AI & the future of accountancy and advisory

AI processes a business's records according to rules the firm sets. A purchase invoice is recognised, the supplier, amount and VAT are extracted, and AI suggests the ledger account based on earlier entries. Bank transactions are matched to open invoices. Work in the firm shifts from data entry to review: an entry that breaks the pattern, a supplier with a new bank account number. With structured e-invoices, invoice data moves directly from system to system, with no PDF or scan. An entry can therefore be ready before anyone has looked at the invoice. Once delivery and payment also feed in data directly, bookkeeping that follows each transaction as it happens comes within reach.

With data analytics, an auditor can examine all the journal entries supplied for a year, alongside or instead of a sample. AI looks for what breaks the pattern, such as weekend postings, round amounts just before year end or payments to a new supplier without a purchase order. An anomaly is not yet an error, but it is a reason to look further. The next step goes from all entries to all sources: AI sets the books against contracts, emails, shipping data and external registers. That reveals discrepancies the figures alone would miss, such as a recorded delivery with no lorry movement behind it.

Tax advisers use AI to search legislation, case law and tax authority guidance. Linked to tax databases, language models compare rules and find relevant rulings, with a reference to the source. Research that used to take an afternoon is done in minutes. The next step is looking ahead. In 2016, researchers had a model predict the outcome of cases at an international human rights court from the text of published judgments, with an average accuracy of 79 per cent. A possible next application: estimating in advance how a tax position might fare in court. AI can also link a draft bill to available client data and calculate the financial consequences. The adviser then sees sooner which businesses will be affected and how large the impact could be.

Businesses use AI for credit decisions, recruitment, pricing and customer contact. Boards, regulators and customers want to know whether those systems work as intended. Assurance over AI is an independent examination of agreed properties of an AI system or of how it is controlled. The accountant tests, for example, whether the data is sound, whether decisions can be explained and whether oversight is adequate. Large firms already offer such services. The European AI Act sets requirements for high-risk systems, and certifiable international standards exist for managing AI. The engagement determines which criteria are examined; a certificate does not guarantee that every decision the system makes is right.

An AI agent carries out a task of several steps on its own, within limits people set in advance. Agents already process source documents and prepare parts of tax returns. In trials, firms add requesting documents such as annual statements and bank statements, sending reminders and tracking what is still missing. Chasing paperwork disappears from the employee's day. The next step: agents of the firm, the bank and the client working directly with each other. The firm's agent requests the annual statement from the bank's agent, which checks that the client gave consent and supplies the document. The business owner no longer has to search for anything, and the file is complete before the adviser starts on it.

In management and strategy consulting, much of the time went on preparation: gathering market data, analysing figures, writing a first report. AI speeds up that work, and the value moves from the report to the conversation. A market analysis that used to take a team weeks can now be ready in days. During a board meeting, AI calculates live what a proposal means: what a price rise does to volume and margin, how closing a site affects revenue and staffing. The board sees the consequences of a choice while it is still discussing it. The adviser asks the right question and weighs an outcome against what they know of the client. The adviser works with the client; AI handles the repetitive tasks and the number-crunching.

With up-to-date books, a firm can regularly refresh the forecast of revenue, costs and cash flow, and AI runs the numbers. For a business owner, next month's question is often more pressing than last year's: can we afford that investment, and is there enough cash for the summer? The system flags what stands out, such as a margin slowly eroding or customers paying later than last year. The next step is the forecast that warns before things go wrong. Research shows that certain machine learning models predict bankruptcies more accurately than traditional models. Continuous analysis can reveal the first signs of financial trouble well before the annual accounts show them.

A firm can offer its professional knowledge through AI assistants that clients consult themselves. AI links a question to the right information in the knowledge base and answers in plain language. Knowledge that now lives in people's heads and in memos becomes a product: an assistant for questions about staff expenses and benefits, a model for calculating the transfer of a family business. The firm earns from its knowledge without billing hours for every question. A possible next step is an agent that tracks rule changes and checks a client's payroll. It can adjust calculations and pass questions that need judgement to the adviser. The firm then also offers continuous monitoring.

A digital twin is a digital model of an organisation or supply chain that is continuously updated with real-world data: stock, lead times, capacity and costs. Advisers build such a model with the client and use it to explore how a change ripples through the whole chain. In trials they combine the digital twin with AI, which runs thousands of variants: what does a dearer raw material do to margin and lead time, and what happens if a supplier on the other side of the world drops out for a month? The model remains a simplification. But it shows where a chain is vulnerable before things go wrong, and once the engagement ends it stays behind as the client's own tool.

A financial AI assistant answers questions based on the books and the available documents. That lets a business owner look into simple financial questions themselves. Accounting software is already gaining such assistants: the owner asks in plain language how cash flow looks or which invoices are still unpaid. Clients then handle some of the tasks that now fall to the firm. The firm gives an independent opinion when a bank or investor asks for one. It also supports the owner through an acquisition, a succession or a dispute with the tax authority, where knowledge of the business and personal advice are needed. The firm gets fewer calls about the cash position, and more about what the owner wants to do with it.

An AI-first firm lets AI do the work people do not need to do, including the repetitive and predictive work: processing, testing, analysing, forecasting and drafting. Accountants and advisers assess what AI finds, weigh the risks and talk decisions through with clients. In a scenario for the next five to ten years, the shape of the organisation changes with it. Agents do much of the operational work, and a professional directs their own team of agents. The pyramid with many entry-level roles at the base narrows, with more specialists and fewer layers in between.

When audit and advice use AI, accountants work more often with specialists in data and models. A data auditor checks that a client's data is complete and reliable before the analysis begins. A specialist in AI assurance tests models, at clients and within the firm. Experienced accountants are moving into these roles too. New specialisms are taking shape: the agent auditor who continuously tests how agents behave, the coordinator who deploys and steers a team of agents, and the specialist who audits tokens and smart contracts. Specialists in data, models and code are taking a bigger share of the work and of the firm's make-up.

In an audit team with AI, people and AI systems work on one file, each with its own task. AI reads contracts and minutes, reconciles the figures, analyses every entry and drafts a first risk assessment. The team decides what further testing is needed, including on items AI did not flag, and discusses anomalies with the client: why was this provision reduced, who approved that payment? The opinion and the signature stay with the auditor. The next step: agents build most of the file throughout the year. People handle the conversations, the borderline cases and the judgement. The peak of busy season could then largely disappear from the profession.

Accountancy is built on certainty: an opinion has to be right, advice has to hold. That care is the profession's strength, and mistakes have serious consequences. A team used to checking every step itself first wants to see that a system works reliably. In a contained trial, staff compare AI's outcomes with their own assessment, separate from the live file. Experienced staff test the systems, because they are the first to spot where the results do not match practice. The benefit shows in hours freed up for the client. The professional standards still apply; the way of working changes.

A future academy is where an organisation trains its own employees and leaders of the future. The work changes faster than courses can adapt their programmes, and new cohorts of graduates are years away. So the organisation takes the retraining of its people and the training of new colleagues into its own hands. Existing staff learn to work with AI, agents and new technology, newcomers practise tomorrow's work from day one, and experienced colleagues pass on what they know.

In an ecosystem, an organisation takes on bigger questions than it could handle alone: with start-ups that bring new applications, scientists who examine what is reliable, partners who contribute knowledge and data, and governments working on shared infrastructure, such as digital reporting. That network does not stop at the border. International networks share methods, models and agents, and what one member develops, another can apply sooner. Smaller firms pool their strengths this way and work with the same technology as a large international network. What a firm can do then depends less on its size and more on its ecosystem.