Telstra off-shoot dishes up AI

When New Zealand pizza franchise Hell Pizza made the journey across the Tasman to set-up shop in Brisbane, it needed to get noticed fast.

This is when speech recognition technology company, Inference Communications, made a deal with the devil, so the pizza chain could take advantage of inexpensive off-the-shelf automated voice technology as a marketing tool.

“Fifty-three per cent of our business is done over the phone,” Hell Pizza managing director Mike Hird explained.

“Because the phone is the first point of contact for many customers I needed to be sure that their first experience was a good one and that they would come back for more.”

Hell Pizza needed the ability uniquely engage with its callers to strengthen their brand in Australia – the challenge was on how to do so over the phone.

Formed in 2004, Inference Communications is a privately owned Australian speech recognition company.

It was the brain child of several Melbourne Business School graduates looking to further develop and commercialise raw technology developed in the now defunct Telstra Research Laboratories (TRL).

The technology, developed by Inference Communications CTO and computer scientist Dr Brad Starkie at TRL, called grammatical inference, is the end product of years of research into artificial intelligence. The technology is now used to develop the back-end of speech recognition systems.

“As a company we began by developing technology to build speech recognition systems. We automate processes, build language models, web interfaces and pronunciation models,” Starkie says.

“Our major advantage is that we have unique technologies to build software very rapidly, and we’ve written software that writes software.

Our basic strategy is to automate as much of the development process as possible, so that the end product is fast, cheap and reliable.”

Traditional speech recognition systems are developed by writing and feeding the software the hundreds of different ways in which a caller might ask or respond to a question for a specific application.

Faced with this complex and laborious process, Hell Pizza was instead looking to utilise speech recognition technology in a less expensive way because the system costs could only be shared between the two stores the company started with.

“Traditionally, you would have to go out and invest a fair amount of capital to buy your own platform and put it in a data centre, which is hundreds of thousands of dollars. But with the cloud computing approach, companies need not endure heavy up front costs,” Inference Communications product manager and co-founder Callan Schebella said.

“When you do it in the network you literally pay for it as you use it. So if your company is just starting and you don’t make many calls, you can still benefit from the technology.”

Page Break

Inference Communications implemented a store locator module – a pre-packaged product customised with Hell Pizza’s brand and image, and devilish voice of an actor.

The pizza franchise was given a secure web portal to work with, where they can log into from any location, and tweak their system to their desire, by creating new stores or removing old stores.

By using grammatical inference, this back-end process is automated and developing a solution takes just weeks.

This ability to personalize messages to specific callers stems from the enhanced intelligence of the technology and from the fact that the system is capable of leveraging the content of the organisation’s CRM database.

“This meant that instead of only being able to record a standard greeting and the usual instructional prompts, I was able to record multiple greetings and also had the flexibility to play those prompts in a randomised way to each caller,” Hell Pizza’s Hird said.

“This flexibility ensures the brand doesn’t get stale and keeps out customers engaged and interested. This is particularly important to our brand which is about being over the top and theatrical.”

The Inference Communications system not only automatically recognises how often customers have called but which store the have previously been connected to.

According to the company, 30 per cent of Australian businesses are using speech technology. Investment in the technology is poised to grow by 23 per cent per year over the next five years.

A surge in the uptake of speech technology around the world is being driven by advances in the technology like improved intelligence.

Search giant Google is pushing the trend too, through its implementation of GOOG411 service in the US.

The affordability is being driven by off-the-shelf software, external hosting providers and pay-as-you-go services models.

Schebella boasts that the system can be built cheaply, because their grammatical inference technology eliminates the need for translators and psychologists to come in and put voice data in a machine and that the constant data received via customers phone calls, enables the company to fine-tune their systems to better cope with the colourful Australian English.

“With grammatical inference we provide examples on what we want the system to understand and the machine will in real-time infer from those examples many thousands of ways of saying a particular phrase,” he said.

Starkie adds that standard voice xml platforms are used to run the technology on, so businesses don’t need anything out of the ordinary to deploy the application on.

“The system’s thinking is all done in advance, derived into a ‘language model’, and then clients chose what language model to feature in their system,” he said.

Now with six stores, Hell Pizza have been able to continually add stores, keeping the same call system, leading to improved customers satisfaction and a reduction in the length of calls times.