This week Xero hosted a seminar at its London headquarters about artificial intelligence and its impact on accountancy. In response John Stokdyk set off to find evidence of this growing trend.
The accounting software industry is buzzing with news and gossip about artificial intelligence and machine learning. Such is the speed of change in this area that the conversations are moving on from last year’s fear, uncertainty and doubt phase of will machines take our jobs?
Instead, software developers and accountants are now talking about how artificial learning systems are actually being applied to accounting tasks, now.
Earlier this month, Karbon’s Ian Vacin blogged on our US sister site about the big cloud accounting platforms are aggregating customer data to build algorithms that auto-categorise and auto-complete many of the tasks within their products. For accountants, he said, these innovations would ease their workflows by eliminating some existing processes through “learned, repeatable behaviour”.
As Vacin noted, Intuit QuickBooks, Xero and Sage all exploring auto-recognition to categorise transactions based on crowdsourced behavioural patterns.
Machine learning arrived within Xero via the Find & Recode feature introduced in 2015. Since then, the underlying data platform (Amazon Web Services) has been scanning code corrections to find common patterns. From these, Xero built a predictive model that it can apply to any new coding scenarios that show up.
The machine learning system learns what the accountants change and what it relates to, the Xero blog explained. So it will recognise a coding for time billed switched from Sales-Materials to Sales-Labour the next time the business creates a similar invoice.
Xero has been refining its algorithms based on the 3m corrections entered every month by accountants and bookkeepers and claims it can achieve 90% accuracy after 50 invoices. In March, Xero took a step further by running a beta test where the coding dialogue box was removed for some users, with the coding engine doing the work automatically instead.
Xero UK managing director Gary Turner told AccountingWEB the rationale behind this move was, “What if we could prevent the mistakes from happening in the first place?”
To our knowledge, AccountingWEB members Kent Accountant and Glennzy aren’t part of that experimental Xero user group, but they are moving in a very similar direction. In response to an Any Answers post from BromleyBob, they explained how they use Find & Replace when dealing with adjustments to clients’ year end accounts.
Kent Accountant uses the tool to make any corrections/amendments in the client’s books and posts year-end adjustments as journals in Xero when preparing the annual accounts.
Glennzy took a similar approach and “saves loads of time” using Find & Recode. “I am moving to doing everything in Xero including tax provisions, then just posting the [trial balance] into [accounts production] to file [statutory] accounts,” he said.
“I had a few incomplete records jobs where I scanned the bank stats into OcRex, then into Xero so had full audit trail instead of using spreadsheet.”
On the issue of whether or not to amend the client’s underlying accounts for the previous year, and with an eye on the digital horizon, Glennzy commented, “I cannot see why you wouldn't amend the accounts as it just means you’d be doing it every year. With MTD surely you will need to have the Xero correct or you will end in a right mess.”
Xero, meanwhile, is also talking about applying machine learning and categorisation to bank reconciliations and making the chart of accounts “invisible” to users.
At the Thomson Reuters Synergy event in May, Gary Turner explained that the Xero database holds 10m different chart of accounts codes. “It’s like the wild west - completely unstructured,” he said. “You can see where the controller brought in the system they’ve previously used and all kinds of multi-segment codes, departments and so on.”
When Making Tax Digital finally arrives, it will accelerate the adoption of these machine learning solutions. “[MTD] assumes everything is clean and ready for filing. And we know that’s not the case. Some small businesses have hundreds of chart of accounts codes. How the hell do you go from that to filing?”
Machine learning can play a significant role in tidying up that mess, Turner continued. “You should have a very small set [of codes], and tag them how you want to analyse them.”
Don’t overlook Sage
In contrast to Xero, Sage is more circumspect about going for fully automated categorisations. “I would love it if you could suggest all the reconciliations were done, but we are not focused on automation without having a level of control. Accountants want to be able to press a ‘confirm’ button, so we built that in,” said Sage product marketing director Michael Office.
Whisper it quietly, but AI and machine learning is an area where Sage has stolen a march on its cloud rivals. While the others are talking up their back-end categorisation and analysis applications, Sage was the first to hit the streets with an AI-powered chat bot, Pegg.
There’s still a lot of fluffiness attached to Pegg, which allows the user (or their adviser) to interrogate their live accounting data and to record transactions via SMS chat systems. But when AccountingWEB met Office at the Sage’s London summit in May, talk quickly turned to how Pegg might act as an intermediary or facilitator for the MTD process.
According to Office, Pegg can trigger reminders about submission deadlines and what clients need to do to meet them. For example, if the client hasn’t recorded all their receipts, the bot can remind them and warn that the quarterly records are about to be locked down. “When you plug in the accountant and client, you can start action and interaction,” Office said.
Machine learning on the front line
Earlier this week, James Poyser from Practice Excellence Award entrant inniAccounts told AccountingWEB that his team were already building machine learning into tools to improve the efficiency of routine compliance tasks.
Automated categorisation is a great start, but Poyser thinks there’s a lot more to come from AI for accountants who really get to grips with it. While inniAccounts isn’t the only firm to embrace machine learning, its strategy reflects the advice from The Future of the Professions author Daniel Susskind to sit down with a blank piece of paper and imagine how you could do things differently with available technologies.
Poyser said the philosophy driving his firm was “to help clients to help themselves” so they could make decisions without consulting experts. And the firm is already mobile apps and its web portal to do this. For example, the inniAccounts app (pictured at the top of article) includes “what if?” scenario planning tools to help clients explore the tax implications of their salary choices and other decisions.
“Xero is great, but it doesn’t do tax, so small businesses have to rely on an accountant,” said Poyser. “We’re looking at that data as we process it for new ways we can use it so they don’t have to ask. For example, we’ve done thousands of VAT returns. We can analyse them to understand the risks and steer clients away from them.”