BudgetIQ AI: Building a Continuous Budget Cycle

BudgetIQ AI is an evolving full-stack budgeting platform for institutions that need to connect plans, procurement comparisons, funding decisions, commitments, financial records and next-year estimates. Its core idea is simple: every planned activity should retain a clear path from objective to cost, review, actual result and future learning.

A budget is often prepared as an annual document, while the evidence needed to judge it lives in separate workplans, spreadsheets and expenditure records. BudgetIQ explores a more continuous process for agencies, businesses, schools, churches and nonprofits. The original product question was practical: could someone upload an action plan and prepare an editable, detailed budget even with little or no financial history? A previous budget, revenue and expenditure records could then improve the next planning cycle. A working institutional budget template helped shape the programme, activity, funding and costing relationships; its historical prices were never intended to become current default rates.

How budgeting research shaped the product

Bismark Adams's thesis, The Role of Data Analytics in Improving Budgeting and Forecasting Accuracy in Organisations (manuscript shared with the project), provided a research lens for this practical idea. Using secondary Ghana budget statement data covering 2020 to 2025, the study examines fiscal trends, differences between original budgets and revised estimates, and how selected annual forecasting methods behave. BudgetIQ is an independent application that applies and extends some of those questions to an institution's own planning workflow; the thesis itself did not develop or validate this app.

The research compares linear regression, simple exponential smoothing and Holt's linear trend for a short revised revenue series. Its expanding one-year-ahead evaluation used only two held-out years. The trend methods performed better than simple exponential smoothing in that particular comparison, but two tests on revised estimates cannot establish general forecast accuracy or performance against actual receipts. This finding led to a conservative product rule: show the method, the available years, the evidence stage and prior errors, and keep the budget officer in control of the decision.

This connection is a starting point for collaboration with the researcher and finance practitioners. The thesis studies national fiscal aggregates, while BudgetIQ needs evidence at the organizational, programme and activity levels. Results from one scale should not be copied into another as a validated model.

One connected budget, from plan to evidence

The platform links objectives → programmes → activities → cost lines → financial results. An institution can turn a programme goal into specific activities, cost each activity in detail, compare the plan against available funding and then record what was received, committed and paid. That structure makes a budget useful beyond its grand total: a reviewer can ask which activity a cost supports and how an estimate compares with later evidence.

For example, a fictional Community Learning Centre might plan a staff training programme and a quarterly outreach series. For each activity, a planner can enter frequency, quantity, unit rates and funding source across transport, venue, materials, refreshments, allowances and other relevant lines. A ceiling comparison then shows whether the complete plan fits the available allocation. These example activities are illustrative; the same model can be adapted to other institutional settings.

What the current MVP can do

An organization can create an account and work with its own saved projects through a backend service and PostgreSQL database. The frontend and API run as separate services on Render. Passwords are hashed, requests use account authentication and project data is stored server-side. This replaces browser-only project storage, although the current account and security model still needs further review before sensitive operational use.

• Start from a plan: import text-based action plans from supported document and spreadsheet formats, review extracted suggestions and edit the resulting activities.
• Build a detailed budget: organize objectives, programmes and activities, then adjust quantities, rates, frequency and cost lines.
• See constraints: compare totals against funding ceilings and identify gaps; create alternative scenario arithmetic for different funding and price assumptions.
• Track the cycle: record projected and received revenue, commitments and payments by quarter, with budget-versus-actual views.
• Review purchases and commitments: compare recorded supplier offers, obtain an internal recommendation, control requests against an activity and funding release, and record invoice and payment references.
• Keep a review history: save plan snapshots and planning notes; Treasury and Procurement also record account-linked decisions.
• Share outputs: export budget information in PDF, Excel or CSV for review outside the app.

Document extraction currently relies on transparent rules and editable suggestions. A user should check every imported activity and rate. The general budget review log remains self-reported; the newer Treasury and Procurement workspaces enforce specific account roles and decision sequences on the server. Neither constitutes an official institutional sign-off.

Separating the stages of financial evidence

A crucial design decision is to distinguish an original approved budget, a revised estimate, a provisional result and an audited actual. These figures answer different questions. Comparing the revised estimate with the original budget describes changes to the plan; comparing an audited actual with the approved budget describes execution. Treating these stages as interchangeable can make a forecast look more accurate than it really is.

The Evidence workspace accepts annual financial records and checks imported CSV or Excel data for issues such as duplicate years, missing audited years, inconsistent currency or invalid amounts. It preserves plan snapshots and shows revision and execution variances separately. A missing result remains missing; the system should not silently replace it with a planned amount.

Forecasts should explain themselves

BudgetIQ includes a next-year draft informed by recorded spending and an evidence view that compares four simple annual forecasting methods: last actual, linear trend, simple exponential smoothing and Holt’s trend. Where there are enough consecutive audited annual results, expanding one-year-ahead checks report error measures including MAE, RMSE, MAPE and bias. A small or interrupted series is flagged as preliminary. These checks do not establish that any method will predict future funding or expenditure reliably.

Every suggested figure should make its origin understandable: a user-entered rate, a prior approved budget, actual spending or a stated assumption. The intended direction is to show uncertainty and weak evidence clearly, let a budget officer adjust the result and preserve the reasoning behind that adjustment. The goal is better judgment, not automatic approval.

The next research step is a permitted, carefully documented dataset: several years of approved and revised budgets alongside actual receipts and payments, with consistent years, units, funding sources and categories. Activity-level costs, quantities, procurement rates and quarterly results would make the planning question more useful to test. The team should check provenance and access rights, remove personal or confidential details as appropriate, reconcile totals with source records and hold later periods aside for honest evaluation. Until that evidence is available, the app's estimates remain editable planning aids rather than claims of accurate prediction.

Procurement comparison linked to the plan

A project owner can open an internal procurement case against a costed activity, record its estimate, funding source, quarter, proposed sourcing method and reason. An invited procurement account enters supplier offers with their references, prices, responsiveness, technical scores and evaluation notes. The evaluator declares that they have no conflict of interest before recommending a responsive offer with a written rationale. If only one offer is available, the case requires an explanation. A separate approver can return or approve the recommendation.

After internal approval, the project owner can create a linked draft Treasury request with the selected supplier and quoted amount. That handoff does not reserve funds or award a contract: the Treasury request must still pass its own funding and approval checks. The application does not decide whether a sourcing method is legally appropriate, verify supplier eligibility, publish tenders or issue a purchase order. Offer references and external system references are entered by users, not independently authenticated.

Treasury-style commitment control

In the Treasury workspace, the project owner records an internal funding release by source and quarter. A proposed expenditure request must fit within the project ceiling, the linked activity cost and the relevant source-quarter release. Once submitted, it reserves that capacity; a rejected request releases it. The owner cannot review their own request, a separate reviewer must assess it, and a different approver must decide before the owner records a commitment. Account-linked events retain the actor, action, time and note.

A recorded invoice must have a delivery or service-completion reference and cannot exceed the commitment. Payment references can be entered in parts, but their cumulative amount cannot exceed the matched invoice. These are user-entered records: the system does not verify delivery, certify an invoice, transfer funds or reconcile a bank statement. The older Execution tab is a separate planning ledger and is not automatically reconciled with Treasury, so its entries should not be added to Treasury totals as if they were one financial ledger.

Designed for review and responsible adoption

The new roles and checks are a working foundation, not a complete institutional control framework. The roadmap still includes verified supplier records, controlled evidence attachments, legally competent approval paths, stronger identity checks, versioned submissions, reversals and period close, comprehensive audit review, organization-level administration and approved integration pathways. Scenario comparison should help a team discuss a full-funding plan, a ceiling-constrained plan and higher-price assumptions without overwriting the approved baseline.

Before a real institution relies on the platform for confidential records or official decisions, the product requires security assessment, backup and recovery checks, access and retention policies, calculation reconciliation, data-quality review and a controlled pilot with qualified finance and procurement staff. Forecast performance should be monitored against later audited results, not judged only from a small historical sample.

Where collaborators can contribute

BudgetIQ is open to focused input from budget officers, accountants, procurement specialists, programme managers, data scientists, security engineers and product designers. Useful contributions include testing activity-costing categories across sectors, defining practical funding-ceiling decisions, reconciling exports against source records, reviewing access controls, and designing forecast validation with sufficient audited data.

The immediate development question is whether a team can take a real action plan, prepare an understandable budget, revise it responsibly and later explain the gap between the approved figure and the result. If that cycle works reliably, more advanced forecasting has a sound foundation.

Public guidance informing the controls

• Public Procurement Authority: GHANEPS use for public procurement ↗
• Ministry of Finance: commitment control guidance ↗

A useful budget remembers both what an institution planned and what actually happened.