AI and automation are equally important when you’re looking at digital tax compliance solutions. You can’t talk about one without the other.

Right now, it may seem like AI is a stadium-filling rock star, and that, by comparison, automation is playing small clubs. In reality, both can have a huge impact on the work you do. AI’s potential is yet to be fully understood, and automation has affected millions of lives for centuries in its different forms.

As a decision-maker in an accountancy practice, you need to understand the role both play. You must know the differences and where they add value, including in areas like Making Tax Digital for Income Tax (MTD for IT). Only then can you make effective buying decisions and put tech to the test.

Who wrote this guide, and why should you trust it?

This guide was written by IRIS accountancy experts Eva Mrazikova and Jon Cooper.

We want this guide to first help you with choosing tax compliance software in terms of AI and automation. From there, we move onto general governance around AI, to help you during this fast-changing time.

This guide does mention IRIS software to provide examples. That’s not intended as a hard sell. We want to give you real examples of automation and AI, and it would be odd to talk about Sage or CCH when we know our own software inside-out.

Building this software for nearly 50 years has given us experience we hope you’ll find helpful. In that time, we’ve developed HMRC-recognised software and earned certification for Cyber Essentials, ISO 9001, and ISO 27001.

We’ve also helped accountants make the most of change many times over. For instance, there wasn’t much of an internet back when IRIS started in 1978.

About our IRIS experts

Eva Mrazikova: Eva is an accountant who blends financial expertise with a passion for innovation and technology. With 20+ years of experience across accounting, marketing, and technology, she now leads go-to-market strategy, product positioning, and customer engagement at IRIS Software Group.

Jon Cooper: Having spent 10 years in practice running a team of accountants and bookkeepers, Jon knows first-hand the pressures firms face, and he is passionate about making their lives easier. At IRIS, he now helps connect software and services to the real needs of accountancy firms.

This guide answers the following questions in depth

Let’s start with a very rapid-fire Q&A. It will give you a sample of what’s in the guide and point you straight to the info you want.

How long have AI and automation been around? Longer than you’d expect. Turing’s eponymous test dates to 1950, while the phrase “artificial intelligence” was coined in 1956 and the first public demonstration of a neural network followed two years later. Automation goes back yet further, to the industrial revolution.

We look at this, and the popularity of AI in the last few years, in our AI and automation backgrounder.

What’s the difference between AI and automation – and when should you use each? Long story short, automation is for predictable outputs like calculations, notifications, and moving data. AI helps with interpretation and judgement, such as reading a receipt, trendspotting, and drafting a new letter.

We explain where each one belongs and what influenced our approach to both in the difference between AI and automation.

Which accountancy software has the best AI and automation for tax compliance accuracy? With technology moving so fast, it’s hard to give a definitive answer, but established vendors – IRIS included – maintain frontrunner status in this field thanks to experience, integration, and the need to protect a trusted brand through careful implementation.

Find out much more about the best AI software for tax compliance accuracy.

What does good AI and automation look like in tax software? Good automation centres around a single point of truth, which commands an entire customer journey from onboarding through to filing and beyond. Meanwhile good AI handles laborious, thought-intensive tasks as part of your workflow, such as checking a return against previous years for anomalies.

There’s much more, including how AI and automation help an accountant in MTD, in our section on what good AI and automation look like in tax software.

How should you choose and govern AI in your practice? There are three things to keep front of mind: the “black box” problem, which means you must act responsibly when it comes to AI; that checking the output is a job in itself – one that can grow the more you add AI or data; and the “garbage in, garbage out” rule. The latter means the better quality of data you put in, the better results you’re likely to get. You also need to consider the ethics; you remain responsible for your clients’ data, regardless of who sells you the product.

The full governance section covers how to choose and govern AI in your practice.

How do I implement an AI system without wasting money? You need to set clear objectives, check it conforms to GDPR compliance, run a risk assessment, and start small to test the solution before spending more. Evaluate using a structured, people-centred process and do so regularly.

We take you through the whole process of implementing an AI system without wasting money.

Do you use AI at your accountancy practice?

Choose an answer

No, I don't use any. Yes, AI chatbots like ChatGPT/Claude – for general research, writing letters, etc. Yes, but only AI built into our accounting software – for tasks like compliance. I use both AI chatbots and AI built into my accounting software.

Vote

What are AI and automation?

Let’s look at the wider role AI and automation play. We’ll provide a brief backgrounder before moving onto seeing how they do different things in compliance software. For now, remember that AI acts like an extra “brain”, while automation’s job is to act on triggers and move processes along.

How long has AI been around?

We’ve had AI for a long time – longer than many might expect. Alan Turing’s famous test, to see if computer intelligence is indistinguishable from a human’s, was set out in his 1950 paper for the journal Mind. In fact, the phrase “artificial intelligence” was coined a few years later, for the 1956 Dartmouth Summer Research Project. Only two years after that, the world had its first demonstration of a ‘neural network’ – a set of processes inspired by the human mind.

AI went on to develop fairly consistently until its recent rise to prominence. AI technology and methods fall under categories ranging from Machine Learning (which learns from data and spots patterns) to computer vision which can identify imagery and real-world objects.

What are the main types of AI?

🧱 Machine learning – This teaches machines to learn from data (e.g. Netflix suggestions, fraud detection).

🧠 Deep learning – This is a powerful type of ML using brain-like layers to handle complex tasks (e.g. self-driving cars).

🗣️ Natural language Processing (NLP) – This helps machines understand and generate human language (e.g. chatbots and translation).

👁️ Computer vision – This enables machines to interpret images and video (e.g. facial recognition and object detection).

🤖 Robotics – This combines AI with machines to interact with the real world (e.g. warehouse robots and drones).

📚 Expert systems – This uses human-coded rules to make decisions (e.g. early diagnostic tools).

Why is AI so popular now?

This all began when OpenAI launched ChatGPT on 30 November 2022.

Their ‘Large Language Model’ (LLM) used huge amounts of text data to feed its algorithms. It caused a lot of excitement because ChatGPT allowed people to yield impressive results with text and code in seconds.

In only two months, an estimated 100 million people were using the free software. Compare that with Instagram, which took two-and-a-half years to attain this level of success.

How long has automation been around?

We’ve used automation for centuries, in one form or another. Where it gained real momentum was during the industrial revolution. During that time, it was machinery that made tasks much more easily repeatable.

From there, automation moved into the office. In our industry, we’ve seen it everywhere, from accounting to payroll and invoicing. In many ways, it’s as powerful as AI, but because we’ve had this way of working for so long – and in so many forms – we’ve become used to it.

What are the main types of automation?

⚙️ Mechanical automation – This is when machinery performs physical tasks automatically (e.g. production lines and packaging systems).

📋 Business process automation (BPA) – This automates routine business activities (e.g. payroll, invoicing, expense processing).

🔄 Workflow automation – This moves tasks or information through predefined stages (e.g. approval workflows and onboarding processes).

🤖 Robotic process automation (RPA) – Here, software bots carry out repetitive digital tasks (e.g. transferring data between systems and creating reports).

☁️ Cloud automation – This automatically manages software, systems, and infrastructure (e.g. backups, software updates, and security monitoring).

Now, it’s time to see how this all applies to you as an accountant. What AI matters? How can the right automation make a big difference to your tax workflows?

How do AI and automation work in tax compliance software?

Right now, it makes sense to speed up compliance work with automation and AI. In modern accountancy, tax returns are still the bread and butter of the profession. But they are an increasing burden, as Making Tax Digital for Income Tax (MTD for IT) turns one tax return into at least four submissions and a final declaration.

All that puts extra pressure on you and increases the likelihood of mistakes. It also stands in the way of growth. Compliance might be the foundation of accountancy, but, because it’s so time-intensive, it also stops you from delivering what the profession does best – advice.

Which accountancy software has the best AI and automation for tax compliance accuracy?

For tax compliance accuracy, the strongest AI and automation currently sits with established practice software vendors rather than newer AI-first tools. We’re not making a claim about any single product, and the field moves quickly enough that any ranking would date. What it reflects is three things that established vendors have that you can’t create overnight:

• Experience: because the vendor has lived through changes and recruits former accountants and top developers.
• Integration: because AI and automation need to move seamlessly. The less the software integrates with other functions, the more stop-and-start an accountant’s day becomes.
• Trusted brand: because damaging this relationship, especially in a profession where accuracy is everything, is not worth it. Your business suffers and the vendor will (rightly) lose that relationship.

IRIS ticks all three of these. It’s why we take a curious but careful approach to adding AI and automation.

What’s the difference between AI and automation in accountancy software?

At IRIS, our experts understand the strengths and weaknesses of automation and AI: with this in mind, we deploy automation where outputs need to be predictable, and AI where there needs to be data interpretation.

Let’s look at this in more detail now.

AI can creatively sift information – it’s a very powerful predictive solution that has a penchant for trend-spotting, while sometimes using web access to supplement its own data.

On the other hand, if you ask an LLM to count from 1-20 in voice mode, it might struggle.

Automation, however, picks up the slack when you want a more predictable response to a given situation. For example, automation might trigger a notification when a client invoice is overdue. Automation could do more with this. For instance, it could then bring up a template reminder letter, fill in the name, and prepare the document for sending.

You could use an LLM (large language model AI) like ChatGPT to draft this letter. However, because of how the most popular LLMs operate, it would be different every time.

But what about reading a document? Here, automation will be less effective. AI is good at this. It can look at a receipt, for example, and work out – from its sample data – what those numbers are and what they represent. It can then fill out a form for you in a system.

What are some examples of good AI features in tax software?

If a task needs judgement, this is where AI can help. Imagine it doing some preliminary “thinking” before you as the accountant provide your check and analysis.

What does good AI tax software look like? By IRIS standards, it should:

• Look at a return and compare it against historic information and make sure there is nothing odd. 
• Read a document (online or on paper) and then turn that into meaningful, populated data.
• Check work for gaps against the newest regulator guidance.
• Explain what methods it used to do something and await human review.

The IRIS approach is very deliberate, building AI into your process as opposed to pushing accountancy through AI. We took our time to plan and add features because of how sensitive your job is. It’s the work that has to work.

Decades of experience means we also were careful with your data. We understand that much of it belongs to your clients. When an AI process runs, personal data is stripped out; everything is handled in the UK.

What does a day with good AI tax compliance look like?

Think of AI as doing all the busywork that would otherwise stop accountants from using their full analytical skillset.

Let’s imagine you are helping a client in busy season. Normally you would be retyping information, worrying about missing data, and have an eagle eye on the wants and wishes of HMRC. Here’s how AI could help you:

• The client has sent over a P60, a bank statement, and a dividend voucher in PDF format. Instead of awkwardly retyping or cutting and pasting, the system now pulls that information straight into the return, ready for you to check.
• After you have filled in the return, you run the anomaly detection. This looks back at the previous two years of returns and flags if anything is odd, missing or seems completely new compared to previous submissions.
• Before you send everything to HMRC, the software runs a check. It goes through all the current rules and help sheets, so you feel assured before pressing “submit”.

So, the software is not taking the expertise out of your hands, but what it is doing is removing needless work, reducing the risk of error from retyping, and running checks and balances to supplement your own.

What are some examples of good automation in tax compliance?

If an accountant will find something repetitive, then automation should pick up the slack in their software. By IRIS standards, good automation should:

• Make reliable calculations, so you’re not messing with spreadsheet formulas or squinting into a calculator.
• Trigger notifications so you are spinning far fewer admin plates.
• Move data for you – or make it sharable – so you retype as little as possible.

To do this well, automation needs support from a single source of truth. For accountants, that’s a central client list. For example, both IRIS Accountancy Suite and IRIS Elements have this.

With good automation, you can:

• Upload data directly from clients using a portal. This means no retyping or sending files back and forth.
• Use the data throughout the client journey. What is used to onboard a business can feed through into filing accounts, and that information can also go straight into the tax area.
• Link to other software, sometimes with the help of an API, to call data without any re-writing or copy-and-paste.

With live client information held in one place, updating Companies House or sending figures to HMRC has never been easier – just a few clicks. This means you can use that time to improve your client offering and reach out to prospects.

How should you govern AI in your practice?

The reality is that although IRIS might update this guide, AI will always try to outrun it.

That’s why governance matters. You must be ready for every pitfall while taking full advantage of what it offers. One report suggests AI-first businesses (across all industries) take 80 days longer to recover from a cyber security incident.

Adopting the right mindset will mean no matter how AI grows and changes, you will always be able to make the right decisions about how to integrate it into your practice.

What is the AI black box problem?

Earlier, we mentioned that if you asked AI to write a letter, then no two documents will be the same. This is because the type of AI used, an LLM, has an element of randomness to it. Even the developers of an LLM – which sifts requests through vast neural networks fed by unfathomable amounts of training data – will not be able to fully predict the output of their chatbot. Hence the working out is hidden in a so-called “black box”

As such, there must always be a human element to working with AI. Even carefully curated AI – such as a receipt scanner, which has a far narrower focus – will have less volatility, but it can still make the odd mistake.

Why does AI output still need checking?

AI is great but comes at a price. That price is the time taken to check the outputs. If you have software that focuses on clear datasets and uses AI to do busywork, then checking that will be simple. For example, if it pulls data from a PDF, and you have the PDF up on-screen, then there’s not much to do; you give everything a ‘once over’ before accepting the system has done its job. It’s likewise simple to spot check year-on-year anomalies.

Where it gets more difficult is when you use multiple AIs. If you have one bot triggering another, and another, then chances are you are further removed from the thinking processes that delivered the result. That means more time needs to be taken checking work.

Likewise, the bigger the dataset, the more vigilance. We’ve probably all had a chatbot hallucinate when it surfed the web to answer a question. That is the current state of consumer-level AI.

Why does the quality of your data matter to AI?

There’s a useful phrase to remember – “garbage in, garbage out”. That means you should link AI to a reliable system that has properly cleaned data. Don’t have AI try to sift a spreadsheet because you think it has grown out of control and is rife with mistakes. It’s not there to fix poor data.

How do you evaluate an AI system before buying?

Have a system for weighing up AI before buying.

Gather:

• Anecdotal evidence: From peers and networks who have used the tool.
• Descriptive evidence: White papers and case studies from the vendor.
• Correlation evidence: Compare efficiency before and after. Measure it, rather than go with feelings.
• Output evidence: Look at practical outcomes like billing and see if the numbers are going up

Who is legally responsible for client data in AI tools?

Whoever is selling you an AI product, it is important to be vigilant. You are responsible for your clients’ data, so always – even if it’s IRIS – check the Ts & Cs on any software you are adding.

This goes double for more consumer-focused products like Google Gemini, ChatGPT and Claude, to name a few. Some of these will offer an “Enterprise” tier, which affords greater security at a cost. Make a judgment call – don’t be tempted to use a “Pro” tier piece of software because it’s cheaper, when you need the “Enterprise” level. Nevertheless, ensure it conforms to GDPR.

Even when you have the right agreement in place, check with IT that there are the correct safeguards in place. This includes any sharing and privacy settings; you don’t want to be accidentally training a chatbot on your client data.

How do you implement an AI system without wasting money?

How do you set clear objectives for AI?

Before adding any AI, it’s important to define your goals. What is it you are hoping to achieve? Are you adding ChatGPT just because you like it, or will it have a clear set of purposes?

It’s best to define what the software will be used for, because there will be a cost associated with it to your practice – DON’T use the free tier of any consumer product.

Will the AI tool work with your existing systems?

Can you add the tool to a workflow? Is it easy – and safe – to upload information or link the AI to your SharePoint?

What risks should your AI assessment cover?

LLMs and other forms of AI can bring risks. What happens if the provider gets hacked, for example?

It doesn’t end there. If you are a bigger firm and using it to sift job applications, there’s always the danger of bias. When it comes to AI, remember that ethics are just as important as security and logistics.

How many AI licences should you start with?

Licences are not always cheap, and some general AI services will eat through “tokens” fast to generate content. People complain about this with LLMs like Claude all the time. For this reason, it might be wise to start with the smallest number of users and scale up. For “Enterprise” plans for software like Microsoft CoPilot, there might be a minimum number of seats. Start with that minimum, then get feedback. It might be worth upscaling, or the AI might not meet expectations.

Set two dates in the calendar. The first is for when the initial sample test is carried out. The second is for after you have expanded the usage and the contract is about to expire. On all these occasions, run checks to see if you are getting real value for money.

Which IRIS tax software has AI and automation built in?

IRIS builds AI and automation directly into the tax workflows you know and trust.

AI-assisted preparation pulls data from documents like P60s, P11Ds and bank statements, and our AI compares each return against previous years.  You are then able to send over the return worry free, because our system will check your final declaration against HMRC rules.

Decades of automation knowhow drives your data with the help of a central client list – from accounts to return – and helps you get data from your favourite bookkeeping solutions.

IRIS Elements Tax was built for the cloud. HMRC-recognised, and MTD-ready, it’s a strong choice for sole traders or large enterprises (depending on your plan). From £10/month, with a 30-day free trial.

IRIS Accountancy Suite Tax is the established desktop favourite, ready to handle even the most complex compliance asks. Personal Tax, Business Tax and Trust Tax share one client database.

Both are HMRC-recognised, MTD-ready, backed by ISO 27001:2022-accredited data governance, and now include AI anomaly detection at no extra cost.