Everything you need to know about MTD and AI| White Paper

Foreword

The shift from "digital record keeping" to "intelligent tax systems."

The UK accounting profession is currently living through one of its most disruptive periods in a generation. Making Tax Digital (MTD), HMRC's landmark initiative to digitise the entire tax system, is reshaping how millions of sole traders, landlords, and their advisors report income and comply with tax obligations. With MTD for Income Tax Self Assessment (ITSA) coming into mandatory effect from 6 April 2026 for those earning above £50,000, and rolling out to the £30,000 and £20,000 thresholds in 2027 and 2028 respectively, accountants are right to have MTD top of mind. But MTD is only the first act. While the profession is busy digitising records and retraining clients on quarterly submissions, a parallel, and potentially even more transformative, shift is underway: the arrival of practical, production-ready Artificial Intelligence in the world of tax.

This white paper, (aka Making Tax AI), explores that next chapter, which doesn't ask whether AI is coming to tax, but whether the profession is ready to meet it, and more importantly, how you can ensure that when AI arrives in your practice, it arrives on your terms.

1. Are We Ready for AI in Tax?

The short answer is: probably more than most accountants think, but less than others would have you believe.

The accounting profession has always been an eager adopter of digital tools, from ledger software in the 1980s to cloud accounting platforms like Xero and QuickBooks in the 2010s. MTD itself is essentially a forced digitisation event, accelerating practices that leading firms were already pursuing. So all the infrastructure, the digital records, API-connected software, cloud-based workflows, is either built or in-progress right now, largely because HMRC demands it. The good news is that this same infrastructure is precisely what AI needs to function.

The readiness factors working in our favour:

The mandatory adoption of MTD-compatible software means that, for the first time, the majority of the UK's small business financial data will exist in structured, machine-readable, API-accessible formats. This is the single most important precondition for AI to work effectively in tax. You cannot automate what is not yet digitised. The cultural readiness within accounting firms has also improved markedly. A profession that was once sceptical of cloud accounting is now overwhelmingly cloud-native. The generation of accountants entering practice today grew up with AI-adjacent tools and are comfortable with automation.

The readiness factors working against us:

Data quality remains inconsistent. AI is only as good as the data it processes, and many small business clients still struggle with basic record-keeping hygiene, from misclassified expenses, mixed personal and business transactions, missing receipts. MTD may help improve this over time, but people are people and the transition period will be messy. Regulatory clarity around AI use in tax is still developing. HMRC set out expectations for tax software developers using generative AI, but stopped short of issuing formal guidance on the use of AI in tax filing. The profession is largely operating on its own professional judgement for now, which introduces risk for firms that move too fast without appropriate governance.

Trust remains a significant barrier among both firms and clients. Particularly for older more risk-averse clients, the idea of an AI black-box making decisions about their tax affairs is uncomfortable. Accountants will need to be the trusted intermediary who explains, validates, and takes responsibility for AI-assisted outputs.

Verdict: The profession is ready to begin adopting AI carefully, incrementally, and with proper oversight. The firms that start now, with a thoughtful approach, will have a significant competitive advantage by the time the £20,000 MTD threshold arrives in 2028 and the total addressable market reaches its peak.

Find out more

2. What's on the Landscape for AI in Tax Right Now?

The AI tax landscape in 2025/26 is a mix of mature, embedded capabilities and emerging, standalone AI-native tools. Understanding the difference matters.

Embedded AI in existing platforms

The major cloud accounting platforms are integrating AI quietly and effectively. Xero, QuickBooks, Sage, and FreeAgent all now include features powered by machine learning, automatic transaction categorisation, bank rule suggestions, anomaly detection in expenses, and predictive cash flow modelling. These are not headline-grabbing AI announcements; they are practical, already-in-use capabilities that most accountants interact with daily without necessarily thinking of them as "AI."

AI-native tax and compliance tools

A new category of purpose-built AI tools is emerging specifically for tax practitioners. These include platforms that can review prior-year tax returns and flag optimisation opportunities, tools that generate draft tax computations from categorised bookkeeping data, and systems that monitor legislative changes and assess their impact on a firm's client portfolio automatically.

Large Language Models (LLMs) as practice tools

General-purpose AI assistants, including Claude, ChatGPT, and Microsoft Copilot, are increasingly used in accounting practices for drafting client communications, summarising complex legislation, generating first-draft tax planning memos, and producing training materials. These are not tax-specific tools, but their application in tax is real and growing.

HMRC's own AI ambitions

HMRC is itself investing heavily in AI. The department is deploying AI to detect tax evasion, using generative AI to assist caseworkers, and developing digital interfaces that will eventually allow taxpayers to interact with HMRC's systems in more conversational, intelligent ways. This matters for accountants because the regulator they deal with is itself becoming more AI-driven, meaning the profession needs to understand AI not just as an internal tool, but as the way their regulatory counterpart will increasingly operate.

Agentic AI and workflow automation

The frontier of AI in tax, and where the most exciting developments are happening, is agentic AI: systems that do not merely assist with a task but autonomously complete sequences of tasks. In the context of accounting and tax, this means AI "agents" that can reconcile a bank account, categorise transactions, identify anomalies, draft a quarterly MTD update, and flag exceptions for human review — all without manual intervention at each step, using "bounded agency" (i.e. clear, effective guardrails).

How Can AI Work in Tax?

AI is not really a single technology, it is a family of capabilities, each suited to different types of tasks. Understanding which type of AI applies to which type of tax work is essential for any accountant building a practical AI strategy.

Pattern recognition and classification

This is where AI has the longest track record in tax. Machine learning models trained on large datasets of financial transactions are remarkably good at categorising income and expenditure, identifying recurring patterns, and flagging transactions that look out of place. For bookkeeping-intensive tasks like bank reconciliation, expense categorisation, and VAT coding, this type of AI is already mature, reliable, and cost-effective.

Document processing and data extraction

AI can read, interpret, and extract structured data from unstructured documents like invoices, receipts, contracts, payslips, P60s, and bank statements. Optical Character Recognition (OCR) combined with AI interpretation means that a client photographing a receipt on their phone can have that expense automatically categorised, coded, and entered into their accounting records. This directly supports MTD compliance by reducing the friction of digital record-keeping.

Natural language generation

AI can draft written communications such as client letters, tax planning memos, queries to HMRC, explanatory notes on tax computations or working papers. Used carefully, with human review, this can significantly reduce the time partners and managers spend on routine written communications.

Tax research and legislative monitoring

AI tools can monitor legislative changes, like Finance Acts, HMRC technical guidance, tribunal decisions, and surface relevant changes for a firm's client portfolio. This is particularly valuable for specialist tax practitioners dealing with complex areas like R&D tax credits, international tax, or property taxation, where keeping current is time-consuming and high-stakes.

Predictive modelling and tax planning

AI can model the tax implications of different scenarios like incorporation decisions, pension contributions, and property disposals, much faster than traditional spreadsheet modelling. This shifts tax planning from an annual event to an ongoing, data-driven conversation between accountant and client.

Workflow orchestration and agentic automation
The most advanced applications involve AI agents that orchestrate entire workflows, connecting bookkeeping data to tax compliance to client communication to HMRC submission, with humans in a supervisory or escalation rather than operational role. This is where platforms like the Operating System at Bots For That are operating and enabling accounting firms to automate routine compliance workflows entirely while their people focus on judgement, advisory, and client relationships.

Ask a question

4. What AI Wouldn't Work in Tax?

Being realistic about AI's limitations is as important as being excited about its potential. The following are areas where, despite the hype, AI is not yet a reliable standalone solution. And for the record, humans + AI, nearly always give the best outcome.

Complex judgement calls requiring professional accountability

AI cannot replace the professional judgement of a qualified accountant in complex, fact-specific situations. Whether a particular expense is wholly and exclusively for business purposes, whether a transaction has the hallmarks of a notifiable arrangement, whether a client's disclosure is adequate - these require not just knowledge but professional accountability. AI can inform these decisions; it cannot and should not make them.

Novel or edge-case tax law

AI models are trained on historical data and existing legal frameworks. They perform poorly on genuinely novel situations like new legislation in its first year of operation, tribunal decisions that conflict with established practice, or unusual client circumstances that don't match the training data patterns. The early years of MTD, in particular, will generate novel compliance questions that AI tools will struggle with.

Client-facing emotional and relationship work

Tax can be a source of significant stress and anxiety for clients, particularly around investigations, penalties, or unexpected liabilities. The empathy, reassurance, and personal relationship that a skilled accountant provides in these moments is not replicable by AI. Clients who receive bad tax news need a human on the other end of the phone.

Unstructured or poor-quality data environments

AI is dependent on data quality. A client who keeps records in a mixture of handwritten notes, WhatsApp messages, and a shoebox of receipts is not a candidate for AI-led compliance, at least not until the data is cleaned and structured. MTD will eventually help, but firms should not expect AI to solve upstream data quality problems; it will simply automate errors faster.

Highly political or reputationally sensitive tax positions
Tax avoidance, aggressive planning structures, and positions that are technically defensible but reputationally questionable require human judgement about risk tolerance, ethics, and client relationships. These are not decisions to be delegated to AI.

5. An AI Readiness Checklist for Accounting Firms

Use this checklist to assess where your firm stands before implementing AI tools in your tax practice.

Data and Infrastructure

  • Client bookkeeping data is held in cloud-based, API-accessible accounting software
  • Bank feeds are connected and regularly reconciled for the majority of clients
  • The firm uses MTD-compatible software and is actively preparing clients for quarterly reporting
  • Document management is digital (invoices, receipts, and client correspondence stored electronically)
  • The firm has a defined data retention and security policy

Process Readiness

  • Key tax and bookkeeping workflows are documented and standardised
  • The firm has identified which workflows are highest-volume and most repetitive (primary AI automation candidates)
  • There is a clear client onboarding process that captures data in structured, usable formats
  • Quality review processes exist that can serve as a "human in the loop" check on AI outputs

People and Culture

  • Partners and senior staff are aware of and open to AI tools in practice
  • At least one team member is designated as an "AI champion" responsible for evaluating and implementing tools
  • Staff understand the difference between AI as a tool they control vs. AI replacing their role
  • Client communication about the use of AI in their affairs has been considered

Governance and Risk

  • The firm has considered its professional indemnity insurance position in relation to AI use
  • There is clarity on who is professionally responsible for AI-assisted tax outputs
  • The firm's engagement letters address the use of technology and automation in service delivery
  • A process exists for identifying and escalating AI errors or anomalous outputs
  • The firm has defined or at least considered an official policy and strategy for AI

Commercial Readiness

  • The firm has modelled the impact of AI automation on its fee structure and recovery rates
  • There is a plan to reinvest time saved by AI into higher-value advisory services, growth or other lines of opportunity
  • The firm has a view on which AI tools to pilot and a framework for evaluating them
Give us a call

6. Getting Started with AI in Tax: A Practical Roadmap

For most accounting firms, the right approach to AI adoption is not a big-bang transformation but a deliberate, phased program. The good news is that the starting point is already in front of you — it is called Making Tax Digital.

MTD is not just a compliance obligation. It is the foundational digitisation event that makes AI in tax possible. By mandating digital record-keeping, connected software, and quarterly data flows, MTD creates exactly the structured, machine-readable data environment that AI needs to function. The practical roadmap for AI adoption therefore begins not with a new technology decision but with the MTD implementation work your firm is already doing.

Phase 1: MTD and ITSA Compliance Readiness (The burning platform — now)

The immediate priority is/was ensuring every in-scope client is MTD-ready before the April 2026 mandatory deadline. This means migrating remaining clients to MTD-compatible software, connecting bank feeds, establishing quarterly submission workflows, and training clients on their new obligations. Done well, this phase is more than compliance hygiene, it is the data infrastructure investment that makes everything else possible. Every client brought onto clean, connected, cloud-based bookkeeping is a client whose affairs are now, in principle, automatable. Treat MTD readiness as the foundation of your AI strategy, not a distraction from it. Given we're weeks from MTD go-live for ITSA, I'm going to assume (and hope) you've got this part down.

Phase 2: Automate the Quarterly Cycle (Months 1–3 post-MTD go-live)

With the MTD infrastructure in place, the next step is layering AI onto the quarterly compliance workflow itself. Automated transaction categorisation, bank reconciliation, expense coding, and anomaly flagging can all be applied to the data flowing through your MTD-compatible software. The goal is a quarterly update process that is near-touchless for the accountant where AI handles the processing, the accountant reviews exceptions and approves the submission. Firms that achieve this will find that preparing a quarterly MTD update for a well-maintained client takes minutes rather than hours. Start with your cleanest, most straightforward clients to build confidence, then extend to the broader portfolio as the workflow matures.

Phase 3: Year-End and Final Declaration Automation (Months 3–6)

Once the quarterly cycle is running smoothly, extend AI into the year-end process. This means pulling together the four quarterly summaries, identifying required adjustments, reconciling against prior-year figures, drafting the Final Declaration, and flagging any discrepancies or planning points that warrant accountant attention before submission. The ambition is an end-to-end MTD compliance workflow, from opening bank balance to submitted Final Declaration, that runs almost autonomously for straightforward clients, with the accountant acting as reviewer and sign-off rather than preparer. This is where the significant capacity gains begin to materialise across the portfolio.

Phase 4: Proactive Tax Planning from Live Data (Months 6–12)

Here is where MTD's hidden gift to the profession becomes clear. Because clients now have real financial data flowing quarterly rather than annually, accountants have, for maybe the first time at scale, the ability to offer proactive, mid-year tax planning as a standard service rather than a premium add-on reserved for the largest clients. AI can model projected tax liabilities from the quarterly data, identify planning opportunities in real time (pension contributions before year-end, timing of capital expenditure, income-shifting between spouses), and prompt accountants to initiate timely client conversations. The shift is profound: instead of telling clients what their tax bill was, accountants can tell clients what it will be, and what to do about it now.

Phase 5: Full Advisory Transformation (Ongoing)

The compliance workflow is now largely automated. The accountant's role has fundamentally shifted from preparer to reviewer, from reporter to advisor, from reactive to proactive. This final phase is about deliberately reinvesting the capacity that AI has freed into higher-value work: quarterly management accounts, business performance benchmarking, strategic scenario planning, and the deeper client relationships that generate referrals and long-term firm value. New service propositions emerge naturally from the data infrastructure MTD created. New fee models follow, moving away from time-based billing for compliance towards value-based pricing for advisory outcomes. The chain is complete: MTD forced the digitisation, AI automated the compliance, and accountants were liberated to do the work they trained for.

7. Use Cases and Examples

Use Case 1: Automated Bank Reconciliation for MTD Quarterly Updates

A sole trader client runs a small design consultancy with 200–300 transactions per month across two bank accounts. Previously, their bookkeeper spent two to three days per quarter manually categorising and reconciling transactions. With AI-assisted reconciliation, the system automatically matches 85–90% of transactions to the correct category based on payee, amount, and description patterns. The bookkeeper reviews and approves the AI's work in under an hour, and the quarterly MTD update is prepared automatically from the categorised data. Time saving: approximately 80%. The bookkeeper now uses the freed time to prepare for a quarterly client review meeting, which is a new revenue-generating service that previously wasn't possible within the budget.

Use Case 2: AI-Assisted VAT Return Preparation

A retail client with mixed standard-rated, zero-rated, and exempt supplies generates complex VAT workings each quarter. An AI system trained on the client's historical VAT records automatically applies the correct VAT treatment to new transactions based on product codes and supplier classifications. Partial exemption calculations are modelled automatically from the categorised data. The accountant reviews the draft return, adjusts two line items where the AI has incorrectly classified a new product category, approves the return, and submits. A process that previously took four hours now takes forty-five minutes.

Use Case 3: Tax Planning Scenario Modelling

A husband-and-wife business considering incorporation wants to understand the tax implications. An AI tool is provided with their current income, business profits, dividend history, pension contributions, and mortgage details. Within minutes it generates a comparison of their current tax position versus post-incorporation, modelling optimal salary/dividend splits, the impact of director pension contributions, and the corporation tax position, including the transitional costs of incorporation. The accountant reviews the model, adjusts assumptions around timing, and presents a professional report to the clients. A process that previously required a half-day of spreadsheet work now takes two hours end to end.

Use Case 4: Proactive Tax Legislation Monitoring

An AI system monitors HMRC guidance updates, Finance Act amendments, and First-Tier Tribunal decisions daily. When a new ruling on the treatment of home office expenses for the self-employed is published, the system automatically identifies the 47 clients in the firm's portfolio who claim home office expenses and flags them for review. The firm's tax manager reviews the ruling, determines its impact on client claims, and sends a proactively drafted client communication explaining the change. The entire process takes three hours rather than the days it might previously have taken, and the clients receive timely, proactive advice rather than finding out at their next annual review.

Use Case 5: AI-Powered Client Onboarding and Compliance Health Check

A new client joins the firm with three years of accounts prepared by a previous accountant, a mixture of digital and paper records, and no existing cloud accounting setup. An AI tool processes the prior-year accounts and tax returns to extract key figures, identify historical patterns, flag potential compliance risks (an R&D claim that may have been under-claimed, an undeclared asset disposal), and generate a structured onboarding report. The partner uses this report to have an informed first client meeting that adds immediate visible value and identifies additional billable work. A process that previously took four to six hours of senior staff time is compressed to under two hours.

8. The Road Ahead: Making Tax AI — The Profession's Real Opportunity

MTD was imposed on the profession. Making Tax AI is an opportunity the profession can choose.

The firms that will thrive over the next decade are not those that resist AI or those that adopt it uncritically, they are those that integrate it thoughtfully, maintain professional accountability for its outputs, and use the time it saves to deepen client relationships and deliver genuinely higher-value advisory work. The analogy to MTD is instructive. In 2019, many accountants resented the quarterly reporting obligation. By 2023, forward-thinking practices had realised that quarterly touchpoints with clients created more opportunities for proactive advice, stronger relationships, and higher fee income. AI, deployed wisely, offers the same transformation, but maybe more profoundly and more permanently.

The journey from Making Tax Digital to Making Tax AI is not a giant leap. It is a logical next step and the firms that take it deliberately, today, will be the ones writing the success stories that inspire the rest of the profession to follow.

Try the MTD Command Centre

This white paper was prepared for accounting professionals navigating the intersection of MTD compliance and AI adoption. It reflects the current regulatory position as at February 2026 and should be read alongside current HMRC guidance and professional body updates.

About Bots For That

Bots For That develops the AI-Native Operating System for Accounting Firms designed to automate everyday accounting work and building the workflow and AI infrastructure that modern accounting firms run on. Where cloud transformed bookkeeping, our AI-native infrastructure transforms firms.

Back To Homepage