When ChatGPT first came out, no one was quite sure how it would change the way accounting firms worked. However, a lot of people in many different places were confident that it would. Eighteen months later, we can see how the advent of generative AI is changing the tools that we use every day.

Generative AI (often abbreviated to GenAI) is different to the machine-learning intelligence that plans our routes in Google Maps or finds the next series to watch on Netflix. GenAI is a language-based model that, in essence, predicts the probability of the next word or character in a sequence.

In 2023, this technology proved to be so powerful that ChatGPT and other chatbots could hold a plausible conversation on a seemingly infinite array of topics. At times it feels like you are talking to an overeager child-savant: one who knows the entirety of information on the internet but who occasionally pretends, by making up answers.

GenAI is now popping up like mushrooms after summer rain. ChatGPT has a handful of competitors in the Englishspeaking world and many more from Chinese and European companies. We can also see GenAI appearing within established software, from the Microsoft Office Suite through to practice management software and enterprise resource planning (ERP). And we can also see new types of tools emerging which are largely built on the functionality of GenAI and challenging the existing players.

AI in practice software

The best form of validation is adoption and software companies have been adding features driven by generative AI at a very rapid clip.

One of the first movers was practice management software Karbon. Karbon is best known for its excellent email management. It has a triage function to quickly process large volumes of messages into tasks and automatically file to client groups.

Just four months after the release of ChatGPT 3.5, Karbon used the platform to assist accountants in reading and writing emails. Karbon uses AI to assess urgency, sentiment and tone to prioritise your inbox and show you which emails need your attention first.

Partners who are stepping into a conversation between a junior accountant and a client can read a summary of an email thread to pick up the main points. And everyone in the firm can quickly draft emails with a professional tone – or make them more conversational, direct, apologetic or neutral.

For ongoing work, accountants can set a task to automatically draft a personalised update that reflects progress within the job, along with any other prompts you wish to provide.

Several reporting apps have also adopted GenAI very quickly. Etani, a platform that connects client data to Microsoft’s business intelligence tool Power BI, has added a chat interface called Eva to query combined business data. You can ask Eva questions that pull answers from information stored within Xero, HubSpot, Shopify and Xero Practice Manager, among other apps. For example, you can ask Eva: “Generate a custom report detailing our team’s billable hours this month versus last month, segmented by member and client, and pin it to my dashboard”.

Xbert takes an AI-centric approach to practice management. It uses a collection of 80 machine-learning algorithms that monitor every client’s accounting file, looking for potential issues. One algorithm checks for wrongly paid or unclaimed GST by checking when the invoice was raised – and if it was raised. It will also alert you if a client has paid GST to a supplier when it shouldn’t have, and claimed tax by mistake. Another algorithm alerts you if an invoice or bill is associated with a cancelled business. It will also check tax validity on attachments and make sure they match the contact record.

Xbert works by identifying exceptions and so is useful for both bookkeeping and auditing. Rather than discover issues 12 months after the fact, accountants and bookkeepers can clean the file every month.

Aider, a practice management tool for scaling advisory services, launched an advisory assistant that helps accountants prepare for client meetings, generates sales scripts for selling advisory services, and conducts high-level analysis of data trends. Rather than writing explanatory notes next to graphs in a financial report, accountants can read the summaries generated by Aider. Junior accountants can use this as a guide in conversations with smaller clients that otherwise wouldn’t pay for a similar conversation with a partner.

AI everywhere

Hundreds of other business applications have also embraced generative AI and added interesting and powerful features.

Zoom and Microsoft Teams have introduced AI assistants that automatically transcribe calls and will provide you with a summary afterwards if you missed the meeting or need a reminder of what was discussed.

These assistants can even hold a conversation about the content of the meeting and answer specific questions such as, “When did the client want to start the project?”.

The assistants add chapters to recordings, so you can jump to a specific area and can create a list of actions for attendees based on the commitments agreed on the call.

HubSpot, a popular suite for sales, marketing and customer service, has also added an AI assistant that brings various capabilities to each function. Perhaps it is most useful is in its Marketing Hub where the assistant can outline, draft, rewrite and repurpose content, and use it for blog posts, social media or generating images.

Even though GenAI is built on languagebased models, it is surprisingly effective at interpreting words into images. Canva, an online design tool, has wholeheartedly embraced GenAI. Features include an AI art generator that can turn text instructions into images or even short videos.

Adobe Photoshop has released similar features, including a ‘magic fill’ that expands an image beyond what was captured in the original. For example, you can expand a photo of a mountain and the software will add more rocky terrain that matches the shape and colour of the original photograph.

AI in accounting software

For several years now, cloud accounting software has used AI to speed up the reconciliation process. Automated bank feeds are these days an entry-level feature,as is the auto-matching of transactions powered by machine-learning techniques.

Accounting software companies are also moving to adopt GenAI, although at different paces.

Xero was an early leader in applying machine-learning AI to improve transaction matching in its bank reconciliations. Last year, it rolled out a more powerful version that will add new contacts to Xero that aren’t in its contact list. (Previously, Xero would predict a contact and account code for transactions only where the business already had a relevant contact.)

Xero’s forecasting tool, Analytics Plus, now also includes predictions for regular invoice and bill payments.

However, it is taking a more tentative approach to generative AI. At last year’s Xerocon Sydney conference, the company announced two experiments. The first is in Xero Central, its customer support and learning site, where GenAI will improve the built-in search function to boost efficiencies.

A second experiment will use an AI assistant to help businesses set up their Xero organisation during the onboarding process. The AI assistant will use Xero Central articles to provide answers to user questions, and guide them on how to complete tasks.

MYOB is using GenAI among its software developers to reduce the time they spend creating code. The company says it has rapidly improved delivery times, which means it can release features faster.

Meanwhile, Intuit is going all in. Last year it announced with much fanfare the launch of GenOS, an operating system for the whole company. GenOS comprises four parts: a collection of large language models fine-tuned for tax, accounting, personal finance and marketing; a dedicated development environment where developers can rapidly experiment with and refine GenAI experiences; an intelligent layer of software that selects from multiple large language models; and a library of customer interfaces to display GenAI features.

This operating system is behind Intuit Assist, a generative AI assistant that Intuit has added to four of its flagship products; QuickBooks, marketing platform Mailchimp, personal tax software TurboTax and personal finance tool Credit Karma.

Intuit Assist appears as a chatbot inside QuickBooks and will answer questions about the accounting file. For example, you can ask Intuit Assist to show your profit and loss for the last month, and then to compare it with the previous month. Or you can ask it to show you your top-selling products last month.

Intuit Assist appears as a chatbot inside QuickBooks and will answer questions about the accounting file. For example, you can ask Intuit Assist to show your profit and loss for the last month, and then to compare it with the previous month.

The answers appear within the chatbot window as specific numbers, rather than as a full-screen report that a user needs to interpret. Intuit Assist will then suggest follow-up questions as buttons based on your previous questions, similar to the way Google displays a section called ‘People Also Ask’ in its search results. A user can then select one of the options or type in their own question.

Even ERP companies are adding GenAI to their software.

NetSuite announced last year features such as NetSuite Text Enhance that can write targeted and personalised collection letters and generate summaries and narratives for financial reports. The feature is also active in its modules for sales, support, supply chain, marketing and HR.

It is remarkable to look at the degree of change in business software since the launch of ChatGPT 3.5 in November 2022. GenAI now feels almost human, and its ability to instantly analyse text, write words and create images indicate it will reshape how professionals work.

Medicine, law, architecture, engineering, as well as accounting – access to all this knowledge will become easier and cheaper, and the value will move to reviewing GenAI content and reports for quality and accuracy.

If it reaches its full potential, GenAI will make the journey from desktop software to the cloud look like a bump in the road.

---

Know your genAI chatbots

Even if you are happy with ChatGPT, it’s worth knowing the other major competitors. Most of the chatbots listed below conform to the same style of interface as ChatGPT. The main difference is that some are trained on different computer models, which cost hundreds of millions of dollars to build and train.

The next most familiar AI chatbot is Bing, which is Microsoft’s version of ChatGPT (Microsoft is a majority owner of Open AI, the company that makes ChatGPT). Bing is built into the Microsoft Edge browser to improve search results, and into Windows to make it easier to search for files, applications, settings and other locally stored content.

Microsoft has also added AI to its Office suite under the product name Microsoft Copilot. Copilot can summarise threads and draft emails in Outlook, supply formulas in Excel and create slide decks in PowerPoint. While it is quite slow to respond to requests and appears to be less accurate than ChatGPT, it’s clear that Copilot will have an enormous impact on productivity for Microsoft Office users once it matures.

Google has also released an AI chatbot called Bard which, as one would expect, is stronger in internet search. While neither Bard nor ChatGPT has direct access to live information, Bard offers a ‘Search It’ feature that lets you initiate web searches directly within the interface. This provides a workaround for accessing more up-to-date information, as all of these AI chatbots create responses based on enormous databases that aren’t updated daily. (The actual frequency of updating these databases is a competitive secret. A major update such as ChatGPT’s transition from version 3 to 3.5 took six months.)

Bard is smart enough to understand intent. If you use Bard to search for flight and hotel prices, it can build your itinerary and give you a packing list. Bard is connected to Google’s Workspace productivity suite (formerly called Google Apps). For example, Bard can summarise or paraphrase emails in Gmail, find specific information in emails – and even generate poems or code based on the content of an email. It can also summarise, translate and find specific information in documents stored in Google Drive.

Google’s engineering heritage also shows, as it has several features for computer programmers such as translating code, explaining concepts and responding to code-related prompts. It can also export conversations in JSON and HTML formats, as well as text.

Claude by Anthropic is also a conversational AI model that emphasises safety. Anthropic describes Claude as being built with an ‘AI Safety via Debate’ methodology, which means it’s designed to consider multiple perspectives on a question before generating a response. This approach is intended to reduce harmful outputs and increase the reliability of the information provided.

Anthropic claims that Claude is better at understanding and responding to nuanced and complex queries to reduce misinformation. Anthropic has focused on understanding context and user intent; it provides more concise answers than ChatGPT and offers sources to back up its statements.

Another chatbot worth mentioning is Perplexity AI, which is an AIpowered search assistant built on ChatGPT. It combines the depth of search with the ease of conversation – a combination between a chatbot and Wikipedia. Perplexity AI stands out for its ability to provide contextrich answers and citations, making it an excellent tool for finding detailed, accurate information.

---

Take away