Security & data handling

Your financial data stays yours, and nothing moves without a name on it.

Accruna holds access to a company's most sensitive operational data. That deserves specifics rather than reassurance, so this page sets out the access model, the approval controls, the audit trail, and — just as importantly — what we are not currently claiming.

The short version

  • Least-privilege access for people and agents alike
  • Human approval required for payments, journals and policy overrides
  • Immutable audit trail with before-and-after values
  • Encryption in transit and at rest
  • No sale, sharing or model training on your data
  • Full export and credential revocation on exit

Control framework

Five controls the engagement depends on.

These are operational commitments, not aspirational statements. Each one has an owner and a way to verify it.

Encryption everywhere

Connections use TLS. Ledger data, documents and stored credentials are encrypted at rest. API tokens are held encrypted and are excluded from application logs.

Role-based, least-privilege access

Access is scoped by role, entity and function. Agents hold the narrowest permissions their task requires — a coding agent cannot initiate a payment.

Immutable audit trail

Every action records the actor, timestamp, affected record and the values before and after. Entries are reversed, never deleted, so history stays intact.

Human approval controls

Payments, journal postings and policy overrides require a named human approval within an authority limit you set. Agents cannot approve their own proposals.

Separation of duties

Preparing, approving and posting are separated. Supplier bank detail changes require dual authorisation; period reopening follows a documented process.

Observable operations

Sync health, rejected rows, exception ageing and approval history are visible to you at all times. You should not have to ask whether something ran.

Access model

Who can see and do what.

Access is granted by role and reviewed on a schedule. Standing broad access is not granted to anyone, including our own staff.

Access permissions by role
Role Ledger access Can approve Can move money
Accruna agentsScoped to a single function, read plus narrowly defined writesNoNo
Engagement accountantFull read, write on assigned entitiesWithin their authority limitNo
Accounting managerFull read, write on assigned portfolioYes, within portfolio limitsNo
Client controllerFull read and write on their entitiesYes, per your matrixVia your bank authorisation
Client adminFull read; configuration rightsYesVia your bank authorisation
Department headRead scoped to their cost centreWithin their budget authorityNo
External auditorRead-only, time-boxed, revoked on requestNoNo

How access is granted

By role, with a named approver, and with an expiry for temporary grants. Delegation carries a date range rather than being indefinite.

How it is reviewed

A permission matrix is reviewed with you periodically, and an exportable version is available at any time. Departures trigger immediate revocation.

How it is verified

Every access event is logged. You can request a report of who accessed what and when, for any period, including our own staff.

Agent security

Autonomy needs boundaries, not trust.

Agents are treated as non-human identities with explicit authority limits. They are not given broad credentials and asked to behave — their permissions are scoped so that a mistake cannot become a large one.

Scoped credentials
Each agent holds its own credentials, limited to the systems and record types its task requires. A coding agent's credentials cannot reach a payment rail.
Authority limits
Value bands, category restrictions and confidence thresholds bound what an agent may post automatically. Anything outside those bounds stops and queues.
No self-approval
An agent may propose an entry; it may never approve one. Approval requires a human identity, and that identity is recorded on the entry.
Bounded retries
Failed operations retry a limited number of times and then fail visibly. There is no unbounded background loop that could post repeatedly.
Full attribution
Every agent action records which agent, which version of its instructions, and which rule or precedent it applied.
Audit log · entry detail Attributed
  • EntryJE-2025-09-0442
  • Proposed byLedger agent v4.2
  • BasisRule R-0142 + 14 precedents
  • Approved byM. Okonjo, CPA
  • Approved at28 Sep 2025, 13:22
  • Value$14,820.00
  • ReversibleYes — reversal preserves original
An auditor can follow this chain from a reported figure back to the source document without asking us for anything.

Why this matters in practice

Most automation risk is not a model being wrong. It is automation having more authority than the situation warranted, and no record of what it did. Both are design decisions, and both are avoidable.

Data lifecycle

What happens to your data at each stage.

From the first connection to the day the engagement ends — stated in advance rather than negotiated at exit.

1

Collection

Only the data required for the engagement. Connections you authorise, scoped as narrowly as workable.

2

Processing

Encrypted at rest, access logged, and excluded from any third-party model training.

3

Retention

Held for the engagement and the agreed retention window that follows, driven by record-keeping obligations.

4

Return and deletion

Full export on request, credentials revoked, then deletion with written confirmation.

Sub-processors

Third parties in the chain

Running this needs hosting, storage and connection infrastructure. We will give you the current list of sub-processors on request, tell you before a new one is added where it affects your data, and we do not permit sub-processors to use your data for their own purposes.

  • Named list available on request, with purpose and location
  • Advance notice of changes affecting your data
  • Contractual restrictions on use, retention and onward transfer

Your rights

Control you keep at all times

  • Access report covering who viewed or changed your data
  • Export of ledger history, workpapers and audit logs
  • Revocation of any connection, effective immediately
  • Deletion request subject to record-keeping obligations
  • Written confirmation when revocation and deletion complete

What we do not claim

Being specific about the limits.

Compliance claims are easy to make and expensive to verify. These are the things we are deliberately not asserting, so you can assess the engagement without having to discount for marketing.

Certifications
We do not currently hold or claim a SOC 2 report, ISO 27001 certification or equivalent. Our controls are designed with those frameworks in mind, and we will say plainly when that changes rather than implying it now.
Regulatory status
Accruna is not a licensed audit firm and does not issue audit opinions. We prepare and maintain the workpapers an auditor will request; the opinion comes from your auditor.
Filing
Statutory tax filing is not part of the core engagement. We maintain the ledger and schedules that filing depends on, and can scope filing separately with the appropriate licensed preparer.
Guarantees
We do not publish accuracy percentages or close-time guarantees. What we commit to is a documented process, named accountability and service levels we are willing to be measured against.
Data residency
Default hosting region is stated during scoping. Specific residency requirements can usually be accommodated, but we will confirm rather than assume.
If a requirement on your security questionnaire is not met, we will say so. Where it is a genuine gap, we would rather you discovered it during evaluation than during onboarding.

Operational practice

How the firm operates day to day.

Security is mostly operational discipline. These are the practices behind the controls above.

Identity and access

  • Multi-factor authentication required for all staff accounts
  • Access granted by role request with a named approver
  • Immediate revocation on departure or role change
  • No shared accounts; every action is attributable to a person

Change and review

  • Rule and threshold changes versioned with an effective date
  • Period locks applied at close and reopening documented
  • Peer review on journal entries above a materiality threshold
  • Quarterly review of access, thresholds and exceptions together

Continuity

  • Backups taken and restoration tested, not merely configured
  • Documented process for a period in which systems are unavailable
  • Named escalation contacts on both sides of the engagement
  • Exportable data so continuity does not depend on us

Encrypted

In transit and at rest

Ledger data, documents and credentials.

Logged

Every action

Actor, timestamp and before/after values.

Approved

By a named human

No agent approves its own proposal.

Security questions

Questions we are asked in reviews.

Send us your security questionnaire and we will complete it honestly, including the parts we cannot yet satisfy.

The accountants and managers assigned to your engagement, and the agents performing scoped tasks within it. Access is role-based, so a team member working on another client cannot see your data, and no one has standing access to all engagements.

You can request an access report for any period showing who viewed or changed what, including our staff. Where a client prefers tighter scoping — for example restricting payroll visibility to one named manager — that is a configuration we set at onboarding.

No. Agents prepare payment batches and match supporting documentation, but releasing funds requires a human authorisation using your own banking credentials and controls. Accruna never holds the authority to move money unilaterally.

This is a deliberate design boundary rather than a current limitation. Even if a future capability allowed it, the authority limit would remain a configuration you control.

No. Your ledger data is not used to train third-party models, and it is not used to train models for other clients.

What does happen is that coding decisions you approve become precedents within your own ledger — so the agent's suggestions for your accounts improve over time. That precedent set is client-specific and never shared.

It is identified, reversed and corrected with the original preserved. Because every entry carries its source, its basis and its approval, tracing the cause is a lookup rather than an investigation.

If a pattern of errors emerges, we will tell you and adjust the thresholds or rules that produced it. We would rather narrow an agent's authority than defend its output.

On cloud infrastructure, in the default region stated during scoping. Specific residency requirements are best raised early — they are usually accommodating, but they change architecture and should not be a surprise midway through contracting.

Sub-processors involved in hosting, storage and connectivity are listed on request, and you will be told before a change that affects your data.

You can request documentation of our controls and an access report, and we will complete security questionnaires in full. On-site or remote control audits are available on larger engagements and are agreed in the engagement letter.

If a control you require does not exist yet, we will tell you that directly. Discovering a gap during evaluation is considerably cheaper than discovering it during an external audit.

Your ledger, documents and history remain in your systems throughout, because Accruna operates on your data rather than holding a separate copy of the business's books. Connections are revoked promptly after the final period is signed off.

Any working data we hold is exported to you on request and deleted according to the agreed retention window, with written confirmation when both are complete.

Send us your security requirements.

If your organisation needs a questionnaire completed, a specific control documented, or a residency commitment confirmed, we would rather work through it before you shortlist us.