Enterprise Financial Management System
Replacing spreadsheet-driven finance with one system: budgeting, enforced approval workflows, reporting, document management, and loan tracking with automatic interest calculation.
- Client
- Confidential enterprise client
- Sector
- Enterprise
- Capabilities
- digital platformsintegration
The problem
The organisation's finances ran on manual processes and spreadsheets. That is a workable arrangement until it is not, and it had stopped working: financial records were fragmented across files, reporting arrived late, audit trails were weak or absent, and the same data was entered more than once in more than one place. Approvals bottlenecked because nothing enforced them -- a request moved when someone remembered it. Files could be lost outright, and answering a straightforward question meant finding the person who held the right version.
The objective was not "a finance system" in the abstract. It was to replace spreadsheet-driven process with structured workflows and a centralised source of truth, and to make the record of what was approved, by whom, and when a property of the system rather than of someone's inbox.
Constraints
Integration with systems we did not control. The platform had to exchange data with external systems that had their own outages, their own release schedules and their own access rules. Poor network connectivity between them was normal rather than exceptional, and our access for testing was limited.
Client-controlled infrastructure. The application runs inside the client's own infrastructure under strict security restrictions. That shaped how we could reach the environment, how we could deploy into it, and how much we could see when something needed diagnosing.
Legacy data that did not agree with itself. The spreadsheets that had to be migrated contained inconsistent formatting, inconsistent records, and the same person spelled several different ways. Migrating it as-is would have moved the problem rather than solved it, so it was cleaned and standardised first.
Stakeholder availability. Decision-making was slow and the key users had limited time, which lengthened every cycle of clarification, validation and sign-off. Sequencing the work so that a delay in one decision did not stall the rest of the build was part of the delivery approach rather than an afterthought.
Our role
Cosmic Limit was responsible for requirements analysis, solution architecture, backend development, web frontend development, database design, systems integration, testing, application deployment, and ongoing system maintenance.
The infrastructure the system runs on is provided and managed by the client's own infrastructure team. We designed and delivered the application and we continue to maintain it; we do not own or operate the environment underneath it, and the security restrictions that environment imposes were a constraint we worked within rather than a boundary we set.
The solution
The core system is a modular monolith -- one deployable application with firm internal boundaries -- supported by a small number of focused microservices, including a notification service. That shape was chosen deliberately over a fully distributed one: the domain is highly transactional and heavily cross-referential, and the operational cost of distributing it would have been paid by a team that does not need to pay it.
Auditability and traceability are core design principles rather than a feature that was added. Approval workflows are enforced by the system, so a step cannot be skipped by agreement, and the trail of what happened to a record is a by-product of how the system is built.
The functional areas delivered are budgeting, approval workflows, reporting, user and role management, document management, notifications, loan tracking, and automatic interest calculation. Loan tracking with automatic interest calculation is the most distinctive of these: interest is derived by the system on a defined basis rather than maintained by hand in a parallel spreadsheet, which is exactly the class of calculation that drifts when it is done manually.
Two things are deliberately not in scope: there is no procurement module and no payment-processing module. The system records and controls financial process; it does not move money.
Implementation is .NET for backend services and business logic, Next.js for the web frontend, and PostgreSQL as the primary relational database. RabbitMQ carries asynchronous communication between services and the background processing of time-consuming jobs -- report generation, email sending -- so those never block a user's request. Redis provides caching. Everything is containerised with Docker and runs on Linux.
Outcome
No numerical results are published here. The measured outcomes that would support figures have not been verified for publication, and stating them qualitatively is the honest form rather than a weaker one.
What changed: manual errors are fewer, because a figure is entered once into a structured workflow rather than re-keyed between files. Auditability is stronger, and approval workflows are enforced rather than assumed. Reporting is faster, because reports are produced from a single source of truth instead of assembled from separate spreadsheets, and information can be found without tracking down whoever holds the current version. Automated reminders keep work that is waiting on someone moving. Data retention improved and the risk of losing records outright dropped, because financial data now lives in a managed database rather than in files held by individuals -- which is the difference between record keeping that is reliable and record keeping that is habitual.
Constraints
What shaped the solution
- integration
The system had to work with external systems outside our control. Those systems had outages, changed on their own schedule, were reachable only over poor connectivity, and offered us limited access to work against.
- infrastructure
The application runs inside client-controlled infrastructure with strict security restrictions, which shaped how we could reach the environment, deploy into it, and diagnose anything inside it.
- data quality
Legacy spreadsheet data had to be migrated: inconsistent formatting, inconsistent records, the same person spelled several different ways. We cleaned and standardised it before migration rather than carrying the inconsistencies into the new system.
- stakeholder
Decision-making was slow and key users had limited availability, which lengthened every round of clarification, validation and approval.
Technology
What was used, and for what
framework
.NET
Backend services and business logic.
Next.js
Web frontend.
datastore
PostgreSQL
Primary relational database.
Redis
Caching.
integration
RabbitMQ
Asynchronous communication between services, and background processing of time-consuming jobs such as report generation and email sending.
infrastructure
Docker
Containerisation and application deployment.
platform
Linux
Production runtime environment.
Outcome
What changed
Fewer manual errors
Structured workflows replaced spreadsheet entry, so a figure is entered once and carried through rather than re-keyed between files.
Enforced approvals and a real audit trail
Approval workflows are enforced by the system rather than agreed by convention, and what happened to a record can be established afterwards.
Faster reporting, quicker access
Reports are produced from a single source of truth instead of assembled from separate files, and information can be found without asking whoever holds the relevant spreadsheet.
Automated reminders
Work waiting on someone is chased by the system, so approvals move without a person having to remember to follow them up.
Reliable record keeping and lower risk of data loss
Financial records live in a managed database with consistent structure and retention, rather than in files held by individuals.
Our role
What we were responsible for
Anonymised by owner decision recorded in the launch content handoff of 2 September 2026. No screenshots or diagrams are published: any future image needs sanitising and a confidentiality review first (spec 22.2, 26).
Facing something similar?
Tell us the constraint you are working against and we will tell you how we would approach it.