Sage CTO spells out AI accounting visionby
Aaron Harris tells AccountingWEB that Sage's journey into the realm of artificial intelligence is not just about automating processes but fundamentally reshaping the way accountants interact with software – a vision that involves large language models, digital assistants and Excel.
When it comes to the impact generative AI will have on the future of accounting, as you might expect Sage’s chief technology officer Aaron Harris is bullish – and he believes the Newcastle-based software giant is developing products to back this vision.
“Generative AI not only offers the ability to analyse information much faster and more efficiently but being able to use natural language reimagines the way that users can interact with software,” Harris told AccountingWEB from his home in Texas.
“It’s our ambition to eliminate the monthly close and bring about a world where all data is real-time and reliable, and generative AI is the last piece in the puzzle,” he continued. “The accounting industry is stuck in cycles of tax filing or month end close, which happen too far after the event to be useful and are far too time-consuming.”
To achieve this ambitious goal, the Newcastle software giant is leaning heavily on ‘reusable’ AI capabilities such as a new digital assistant and a recently developed inbox tool.
What makes these particularly notable from an established vendor such as Sage is that these new services can be used across much of its product set, whether they’re cloud or desktop-based, and in the case of the digital assistant even embedded into Excel.
This reusable element relies heavily on the Sage Digital Network architecture developed on Harris’s watch, which allows all the developer’s services to integrate more coherently with each other.
Narrow AI capabilities
Harris established Sage’s AI team five years ago and emphasised the importance of creating AI tools where the capabilities are explicitly defined, rather than a wide-open, generalised AI system.
“We’re focussed on narrow AI, training it to do one thing very well, very efficiently, as opposed to the more general functionality you’ll find with other platforms,” said Harris. This would ensure that the errors or ‘hallucinations’ which creep in with more generalised products would not occur in this context.
Sage Digital Assistant
One example of the reusable services mentioned by Harris is the Sage Digital Assistant, which will be available to the first cohort of customers this quarter.
The Digital Assistant is designed to interact with the majority of Sage’s products that can also plug into different solutions – with Excel a primary target for its outputs.
While the tool can be used for a wide variety of accounting tasks, one demonstration of its functionality Harris has given to accountant partners is to ask it to project a business’s cash balance at the end of the month.
“The assistant knows to go to the accounting system to get the starting balance,” explained Harris. “It knows the basics of how to do a cashflow forecast, look at receivables, go out to the planning product, payroll etc. It then gives an answer of what the balance is likely to be, which the user can push directly to Excel and work on it there.”
According to Harris, the advantage of being able to extract data directly to Excel is the user then has access to all the data used to get to the answer, rather than a ‘black box’ approach used by many of the new generative AI tools on the market, which often fail to show their workings or data sources.
“Transparency is critical,” he said. “It’s one of the core principles of trust in AI. If an accountant gets burned once with a tool that gives them the wrong answer or one they’re not able to verify, they won’t use it again.”
Another way in which Sage hopes to use its AI capabilities to drive efficiency is through a new inbox product. This stitches in functionality from AR/AP workflow tool Lockstep, acquired by Sage last year, and will eventually spread across the whole Sage network.
“Most AP products have an inbox - something like '[email protected]' – and they usually have basic capabilities to extract details,” said Harris. “Ours is a much more complex inbox tool. For any email that comes in from customers or suppliers that comes into the Sage address we’ve built a generative AI large language model (LLM) that is able to read incoming emails and, based on the ability to understand context and language, predict what the standard accounting workflow will be.”
Harris stated that the top 50 most common workflows have already been mapped out, from customers asking for a copy of an invoice to a client looking for some basic tax information.
“The tool first categorises, then recommends the next step,” he said. “If the customer asks for a copy of an invoice, the AI knows to go and get the invoice, then composes a reply and attaches the invoice to the email. It then allows the user to review before hitting the send button.
“In isolation, it’s a relatively minor change. However, if you’re a large accounting team that gets 100 of these a day, as our research shows many teams do, dealing with each request takes five to seven minutes and you can cut that to 30 seconds, you can see it’s a powerful tool.”
The inbox tool is live in Sage Intacct today, currently in testing on the UK version of Sage 50, and will eventually integrate with almost all Sage products and services.
Harris emphasised that a strong focus for the developer is on accounts payable and accounts receivable, working on collections, matching deposits and receipts.
“Those are the areas our research tells us the pain is for accounting professionals,” he said. “We’ve cut our time for processing invoices in half, making us two or three times more productive. It’s a clear story to tell our users, you know what the productivity gains will be.”
Sage already deploys AI tools to find accounts payable problems such as duplicate invoices, checking not only for the same number but also for the same amount, description, quantities etc. However, according to Harris, it can also spot potential fraud.
“Maybe the invoice has a fake Amazon logo on it? We know what an Amazon invoice looks like because we’ve processed thousands of them so we can spot a fake,” he said. “This is one of the big benefits of the Sage network. We don’t build AI for one customer. The models learn across tens of thousands of customers and become highly tuned for specific vendors when used across the network.”
Harris also teased another upcoming release: a generative AI tool that reads and summarises reports like income statements for small business owners, leveraging OpenAI rival Anthropic for its enhanced analysis capabilities.