# Enterprise Design System: Banking Systems

_Product design case study by Paul Kennedy. [Portfolio home](../../index.md) · [View the full page](index.html)_


As Product Design Lead I launched an internal design system and guided it to maturity while leading design on client onboarding, client reference data and reporting tools.

**Role:**

Product Design Lead

**Sector:**

Capital Markets, Corporate Banking

**Team:**

Design system lead, working alongside 5 other designers and reporting directly to the Director of Product Design

**Platform:**

Internal desktop web applications; components tested in Storybook

**Scope:**

Lead for the design system, plus lead designer on Client Reference Data, Client Onboarding, Active Account Monitoring, SSI management and Regulatory Exceptions Reporting. I presented each user need and proposed solution to senior stakeholders for approval before implementation.

**Outcomes:**

40+
apps on the system
10
apps led
3×
faster engineering
CAD 1m
saved a year
550+ h
saved a year
90%
fewer support questions

I launched the design system and guided it to maturity. Before it, each app team built its own tables and forms. A shared system ended that duplication, made engineering 3× faster (measured in sprint output velocity between years 2 and 3) and gave users one consistent experience across apps. I made sure every component is aligned between the design file and engineering, and tested in code via Storybook before it reaches a production app.

- Launch · Year 1 I launched with a subset of the most frequently used and requested components, chosen by analysing existing applications

- Scale · Year 2 I added more components, and in some cases combined existing ones to give applications more functionality

- Maturity and adoption · Year 3 onwards Design files and code aligned and tested in Storybook; semantic versioning; feedback channels; 40+ apps

- Colour tokens Background, field and text tokens, plus the type scale. Hex values withheld.

### Design system in use: Client Onboarding

A production Capital Markets onboarding app built from the design system. All client data is fictional.

- Row actions The context menu on a request: set as filter, request actions, view, generate email, clone and copy. User need: Set as Filter I proposed the Set as Filter micro-interaction to remove the need to open filter menus and find the filter of interest before applying it. Users can now build up filters fluidly from what they see on screen.

- Filtered by entity type Set as Filter applied from a cell, narrowing the blotter to asset management firms. The filter uses the Autocomplete search variant, with each active filter shown as a removable pill.

- Combined filters Two filters combined and saved as a custom MiFID AMF tab. Users can build unique, personalised views like this and return to them alongside the standard Active, Inactive, Cancelled and Completed tabs. User need: Seeing applied filters End users of the data table flagged a recurring issue: once they applied filters, it took 3 clicks to see which ones were active. The Autocomplete search variant shows every applied filter and sort by default, so it now takes none. I created it in partnership with the wider design team, and once I built it, the design team adopted it across all projects.

- Component design: Autocomplete Built to solve the applied-filters issue above. I built it by combining existing components rather than starting from scratch: the Dropdown input and chevron, selectable Tags for chosen options, the Dropdown menu list, and list items with checkboxes for multi-select. Shown in every menu state, plus error, warning and search variants.

- Atomic elements Autocomplete broken into its atoms: chevrons, pills, menu lists and list item states, on page and layer backgrounds.

- New onboarding request The request form as it opens: a progress rail on the left, required fields flagged for Save as Draft and for Send to Intake.

- Request in progress The same form part-completed: once the entity is found by name or LEI, KYC / AML and Reg Data status load in from downstream systems instead of being rekeyed. Deal team added as pills. User need: Locked read-only fields I observed users still trying to interact with fields in their initial read-only state. I added a lock icon to every input across the design system, so users can see at a glance which fields are read-only before they try to edit them. This cut support questions by 90% and removed mis-clicks in testing completely.

### Stakeholders

The 17 senior stakeholders I worked with across Client Reference Data, Client Onboarding and the Reporting Hubs, grouped by function. Shared across every project on this page.

LEADERSHIP AND SPONSORS · 3

ONBOARDING AND OPERATIONS · 6

PRODUCT AND DELIVERY · 3

TECHNOLOGY · 5

### Accessibility

I pushed to build the system on an open-source enterprise design system with a strong accessibility baseline, and convinced the Director of Product Design and the Head of UI to adopt it as our foundation. Every component inherits that baseline.

- Standards Components follow an accessibility checklist based on WCAG AA, Section 508 and European standards.

- Colour contrast Light and dark themes target WCAG 2.1 AA contrast.

- Automated testing Accessibility checks run on every code change, and a component counts as tested only when it passes with no violations.

- Manual testing Default and advanced states, keyboard navigation, and screen reader checks with JAWS, VoiceOver and NVDA.

### Code assignment: Legal Entity and Connection codes

A 0–1 replacement for a daily spreadsheet: a cron job created it each morning and it was submitted once, at the end of the day. Analysts now assign the CRD codes that identify an entity and its credit connections, replacing those in the legacy system, and submit them downstream in real time. I designed the grid on the design system, with inline entry, live validation and a confirmation step, working with stakeholders from Managing Directors to developers. The spreadsheet was retired after UAT, and validation has since stopped every duplicate or invalid code reaching downstream systems.

### System map

Where a code comes from, what checks it passes and where it ends up. Definitions are taken from the app's own Info panel. Entity names, UENs and codes are fictional throughout this section.

1 · REAL-TIME CODE QUEUE

2 · CODE ASSIGNMENT

3 · CODE FAMILY

4 · SUPERSEDES LEGACY CODES

5 · ACCOUNT SYSTEMS

**UEN:**

The entity's identifying number. Each Legal Entity code links 1-1 to a UEN where one exists.

**Legacy:**

The system that held entity, agreement, source and counterparty codes before CRD replaced them.

### Screens

18 entities waiting for a CRD Legal Entity Code. Summary tiles count Remaining, Validation Errors and Ready for Submission; Submit stays disabled until something is valid.

- 1 Edit in the grid Each code is entered beside the entity name and UEN it identifies, so the analyst checks the right entity as they type, with no separate form to open.

- 2 Validate before submit Bad codes are caught at the cell, before they can reach downstream systems. The toast names the rule that failed, so the fix is obvious.

- 3 Always show progress Remaining, Validation Errors and Ready counts per tab show how much of the queue is left and whether anything is blocking submission.

- 4 Explain codes in place Several CRD codes have similar names. A column tooltip and the Info panel define each one where it is used, so analysts don't need separate documentation.

Most users were familiar with inline editing in Excel, so I created the same features in the grid. Their working practices stayed unchanged: they could still copy and paste cells into the system or type codes inline manually.

I added status colour to the cells after early testing showed that users found it difficult to identify codes that had issues or were ready for submission.

### Client Account Management

Client accounts were managed through spreadsheets and email. With Client Reference Data we digitised account management in-house for the first time, replacing 4 legacy systems with new workflows and data models and saving CAD 1 million a year. It took 80+ design workshop hours with stakeholders.

- Before: current data flow Codes and mappings were kept up by hand around a legacy credit risk system and pushed downstream once a day. Simplified and anonymised from the programme's dataflow model. 1 · Upstream sources Credit approval system Counterparty data, hierarchy, facilities and limits Credit origination system Credit code requests External credit sources Facilities, credit limits and tenors Trading systems Trades and account references Re-keyed by hand 2 · Manual handling Daily code spreadsheet Created by a scheduled job each morning, submitted once at the end of the day Requests by spreadsheet and email Account and code requests tracked outside any system Manual mapping Account and counterparty mappings kept up by the data operations team End-of-day batch 3 · Legacy system Legacy credit risk system Held entity, agreement, source and counterparty codes Account mappings Updated by an end-of-day feed Once a day 4 · Downstream Risk and reporting systems Received new codes once a day Legal reporting Legal agreements loaded by end-of-day feed Problems Codes reached downstream once a day Manual re-keying and mapping Duplicate or invalid codes could reach downstream systems No single source for entity and account codes Source or downstream system Manual step

- After: future state CRD becomes the single place where codes, hierarchies, limits, agreements and mappings are created and checked, then fed downstream in real time. Simplified and anonymised from the future-state model. 1 · Upstream sources Credit approval system Counterparty data and hierarchy, fed in automatically Credit origination system Triggers credit code creation and legal agreement setup External credit sources Facilities, credit limits and tenors Data operations team Semi-manual counterparty creation, now inside CRD Real time 2 · CRD UI and data management Connection · Customer · Trading counterparty Linked by hierarchy mapping; each carries a CRD code Facilities, credit limits and tenors Held against the customer record Legal agreement Set up by a real-time automated process Alias mapping Built into the CRD account workflow Validated before submit 3 · Processing and reporting Reconciliation Aligned with source and legal systems Downstream feed and reports One feed for every consumer Real-time data flow 4 · Downstream Risk engine Transaction reconciliation, credit calculation, violation generation Legal reporting engine Fed from the legal document repository Collateral system Kept in step by the reconciliation process Outcomes of the solution 4 legacy systems replaced, saving CAD 1 million a year Real-time data flow replaces the end-of-day batch CRD codes replace legacy codes, one source of truth Legal agreements set up automatically Mappings move into the CRD workflow Validation stops duplicate or invalid codes reaching downstream Reconciliation against source systems Source or downstream system CRD module Manual step

- New data model The first structured model for account data. Every record joins back to the entity's UEN or the facility's ID, so codes, aliases, limits and agreements connect instead of sitting in separate spreadsheets. Keys and relationships only; simplified and anonymised from the physical model. Entity Facility and limits Legal agreement Governance PK Primary key FK Foreign key One to many

Every credit code in one grid, searchable by entity name, credit code, connection code or UEN.

Counterparty codes were laid out as cards to remove the need to scroll through a row-based view. I proposed adding Quick Launch and Copy to each card, so analysts can open or copy a code without leaving the screen.

Feedback from users showed the slowest step was manually rekeying data from downstream KYC systems into the Client Onboarding app. The team proposed pulling that data in automatically, and I presented the redesign to senior stakeholders. It saves 68.97 days (550+ hours) annually across 1,100 active onboarding requests. The time saving was established through repeated qualitative feedback from end users, who reported about 30 minutes of rekeying per request.

68.97 days × 8 h per day = 551.76 h · 551.76 h ÷ 1,100 requests = 0.50 h (30 min) per request

> “I have enjoyed working with you on the [Client Onboarding] project. Your timeliness, responsiveness and strategic insights have been greatly appreciated.”
> SBM Associate

The request queue and request form are shown above, under Design system in use.

Every update on a request in one side panel, newest first. Filter pills narrow it to one stage (intake, regs, KYC / AML, credit, legal or account opening), and tinted rows flag the events that need attention: deactivated, with client, needs review and submitted.

Users requested a way to view the events within a request chronologically. I redesigned the existing comments side panel to show those events, and added ways to filter the view to specific events or event types.

Managers requested a way to quickly view a summary of client outreach. The team proposed using AI to condense all emails and activity into an easily reviewed summary of progress to date and outcomes.

Users were frustrated at having to open the request to manage a client’s supporting documents, so I designed a side panel, working with the Product Owner and Business Analyst, for document management without opening the client request.

Users requested a way to add multiple products at once when they share details, to further reduce the amount of rekeying needed per request. I worked with the Business Analyst and Onboarding Associates on the design before presenting it back to the Product Owner and senior stakeholders for sign-off. This reduced time spent rekeying product details per request by an average of 60%, measured in timed sessions.

Each hub is shown briefly: what it does and its defining screens, in light and dark. All data is fictional.

### Active Account Monitoring

Monitoring for the EMIR 3.0 active account requirement. It tracks EU CCP and Tier 2 trading in each subcategory against reference-period targets, then produces the Annex III submission report.

Monthly trades per bucket across two reference periods, with the five most relevant starred and flagged on track or short.

### SSI management

A digitised hub for standing settlement instructions, their templates and bank addresses, with an approvals workflow.

Cash SSIs split into internal and external, with approval counts (for review, rejected, approved) as filter tiles. Approvals sit in the first column and status in the last.

### Regulatory Exceptions Reporting

Regulatory trade reporting exceptions in one dashboard, grouped by rejection reason, with tools to view, comment on, replay, archive and reactivate trades.

Exceptions for one regime, filtered by rejection reason, product tab and message type. Aging turns amber at 10 days and red at 30.

## How I led across projects

A sample of leadership moments behind the work above

- 1 Set the foundation I pushed to build the design system on an open-source foundation with a strong accessibility baseline, and convinced the Director of Product Design and the Head of UI to adopt it.

- 2 Set standards the team adopted I created the autocomplete search component in partnership with the wider design team, and it was adopted across all of the team's projects.

- 3 Aligned senior stakeholders I worked with 17 senior stakeholders across product, operations and technology, including 80+ hours of design workshops for Client Reference Data.

- 4 Took user needs to sign-off I presented each user need and proposed solution to senior stakeholders for approval before implementation.

- +44 7456 495994

- nedkeny@gmail.com

- LinkedIn [LinkedIn](https://www.linkedin.com/in/pj-kennedy/)
