How Becide works

Becide is CFO decision intelligence for finance teams. It combines a managed, tenant-scoped financial data lake with an AI analyst experience: it drafts your monthly management report from live, curated data, then answers board follow-up questions in seconds — every figure cited back to the query that produced it.

The core idea is simple: we fix the data first. Your source systems — general ledger exports, billing, payroll, spreadsheets — are ingested, cleaned, and curated into one governed financial picture. Becide is the product layer on top: it reads that governed data, never writes to it, and explains every number it presents.

What Becide is not

• Not a replacement ERP or general ledger — your systems of record stay where they are.
• Not a DIY data-lake project — ingestion, cleaning, and curation are managed for you.
• Not a chatbot bolted onto raw files — the AI only reads governed, curated layers.
• Not an overseas generic AI wrapper — data is hosted and processed onshore in Australia.

Becide is built for finance leaders at growth companies who need board-ready numbers without building a data team. Each role interacts with the same governed data in a different way:

A typical onboarding takes days, not months, because the data foundation is provisioned as part of onboarding rather than built as a project:

• Week 1 — Your tenant is provisioned with its own isolated storage, encryption key, and workspace. You connect your first data source.
• Week 1–2 — Your data flows through the raw, clean, and curated layers, and a knowledge base is built that maps business questions to your real tables.
• Week 2 — You generate your first management report from live workspace data and start asking follow-up questions.
• Ongoing — The pack regenerates on your reporting cadence; the assistant handles ad-hoc questions between cycles.

Request access to get started.

Management Report generates your monthly management pack from curated workspace data: profit and loss, variance to budget, segment revenue, overheads, balance sheet, and cash flow where the data exists — laid out in a board-ready format with an executive summary and key business drivers.

What makes it defensible

• Methodology on every section — each KPI, chart, and table explains how it was calculated and what to read from it.
• Traceability — figures link back to the source queries that produced them, so numbers hold up with the board and auditors.
• Workspace context — standing report instructions and section preferences apply automatically to every generation.
• Refinement — regenerate or adjust the pack with follow-up instructions instead of rebuilding it from scratch.
• Honest empty states — if a connector is missing or a schema is insufficient, the report says so; it never fills gaps with invented data.

Ask Assistant answers ad-hoc CFO questions in plain language over the same governed data: “Are we ahead or behind revenue plan for the quarter, and why?” “Show gross margin by product line for the last four quarters.”

Every answer follows the same discipline:

• Multi-step reasoning you can inspect — the assistant plans, retrieves, analyses, verifies, and then synthesises, and shows that thread.
• Citations and a methodology note on every answer, with charts where they help.
• Read-only access — the assistant queries governed layers only and can never write to your source systems.
• Workspace scoping — answers draw only on the connectors and knowledge base of the active workspace.
• Follow-up conversation in the same thread, with suggested next questions.

Typical answers arrive in seconds — the difference between answering a board question before the next agenda item and taking it on notice for a week.

Data enters your tenant through connectors. Today that means file uploads (CSV) and direct connections to your existing SQL data warehouse; additional accounting, revenue, and document connectors are on the roadmap.

• Every connection is tested before it is saved, with schema discovery and clear error reporting.
• Connector health is visible at all times — broken or untested connections are flagged, never silently ignored.
• Connector credentials are stored in a dedicated secrets vault, encrypted, never shown back to users, never written to logs, and never included in anything sent to the AI.
• Original files are preserved untouched in the raw layer, so there is always a verbatim record of what was provided.

Organisations with multiple entities manage each one as a workspace. Connectors, knowledge bases, report preferences, and conversations are scoped per workspace, so a question asked in one entity never draws on another’s data.

Access is role-based: administrators manage connections and settings, finance roles work with reports and the assistant, and executive viewers see outputs without configuration access. Permissions apply per module, so you control exactly who can connect data, generate reports, or ask questions.

Becide runs entirely within a single Australian cloud region (Sydney). File storage, databases, search, AI processing, and audit logs all stay in that one region — nothing is processed offshore. The platform is organised into four layers, each with a narrow job:

Design principles

• Separation of concerns — the layer that talks to users never touches raw data directly, and the layer that runs AI holds no credentials or persistent state.
• Least privilege — every component gets the minimum access it needs; the AI layer gets read-only access to governed layers and nothing else.
• Statelessness — AI runs are independent, disposable executions; progress is recorded in encrypted records, so no single machine ever holds your session hostage.
• Fail loudly — missing data, broken connectors, and failed checks surface as visible states, never as silently fabricated output.

Your data lake follows a three-layer refinement model. Data moves one way — from raw to clean to curated — and access narrows at each step:

“Cleaned and made sensible” means revenue, costs, cash, and receivables land in consistent shapes across all your sources, and a knowledge base maps business questions to the real tables that answer them. You never manage schemas, storage layouts, or query engines — that is the managed part of the service.

When you ask a question or generate a report, here is what actually happens:

• 1. Your browser sends the request with your authenticated session. The application verifies your identity and resolves your tenant and workspace before anything runs.
• 2. A run record is created with a unique ID — every AI execution is tracked from the moment it starts.
• 3. The orchestration layer plans the work, then executes a reasoning loop: retrieve, analyse, verify, synthesise. Each step uses only approved read-only tools scoped to your tenant.
• 4. Personally identifying fields such as emails and phone numbers are redacted before content reaches the language model, and the model receives only what the specific question needs.
• 5. Progress streams back to your screen as it happens. The final answer arrives with citations and a methodology note.
• 6. The complete run — question, tools used, safety checks — is written to the encrypted audit record.

Your own encryption key

Every client is issued a dedicated encryption key at onboarding. That key lives in an isolated key vault — a separate, hardened service that is not the database and not the file store — ensuring key isolation from the data it protects. Becide never stores or handles the key material directly: the platform can request that the vault encrypt or decrypt on its behalf at processing time, but no person — including our administrators — can read or export the key itself.

Because each client has a separate key, one client’s data can never be decrypted under another client’s key. And because key destruction is part of offboarding, deleting your key renders your data cryptographically unreadable everywhere — including in backups.

Encryption everywhere

Personal information

Personal information is handled under the Privacy Act and the Australian Privacy Principles. It is encrypted, access controlled, and never used to train AI models. The AI receives only what a specific question needs, on demand, with identifying fields redacted first. Read the Privacy Policy for the formal commitments.

Isolation is enforced at every layer independently, so no single failure can expose one client’s data to another:

• Identity — your tenant is resolved from your verified login token on every request; it is never taken from the request body where it could be tampered with.
• Storage — files live under tenant-specific prefixes; operational records are partitioned by tenant.
• Search — each tenant has its own search index; the knowledge base applies a tenant filter on every query.
• Encryption — a dedicated key per tenant means that even a hypothetical storage-level fault would yield only ciphertext unreadable without your specific key.
• AI runs — every tool call is scoped to the requesting tenant and workspace; there is no query path that crosses tenants.

The AI operates under a fixed, enforced contract — not a polite instruction, but permissions it structurally cannot exceed:

Every AI run — every report generation, every question — is recorded in Australia with a unique run ID, the tools it used, and the safety checks that were applied. Questions are stored encrypted, not in plain text.

• Tenant users can view their organisation's activity log inside the product.
• Figures in reports link to the queries that produced them, so any number can be traced end to end.
• Your organisation can request access to its full activity records through our privacy contact.

You own your data and everything Becide produces from it. Retention is governed by your intent, not ours:

• You can export or download your original files at any time — they are preserved untouched in the raw layer.
• You choose how long we retain data, per your agreement.
• When you leave, you can request deletion with written confirmation. We destroy your dedicated encryption key, which renders your data unreadable everywhere — including backups.
• Inactive accounts are never silently deleted: after 18 months of inactivity we contact you with three notices before deactivation, followed by a 180-day restore window.

General-purpose assistants like ChatGPT or Claude are remarkable tools — and the wrong tool for board reporting. Pasting financial exports into a generic chat gives you an eloquent answer over stale, unreconciled data with no citations, no tenancy guarantees, and no audit trail. The difference is not the model; it is everything around it:

Can the AI change or delete my data?

No. The AI has no write tools — not to Becide’s layers and not to your source systems. It can only run approved read queries scoped to your tenant.

Where is my data stored?

In a single Australian cloud region (Sydney). Storage, search, AI processing, and audit logs all stay onshore.

Can Becide staff read my data?

Your data is encrypted with a key held in an isolated vault that no person can read or export. Access is permission-based, logged, and limited to what operating the service requires.

Is my data used to train AI models?

No. Your data, questions, and outputs are never used for model training.

What happens if I leave?

You export what you need, request deletion, and receive written confirmation. Destroying your dedicated encryption key makes remaining data unreadable everywhere, including backups.

How do I trust a number in a report?

Follow it. Every KPI, chart, and table carries its methodology, and figures link to the source queries that produced them — defensible with your board and your auditors.