Manual work everywhere. AI in strategy. Little in control.
Most mid-tier firms now have AI in their strategy, and staff say manual work is their biggest frustration. The risk is AI spreading through working papers faster than anyone can show who checked it.
of mid-tier firms include AI in their strategy
ICAEW, 2026of accountants report excessive manual tasks
Sage, 2024 to 2025, vendor surveyof firms can currently assess AI's impact on their workforce
ICAEW, 2026Professional duties do not change because a tool was used.
The Financial Reporting Council (FRC) published guidance on generative and agentic AI in audit on 30 March 2026. It is principles-based: firms stay accountable for audit quality, and the depth of validation depends on the tool and its use. In January 2026 seven professional bodies, including ICAEW, ACCA and the Chartered Institute of Taxation, issued guidance showing how the tax ethics code applies to AI. Their message is that the principles are the same whether or not AI was used.
Summarised from the FRC (30 March 2026) and the joint professional bodies' guidance (January 2026). The FRC guidance is aimed mainly at the central technical teams of audit firms; read the PDF at source. Not legal advice.
Six problems that keep coming back.
Drawn from the FRC and professional bodies' guidance and from surveys of the profession.
Audit, tax and accounts files increasingly contain text or analysis an AI produced. A reviewer cannot tell what was generated, what was checked, or against what.
Client ledgers, payroll and tax details pasted into free assistants break confidentiality, and the engagement letter rarely covers it.
AI speeds first drafts, so the review queue grows. Quality depends on the reviewer knowing what to look for and recording that they looked.
Audit, tax and advisory each buy their own AI. Costs add up, and no one can describe the firm's AI use to a client or the quality team.
AI helps with onboarding and monitoring checks, but the firm must still verify, decide and keep the record.
Clients ask what part of their work AI touched. Without records, the honest answer is 'we think none'.
A quality review finds what your records cannot explain.
Inspections and internal quality reviews look for evidence that a competent person reviewed the work. If AI helped produce it and nothing shows who checked what, the file looks weaker than the work was. Beyond inspections, a single client data incident can end an engagement, and fee pressure continues while competitors use AI openly.
Quality management already has a place for AI. It needs records.
Firms work under quality management standards that expect risks to be identified, controls to be designed and operation to be evidenced. AI is one more risk in that system. A register of tools, rules on client data and a log of reviews let the quality team treat it the same way as any other control.
- Bookkeeping, reconciliation and invoice processing
- Audit sampling and working papers
- Tax research and drafting
- Client reporting and management accounts
- Client financial data in consumer AI tools
- Working papers with unattributed AI content
- Duplicate tools across service lines
GenWA puts managed accounts and client-data redaction on connected tools, records who accepted each AI output for attachment to the file, and produces Evidence Packs for quality review. Skill Teams for invoice processing and reconciliation, and a shared Library across service lines, are being built next and are not live.
Four agents, trained for this sector.
The rules, as controls.
- FRC guidance on generative and agentic AI in audit (March 2026)
- Tax ethics code guidance on AI (January 2026)
- Money Laundering Regulations 2017
- Client confidentiality
- Firm quality management standards
- UK GDPR and the Data (Use and Access) Act 2025
Built around partnership models and service lines.
Accountancy firms decide by partner vote and often run several service lines. We shape each route to that.
An independent audit by service line and a short briefing for the partners, so one decision serves audit, tax and advisory.
Start the health checkAdvice: policy and quality systemAn AI policy and review rules that fit your quality management system, with training for reviewers.
See advisoryProduct: control toolkitManaged accounts, redaction, rules, log and Evidence Packs, hosted in your AWS or Azure account, or GenWA AWS as agreed.
Find the right productProduct: agents at workInvoice processing, reconciliation checks and client communications, each with a named reviewer.
Meet the teamIncubator: one service line functionWe define, build and run one controlled function, such as accounts production support, in twelve weeks. You own the intellectual property.
See the incubatorAccelerator: for practice-technology foundersBy application only, ten places a year. Senior AI leadership time, advisory and platform access, paid for by equity, project, retainer or licence.
See the acceleratorWhat Guard and Evidence typically hold for this sector.
Control-mapped, not certified. We prepare the evidence; regulators and accredited bodies decide.
March 2026 principles on validating AI outputs, mapped to review and sign-off records.
January 2026 ethics guidance for tax work, mapped to integrity, confidentiality and due care.
Client due diligence and record-keeping, mapped to onboarding checks.
Engagement terms and ethical duties, loaded as rules on what client data may leave the firm.
Data protection and automated-decision safeguards, with redaction evidence.
AI management, information security and baseline cyber standards that clients ask for. Readiness evidence only.
Three role holders, three different problems.
Wants growth without a client data incident. GenWA gives one picture of AI use across service lines and a decision the whole partnership can stand behind.
Must show that reviewers checked AI-assisted work. GenWA records who accepted each output and produces evidence packs for quality review.
Pays for overlapping tools and wants capacity back. GenWA shows spend by team, consolidates tools and automates invoices and reconciliations under a named reviewer.
From this sector.
How can a firm show who reviewed AI output in working papers?
The log records what was asked, which model answered and who accepted it. That record can be attached to the file, and Evidence Packs support quality review. Reviewers still decide whether work can be relied on.
We already use Microsoft Copilot or a supplier's AI. Is that covered?
Existing tools stay in place and you keep your licences. GenWA puts rules, redaction and the log in front of connected tools. Coverage depends on the integrations scoped in the audit.
What do the FRC and professional bodies expect on AI?
The FRC's March 2026 guidance says firms remain accountable and validation should fit the tool. Seven professional bodies said in January 2026 that the tax ethics principles apply whether or not AI was used. Read both at source. Not legal advice.
How do we keep client data in approved tools?
Use managed accounts with client-data redaction, rules on what may be shared, and Usage Insight to see exceptions. Redaction applies to connected, routed tools.
How many accountancy firms include AI in their strategy?
ICAEW's 2026 research found 86% of mid-tier firms do, but only 17% can currently assess AI's impact on their workforce. Sage's 2024 to 2025 vendor survey found 92% of accountants report excessive manual tasks.
Are Skill Teams for invoices and reconciliation available?
Not yet. They are being built next, as is a Library across service lines. We do not describe them as live, and confirm status in writing before you commit. The Log, Rules and Evidence Packs are available now at their stated release status.
How does a firm start?
Take the AI health check and name the service lines where AI is used. First findings arrive within 48 hours and the audit with a 90-day plan within ten working days. The audit fee is credited in full against year one of a platform plan.
Can you show which working papers contain AI content?
First findings within 48 hours of the health check. Audit fee credited in full against year one of a platform plan.