So often when technology companies announce a fancy new feature, they’re adding something to their repertoire.
But this time at Xero, we’re taking a function away - and handing it over to a robot. To explain what our exciting new machine learning system will do, first I have to tell what the problem is in the first place.
Within our software, when small businesses go through their invoices and code their accounts, they often choose the wrong code for what they’re charging.
Even if it’s right, you still have to manually enter the information, unless you go to the trouble of setting up account code defaults for all your inventory items or contacts.
More often than not, a small business user of Xero has no idea they’re coding things wrong - which then sees their accountant or bookkeeper having to go in and recode the accounts.
We had no idea how widespread the problem was until we launched a feature called Find and Recode, which helped accountants and bookkeepers with that menial task.
We’ll only know the true impact of the technology when people start using it.
When we started digging into Find and Recode data around invoice coding that had later been fixed, we saw millions of problems (three million, to be exact). Every one of those mistakes represents wasted time, and a missed opportunity.
But now, we’re ready to put the robots to work to help save businesses and their advisors valuable time and money.
Developed in-house by a specialist team of engineers working full-time for a year across three different locations, Xero is launching a machine learning system that uses detailed statistical analysis to learn what’s changed and what it relates to.
So now, for example, when an invoice for the time a tradie spends on site gets re-coded correctly from Sales - Materials to Sales - Labour, the machine learning system tracks that invoice coding behaviour and the mistakes rectified by your accountant or bookkeeper.
Because we chose a very difficult problem to solve, we had to build our own system, rather than use an off-the-shelf solution, which typically handle large homogeneous data sets. We needed to build a system which learnt from, and applied these learnings to, individual businesses and accounts.
In other words, this wasn’t machine learning across the entire Xero customer base, but a system that could learn and assist the individual business and their partner based on their own specific circumstances.
We’ll only know the true impact of the technology when people start using it. However, the machine learning system reaches over 80 per cent accuracy after learning from only four invoices. By the 50th, it’s consistently reaching over 90 per cent.
Built off the back of our recent migration to Amazon Web Services, this is the first specialist, personalised machine learning in a small business cloud accounting system, and we couldn’t be prouder. The average time to create an invoice in Xero is 1:38. When you consider that half a million invoices are raised in Xero daily, that's 13,600 hours every day.
For every second we shave off that average edit time we collectively save Xero small businesses around a working month every day.
It will initially be made available to a small group of small business customers and their accounting partners today, before being launched to a wider group to beta test, and then to all Xero customers and trialists later this year.
It’s our first step towards building a bespoke, personalised assistant for small businesses and their partners.
Luke Gumbley is a product manager at Xero.
Join the CIO New Zealand group on LinkedIn. The group is open to CIOs, IT Directors, COOs, CTOs and senior IT managers.