Should your small business be using AI?

It feels like every other day there’s a story about AI in the news - Oprah has called it “miraculous” while Geoffrey Hinton, who was awarded a Nobel Prize for his work on machine learning, warned it could be an “existential threat”. What’s the truth? Can there be a middle ground between AI’s fans and its critics?

There are some great reasons to embrace AI - if you use FreeAgent, some of your tasks are already made easier through features that use AI. Our automatic categorisations for your bank transactions and expenses may already be saving you time, for example. And you very likely have other AI-powered favourites, such as spellcheckers, meeting transcription services or search engines, saving you time, hassle or money.

But there are still legitimate concerns to bear in mind. Ultimately, the decision about how to use the new technology will have to be an individual choice for each business. Here are some of the questions you’ll want to ask as you shape your AI strategy.

First things first: what exactly is AI?

AI stands for artificial intelligence: an evolving technology that attempts to simulate human intelligence. The birth of AI as a field of study is usually dated to a workshop at Dartmouth College, New Hampshire, in the summer of 1956, but the instinct goes back much further. “The desire to codify our actions, to make them more efficient and to remove the 'boring' from our working processes has been a constant evolution,” says Craig Clarke, FreeAgent’s Director of Product. “AI is just the continuation of hundreds of years of work to try and remove the drudgery from our lives.”

While much of the current excitement is centred on large language models (LLMs) like ChatGPT and other generative models (named because they generate ‘new’ content, like sentences, images and even music), the technology actually has a much broader application than most people realise. Non-generative AI - also known as analytical or predictive AI - does not generate new content or data, instead it focuses on analysing and making decisions based on existing data. Common examples of non-generative AI include email spam filters, recommendation systems (such as the ones used by Netflix and Spotify) and fraud detection processes (with which financial institutions identify suspicious transactions).

In our June 2024 survey of small business owners, only 29.9% said they were currently using AI for their business. But Dave Evans, FreeAgent’s Head of Data Engineering, thinks the percentage is actually much higher.

“People are probably using AI already, whether they know it or not,” he says. “AI powers so many tools we use everyday - like spellcheckers, voice assistants [for example Amazon’s Alexa, Siri from Apple or Google Assistant], email suggestions, maps and route-finder apps, Google and other search engines… The list goes on.”

AI! Huh! What is it good for?

AI is now really competent at lots of different tasks, like spotting patterns and making predictions across large datasets; scheduling actions; summarising complex, multimedia content; and writing software. It is getting better at generating convincing new content - text, images and video - as well as editing existing content.

What questions should you ask before choosing AI for a task?

What’s the catch? The breadth of potential issues with AI might surprise you, from ethical concerns to digital security risks - and they change according to what sort of AI you are planning to employ. Here are some of the questions you might want to ask yourself before rushing full steam ahead.

Will AI speed up this task?

As AI doesn’t actually understand the answers it produces, you might need a human to check its work and make sure it is correct. (If you’ve seen the ChatGPT response to the question ‘how many Rs in strawberry’, you’ll understand the depth of the issue when it comes to generative AI.) This could sometimes take longer than a human working on it alone.

Can I trust AI answers?

The accuracy of AI largely depends on the quality and quantity of the training data.

Generative models need huge amounts of content to learn from, so are particularly vulnerable to the old computer science concept ‘garbage in, garbage out’. Trained on vast amounts of text, typically scraped from the open internet, LLMs can get facts wrong, and can even reproduce offensive or problematic viewpoints, or accidentally plagiarise existing work. Remember, generative AI can return wildly inaccurate or inappropriate answers.

Predictive models tend to use more structured data - which should have been cleaned in advance to remove outliers and irrelevant information - to make informed predictions. While the scope of what they can do is more limited than generative models, their answers are easier to validate and explain. This may be an essential requirement in use cases such as a bank deciding if a customer is eligible for a loan. You may be familiar with ways predictive AI can get it wrong, though, from less-than-perfect recommendations on Amazon, Spotify or Netflix.

Is my information secure?

Before you use any AI tool, you should know what it does with the data you enter into it. Machine learning systems are always greedy for more data - it’s how they get better - so lots of tools will train their models on your data.

OpenAI, the company behind ChatGPT, uses some data entered into the chatbot as well as its text-to-image generator Dall-E to improve their models (find out ways to opt out here), so make sure that you’re not adding any information that you wouldn’t share outside your company. Samsung got stung by this in 2023, as employees had been using ChatGPT to organise confidential documents - which quickly became not-so-private as ChatGPT used this information to answer other user’s questions. Many accountants using AI have a firm rule to never input anyone’s personally identifiable data into tools - and recommend keeping a record of exactly where AI has been used in any tasks.

What are the potential legal issues with AI?

AI is definitely complicating intellectual property law, with debates arising about whether AI can be trained on copyrighted material and who is responsible if new content created by AI infringes copyright. There are several ongoing lawsuits against AI companies, claiming that AI image generators produce content that ‘supplants’ the artist’s work.

Businesses using AI will need to ensure that the outputs they use don’t plagiarise existing works. Another consideration is whether your business has exclusive control over content you use - the UK Copyright Designs and Patents Act isn’t clear on whether content created using AI can be considered an original work. We will likely see clarifications and rulings emerging over the next couple of years, as well as more legislation around issues of liability for AI decisions, privacy breaches, and discrimination built into AI models.

There is currently no dedicated AI legislation in the UK but in the King’s Speech in July, the government said they would “seek to establish the appropriate legislation to place requirements on those working to develop the most powerful artificial intelligence models”. Keep an eye on upcoming legislation to make sure you’re on the right side of the law.

Does using AI for this task match my business values?

The decision to use AI isn’t all practicality, it’s also about being true to yourself and your business’s priorities.

For example, there are concerns about generative AI taking over creative tasks like design, copywriting, video production, and song writing. More than 10,500 creative professionals, including Abba’s Björn Ulvaeus, the actor Julianne Moore and The Cure’s Robert Smith, have signed a statement saying: “The unlicensed use of creative works for training generative AI is a major, unjust threat to the livelihoods of the people behind those works, and must not be permitted.” You might want to ask if this is a task you would previously have paid a contractor or freelancer to do, and whether using technology or a human would resonate better with your customers.

If sustainability matters to your business, you might want to consider the massive energy use of AI. The United Nations Environment Programme has called the impact of AI “concerning”, as models require such high computing power that they take a lot of electricity and water to power. ChatGPT’s daily power use is equivalent to 180,000 US homes. A 2023 study by Shaolei Ren at the University of California, Riverside found that an average ChatGPT conversation uses 500ml of water. Is your task important enough to use that much energy?

How much will AI cost?

AI can help small businesses reduce operational costs, by automating time-consuming (often mundane) tasks - but what will it cost upfront? The answer depends on the type of software you need, the level of intelligence you’re aiming for and the complexity of the problem you’re trying to solve. Upfront costs have reduced for many off-the-shelf AI tools - such as chatbots, marketing automation, customer service AI, and basic data analytics - in recent years, as competition for customers increases and technology advances. However, for more custom solutions, costs can quickly run into the tens of thousands.

What’s FreeAgent’s approach to AI?

From the very start of FreeAgent, we’ve been developing our automation toolkit to help solve our customers’ problems. We want to support you to get accurate data, see what’s important and stop repetitive tasks taking up a lot of your time.

“Customers don't typically come to us saying: ‘we want AI’,” says FreeAgent CEO Roan Lavery. “They just want to be able to run their businesses more effectively, whether that's getting paid on time, handling their expenses, completing Self Assessment or running payroll. If AI can help them do those things more effectively, that’s awesome.”

With this practical goal in mind, we’ve been developing what we think is truly useful for our customers rather than what might sound a bit flash. FreeAgent has been using machine learning - which identifies patterns in large datasets - to ease the process of explaining bank transactions since 2020. It’s an essential step in fuelling FreeAgent's accounting engine - which powers features like automatic tax return generation, cashflow forecasting and internal reporting. We’ve been enhancing the system continuously since it launched and today it automatically explains more than half of all transactions for new customers on day one of using FreeAgent.

We also use AI to help our users gather accurate data using Smart Capture. And we help users to sort through that data, presenting them with the right information at the right time to make smart decisions.

FreeAgent’s friendly AI-powered colleague, Ruby the Robot can answer some questions by directing you to relevant articles in our Knowledge Base - where, by the way, all the articles are written by humans. But if that doesn’t help, it’s easy to reach one of our UK-based (human) support accountants.

Want to see how FreeAgent could automate some of your admin? Take a 30-day free trial to explore all the ways FreeAgent can save you time.