COBOL data structures, business logic, and UI flows so engineers can read what the mainframe is actually doing before they touch it.
- Connects
- Mainframe
- What it changed
- Onboarding new mainframe engineers is faster. Migration risk is lower because the agent makes legacy code readable first.
Build it in-house. Or buy a closed AI suite. Or run on the platform that does both.
Three quarters of infrastructure, no production use case.
Three to six months of cloud AI infrastructure work before anything ships. The path most large financial institutions are still on.
Composable agents on the firm's own stack.
Bring your own LLM, embedding model, vector store, and data store. 100+ connectors across the firm's actual stack. Deploys in the firm's VPC. Audit trail on every decision by default.
Vendor-locked stack with the wrong connectors.
Limited coverage for the firm's real systems (DB2, Pentaho, Collibra, Broadcom Clarity, mainframe). No path to keep data inside the firm's perimeter.
Agents the firm builds on Dataworkz are agents the firm owns: composable across the platform, deployable inside their perimeter, instrumented with the audit trail their regulators expect.
One platform. Four pillars.
The same Build / Connect / Trust / Measure foundation runs every agent in production at this firm.
Build
Agent Builder generates working agents from plain-English specifications. Engineers describe the workflow; Dataworkz produces the plan, the data pipeline, the retrieval logic. New use cases ship in days, not quarters.
Connect
100+ connectors across the systems this firm already runs on. SFDC, DB2, Pentaho, Collibra, Broadcom Clarity, mainframe. No firm-side rebuild. No data leaves the perimeter.
Trust
Deployed inside the firm's VPC. SOC 2 Type II. Permission-aware retrieval inherits role-based access from the underlying source through to every agent. End-to-end audit trail per agent decision.
Measure
Every agent ships with metrics, traceability, and A/B testing from day one. Every workflow proves its return: time saved, cost avoided, errors prevented, instrumented, not estimated.
This is what production agentic AI looks like in a regulated enterprise.
If you're at a peer firm: wealth management, capital markets, retail banking, three things matter from this deployment.
Composable agents on one platform.
Agents at scale on the same foundation, across multiple workflows and applications. The firm did not stitch together point tools. The next agent ships faster because the connectors, the audit trail, and the deployment model are already in place.
Inside the perimeter, by default.
Deployed in the firm's VPC. Their data stays where their regulators and security team expect it, no exfiltration, no shadow stack.
Audit trail their auditors trust.
Every agent decision logged. Lineage runs end-to-end. The same discipline the firm applies to regulatory data, applied to every agent it builds.
Your firm's work is too important to wait three quarters for an in-house AI platform.
Book a working session with the FS team. We'll walk through the workflows in this case study against the systems your firm already runs on.