Don’t understand predictive algorithms? Xplainable bridges the “how and why” gap of machine learning

"There is so much hype around AI. Let's just focus on some simple implementation of traditional machine learning, and you'll get so much more value. You can do that with explainability," says Xplainable founder Jamie Tuppack.

An early-stage Brisbane AI software and consultancy startup now has four paid pilots to its name, including an ASX200 retail company, for its offering that bridges the "how and why" gap between machine learning-driven predictions and non-technical users.

Xplainable founder Jamie Tuppack says the software runs real-time scenarios and highlights the breakdown of predictions, outlining potential points for optimisation which are then applied within the business.

"We implemented this for an ASX200 listed company that was experiencing too much inquiry through their online contact forms," Tuppack said at the University of Queensland (UQ) Ventures annual ilab Accelerator Pitch Night last month, which he won.

"We were able to prioritise the leads, and also highlight novel insights such as how the length of the inquiry and the use of specific terms directly predicted the customer's likelihood to purchase.

"This is something that you cannot do with traditional machine learning. And the outcome, we managed to achieve an incredible 80 per cent increase in conversion, which is equivalent to $24 million."

Tuppack says the current problem is that users who create machine learning models tend to exist in siloes, often with years of deep technical expertise, but they struggle to communicate this back to their business.

"On the other hand, business users for one don’t know what ML stands for, and those that do, ask questions such as ‘what’s the value add? How long will it take to implement? And most importantly, how do I know I can trust it?’

"This is exactly where we play – Xplainable, bridging the gap between business and technical speak; instead of just providing a prediction, we tell you how and why you got that prediction."

Tuppack tells Business News Australia the idea for the business came to him while working for a resources company where he needed to predict the likelihood of mechanical failures for heavy vehicles and equipment.

"We got really quite accurate results - magnitudes better than a blind guess - but the issue is that it came back to a number," he explains.

"I was giving a number to a maintenance personnel and they didn't know what to do with it.

"What we’ve built at Xplainable is we can tell you the 'why' behind that, such as 'you're experiencing high EGR (exhaust gas recirculation) attempts' or 'you've got wear in your bearings' - targeted heuristics that you can actually look at to fix the problem, as opposed to just a number."

It is a service that can be applied across a wide range of sectors where companies collect data in various forms, with existing work involving house price prediction for a real estate company, compliance and risk checks for an insurer, and maintenance for a resources player. 

"We're highly aware that that there are some companies out there at the moment which because of the hype of AI they know they want AI, but they just don't know what sort because there are so many facets to it," Tuppack says.

"They’re often very early on in the spectrum of data in general. Usually it’s the full lifecycle taking it from the education piece all the way through to implementation and deployment, and it’s actually quite a long process."

The entrepreneur sees strong potential for Xplainable to have further penetration at retail where companies often produce large volumes of data, but he is also optimistic for the role the business could play in tackling topical issues around the bias and discrimination that is inherent within machine learning algorithms.

"With the rise in regulation around AI and algorithmic processes, we’re perfectly positioned to capture that as well," he says.

"It's a side effect of explainability piece, because if you have a column that's representative of a gender or demographic, we can actually highlight whether they're more or less likely to achieve X, and in doing so you can feed that back to your internal stakeholders and ask if they're happy with it.

"I’d like to get in front of government as well, to let them know there are Australian-based companies that are actually trying to solve this solution."

Tuppack says financial loans are another area where the Xplainable model would be useful.

"This way you can have two-way transparency in regards to, 'you’ve been denied a loan on this instance, but if you increased your monthly payments by $200 or how much you get paid by $200, we’ll give you that loan. Currently you’re sitting at 52 per cent risk, but if you do these things you can get it down to 48 per cent'," he says.

"If you explain to people how things work they’re much more likely to adopt. It was drastically noted when we were dealing with enterprise and companies. I'd say double the amount of people are happier with a solution whereby they don't just get a number, but they are told why they got that number."