Every few weeks another announcement lands from one of the Big Four about an AI platform, an audit tool, or a generative AI assistant handling payroll tax queries at global scale. For a practice with ten staff and 200 clients, those announcements can feel like news from a different industry. Some of that distance is real, particularly at the level of custom development and proprietary platforms. At the use-case level, the same capabilities have filtered into specialist tools that smaller practices can actually reach.
What AI use cases have actually landed in accounting firms?
Document extraction, bank reconciliation, and tax research assistance are the three use cases with the clearest track record across the profession. Accounting software already handles much of the base automation; the AI layer reduces the manual work that still surrounds those processes. The Big Four provide the most visible examples, but specialist tools now make similar workflows accessible to smaller practices without requiring bespoke development.
EY has been testing a generative AI system to answer complex payroll tax queries, drawing on a large corpus of tax law to improve speed and consistency for cross-border workforce cases. Deloitte has built generative AI and agentic capabilities into its audit platform to perform initial document reviews and suggest improvements to audit documentation. These are large-scale, custom deployments. The underlying capability, AI extracting amounts, dates, and counterparties from invoices and linking them into working papers, is now packaged in tools like DataSnipper that mid-sized and smaller firms use daily.
Thomson Reuters’ 2025 survey found that among accounting professionals actively using generative AI, 44% were doing so daily or multiple times daily, with a further 29% using it weekly. The tasks they reported are practical: document summarisation, drafting client correspondence, querying tax content, and reconciling accounts. ICAEW has identified eight specific use cases, including translating financial information, assisting with GAAP and IFRS queries, and summarising board minutes.
AP automation is also well established. Vic.ai handles invoice processing and provides financial insights using machine learning. Cash-flow forecasting tools ingest historical data and flag early-warning indicators for stress in accounts receivable. These are working workflow tools, not early-stage experiments.
Why do these use cases matter for a practice your size?
The productivity gains reported from AI-assisted document review, with some early adopters citing over 40% time savings on specific audit tasks, apply to the underlying workflow rather than to any minimum scale of operation. A practice that spends significant time on manual vouching or invoice matching stands to benefit from the same AI layer, even if the absolute savings are smaller.
Thomson Reuters’ survey suggests the shift is already pronounced. Four in ten accounting professionals using generative AI are doing so every single day. That rate of adoption, across a regulated profession where the consequences of errors are material, signals that the tools have moved past early adopter novelty into day-to-day use.
Client expectations are also shifting. Larger practices that automate the low-value work can turn results around faster and, over time, at lower cost. Smaller practices will notice the competitive pressure this creates, as they have noticed previous waves of technology change in the profession.
The economics need modelling carefully before committing. AccountingWEB guidance for UK firms notes that subscription costs, setup time, staff training, and the period of running parallel processes all need to factor into the calculation. For very small practices with limited transaction volumes, a phased approach starting with the single most time-intensive manual workflow tends to produce a clearer return than a broad rollout.
Where will you actually encounter AI in a practice’s workflows?
AI appears at three natural entry points in a typical accounting practice: the document handling layer, where invoices, bank statements, and engagement letters are processed; the research and query layer, where staff look up tax rules, standards, or client-specific questions; and the communication layer, where client correspondence is drafted. These are the points where AI tools already exist in accounting software or are available as add-ons.
In the document layer, tools like DataSnipper extract data from invoices, bank statements, and contracts straight into Excel working papers, maintaining traceable links to source documents. This enables full-population testing rather than sample-based audit procedures and replaces what was previously manual vouching.
In the research layer, AI-powered tools query curated tax and accounting content libraries and return relevant sections quickly. ICAEW notes these are particularly useful for IFRS queries and translating financial information between frameworks. Larger UK practices are also deploying internal AI assistants modelled on PwC’s ChatPwC, connecting staff to firm knowledge rather than the open internet.
In the communication layer, document summarisation and email drafting are among the most commonly reported first applications. They carry lower risk than technical accounting tasks, offer quick time savings, and give staff a practical basis for understanding what AI can and cannot handle before applying it to more sensitive work.
When should a practice actually use AI, and when should it wait?
AI delivers reliable value when the task is high-volume, clearly bounded, and produces outputs a professional can quickly check. It struggles where data quality is low, where context requires professional judgement at every step, or where client confidentiality controls haven’t been properly configured. The regulatory and reputational costs of getting it wrong in accounting mean the due diligence is worth doing before you start.
Workflows that suit AI well include invoice processing, bank reconciliation, and document summarisation. These are clearly bounded tasks where the pattern-matching or extraction work is well-defined and the output can be reviewed efficiently. For these, purpose-built accounting tools with AI already configured for client data handling are the appropriate route.
The clearer limits apply where professional judgement is the substance of the work. ICAEW guidance describes AI as a “junior assistant”, useful for drafting and analysis, but always subject to expert review. FRC standards require auditors to obtain sufficient appropriate audit evidence and document their work; AI supports this, but the professional accountability does not transfer to the tool. A practitioner who signs off on AI-generated outputs without critical review takes on the risk of those outputs directly.
For micro-practices, there is also an economic consideration. Where subscription costs for AI tools run to several hundred pounds per month and transaction volumes don’t support that outlay, the business case doesn’t hold. Modelling actual time savings against actual costs, including setup, training, and transition time, before committing is sensible rather than overcautious.
What does a practice need in place before deploying AI?
AI deployment in accounting sits inside a clear regulatory frame. UK GDPR and the Data Protection Act 2018 apply to any AI tool processing client financial data, and the ICO expects a data protection impact assessment before introducing new technology. ICAEW and the FRC both make clear that professional accountability remains with the practitioner, regardless of how much of the underlying work the AI tool handles.
The ICO’s AI and data protection guidance sets specific expectations around DPIAs, explainability, and human oversight for significant decisions. Using a consumer AI tool for client financial data without understanding where that data goes creates a straightforward compliance risk. The ICO’s fine against Interserve, £4.4 million in October 2022 after a security failure exposed the personal data of 113,000 employees, gives a sense of scale when digital systems handling personal data are mismanaged.
The NCSC’s Cyber Essentials guidance covers the baseline controls needed when processing financial data digitally, including strong identity management, patching, and supply chain risk management. AI tools that centralise more client data also centralise the risk if those controls are weak.
AccountingWEB recommends that UK practices treat AI deployment as a strategic project, including workflow mapping, data governance, and clear staff policies, rather than allowing ad-hoc experimentation with client data. Practically, that means staff should know which tools are approved for client work, what data can go into them, and where human review is required before an output is acted on.
For practices that advise FCA-regulated businesses, AI vendor due diligence and governance accountability are part of what the FCA expects from authorised firms managing operational resilience. The CMA’s ongoing review of AI foundation models is also worth monitoring, given its concerns about concentration in the market and the potential effect on pricing for smaller customers.
